Program

Type: short talk, 30 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

It is always cheaper to retain current customers than to acquire new ones. Decreasing customer churn is therefore crucial for any subscription-based service. It is important to target customers with high risks of churn and to understand which features drive up or down the probability of churn. By understanding the most important features for churn risk for an individual customer, we can target more precise measures to improve customer experience and retention.

In this talk we will tell the story on how we went from customer data to churn predictions. We will go through all the steps of a typical machine learning project, focusing on learning points in each step. We will present the architecture of our resulting churn pipeline and give you examples of how we are going to use the predictions to create business value for TV 2.

Speaker(s):

Astrid Johnsen Tessem

Astrid Johnsen Tessem works as a data scientist at TV 2. Her responsibilities consist of data modelling, predictive analytics, and making data available across the organization. She has previously worked as a developer at Knowit Experience and has a master degree in computer science and machine learning from NTNU.

Silje Hindenes

Silje is a Data Scientist at TV 2 with a background in Mathematics and Statistics from NTNU. At TV 2 Silje is a part of the team responsible for TV 2’s analytics platform, focusing on making data and insights available across the organization

Type: short talk, 30 min Language: en Tags: AI, Cloud, ML Audience: Beginner

Abstract:

I am 14 years old (I will be 15 in May) and I have been coding and building robots since I was 9 years old. I have built 5 robots until now; ic4u and ic4u2- a robot guide dog for blind people, BB4All - a robot that is designed to help teachers and students and to prevent bullying. Hashmet the Android – the Android robot and DO the droid from Starwars.

In my session, I would like to talk about the technologies I have integrated into my robots such as Google Assistant API, Google Maps API, and Google AIY voice kit. My robots have AI capabilities such as voice recognition, image processing, voice commands, and sensor-based automated decision mechanisms. To be able to do that I used Google Cloud, Tensorflow, and Dialogflow platforms and libraries. My motivation to be a part of Confer 2021 is to be able to encourage and show other kids especially girls that technology is fun and its good to be able to use technology for human benefit.

Speaker(s):

Selin Alara Ornek

My name is Selin Alara Örnek, I am 14 years old and I am a student in the 9th grade at Irmak High School, Istanbul. I started to learn to code when I was 9 years old and started building robots when I was 10. I speak English, Turkish, and French. I am learning C++, Python and Java. When I get the chance I really enjoy attending events, talking about my projects, and receiving feedback from all different people. I have 5 robots; ic4u, ic4u2, BB4All, Hashmet the Android and DO. In 2018 I won first prize with my robot ic4u in the hardware category at Coolest Projects International, Dublin. The same year I was also a finalist in the European Youth Awards Open Innovation category with ic4u. In October 2019 I was awarded the “European Digital Girl of the Year, 2019” award in the (11-14year old category) by the Ada Awards in Brussels. In November 2019 I mentioned as one of the world's “Devkid”’s in Google Developers Blog. In May 2020, I also won the Microsoft Youth Tech Star Award in the Women Leaders in Technology Awards. I am also a young member of the New York Academy Of Sciences

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML Audience: All

Abstract:

The goal of this talk is to advocate for a multidispinary approach to AI. Much like how HCI is a multidisplinary approach to Systems desifgn. - To humble Data Scientist in the audience about many aspects of Computer Science and Engineering they probably dont know, and to show how sociology, psychology and other fields are relevant to the training of ML models and AI systems more generally.

Speaker(s):

Julia Flament-Wallin

I'm a mostly back-end Software/Data Engineer with a HCI background and an interest for AI and ethics. I currently wrk in the Norwegian VR PaaS start-up Haptiq.

Type: presentation, 45 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

The process of building an AI looks like is so glamorous but is a long process, and at the end of the day, the tasks related to the AI model are just a 5% or less of the project.

We will see how to start an AI project from zero: defining the objectives, creating the architecture, building the game interfaces, massive data pipelines, defining model strategies, how to parallelize everything, etc.

'The abbey of crime' is an adamant 8-bit game. This game is more complicated than Montezuma Revenge and is a perfect challenge for an AI. Its complexity is about 10^1000 legal moves to solve it.

As AI technology, we will use Reinforcement Learning using Deep Neural Networks and Monte Carlo Tree Search.

The takeaways of this talk will be: understanding all the process involved to create an AI and learning the basics of Reinforcement Learning.

Speaker(s):

Juantomas Garcia

Juantomás leads Sngular's Data Science team and is also Chief Envisioning Officer. Since joining Sngular in 2018, Juantomás has leveraged his extensive experience to harness the potential of new technologies and implement them across the company’s solutions and services.

Juantomás is Google Developer Expert for Cloud and Machine Learning, co-author of the software book La Pastilla Roja and creator of “AbadIA”, the artificial intelligence platform built to solve the popular Spanish game La Abadía del Crimen. He’s an expert on free software technologies and has been a speaker at more than 200 international industry events.

He studied IT engineering at the Universidad Politécnica de Madrid and plays an active role as a tech contributor and mentor to various academic organizations and startups. He regularly organizes Machine Learning Spain and GDG cloud Madrid meetups and is a mentor at Google Launchpad for entrepreneurs.

Type: presentation, 45 min Language: en Tags: AI, Data Science, Ethics, ML, Security Audience: Intermediate

Abstract:

With the constant progress of AI, ML has now become all around us in real life apps instead of just research papers.

Doing so has given rise to a new type of issue that wasn't a problem before: Robustness.

With the nearly unlimited possibilities of the real world and its constant changes, Deep Learning models are frequently subject to tricky data points that cause it to give wrong predictions, ranging from very unfortunate events to people messing with them on purpose.

To mitigate this, robustness in our Neual Networks has become a must. We need to address this issue and test our models extensively before going to production.

In this talk, we will be introducting robustness and give examples of how easily our networks can be fooled, thus shedding light on how we can work around this.

Speaker(s):

Elyes Manai

Elyes is an Artificial Intelligence consultant and teacher as well as a Google Developer Expert in Machine Learning.

He enjoys teaching companies how to use AI/Data to improve their services and make more money.

Currently, he is a mentor at Google For Startups Accelerator and the co-founder of PyData Tunisia.

But most importantly, he is enjoying every part of it.

Type: short talk, 30 min Language: en Tags: ML, Research, GAN Audience: Intermediate

Abstract:

Generative adversarial networks framework immensely attracted the machine learning community attention in the recent years. GANs have succeeded in generating realistic-looking data from noise and have countless applications. On the other hand, Transfer Learning is a smart way that aims to enhance machine learning models in the case of scarce data. The main intuition behind transfer learning is to train models on a plenty of data that is “not” our targeted one but has some similar attributes to it. Then transfer learning in models that contains single networks, such as CNNs, are implemented by sharing the first layers (from the input side) of the model between source and target domains and initialize other layers then retrain the model on our scarce target data. This allows the model to have a good start when learning on the target data and achieve better results than if we trained it from scratch on the target. Now, GANs framework consists of two networks, a generator and a discriminator, and a legit question here is: which network and which part of that network contains the transferable features and hence parameter? Should we fine-tune or freeze the shared parameters? I am trying to answer this specific and direct question by showing imperial results of a series of experiments that I designed based on conditional GANs framework.

Speaker(s):

Awdelrahman M. A. Ahmed

I am a Ph.D research fellow in the department of informatics at the University of Oslo. My current research focuses on machine learning applications in smart grids and energy markets. I obtained MSc in renewable energy from Oldenburg University in Germany. Also I am a Marie-Curie alumnus.

Type: lightning talk, 10 min Language: en Tags: AI, ML, Research Audience: All

Abstract:

Machine learning -- in the form of convolutional neural networks -- has come a long way, from recognizing handwritten digits in the 90s to generating convincing deepfakes and providing highly accurate medical diagnoses based on medical images. In these examples the inputs and outputs are 1D, 2D, or 3D images, or numbers.

But how can a computer generate something more complex, like a shape? In this talk we will investigate one way of doing this, using our method involving implicit spline representations. We will demonstrate how this works on a congenital heart disease data-set, where it is able to reconstruct heart and connecting vessels from CT images with state-of-the-art accuracy.

Speaker(s):

Georg Muntingh

Research Scientist at SINTEF Digital, with a focus on machine learning and geometric modelling.

Type: short talk, 30 min Language: en Tags: AI, ML, Research Audience: Intermediate

Abstract:

An obstacle often encountered in computer vision applications is the time it takes to collect sufficient amounts of data. We'll talk about how Spoor use synthetic (rendered) data to training AI systems for detecting and classifying birds on wind farms with applications to environmental impact studies and operations.

Speaker(s):

Helge Reikerås

Helge is an experienced data scientist and AI engineer with 10+ years of experience. Currently he's the CTO at Oslo startup Spoor.

Type: workshop, 120 min Language: en Tags: AI, Data Science, ML Audience: Beginner

Abstract:

This workshop is divided in two parts. For the TensorFlow portion, the goal is to give the audience the flavor of building neural networks by using Colab notebooks. For the design part, the takeaway is to discover the user impact of the technical decisions when designing and building AI systems.

The goal is to introduce elements from both fields (development and design), so folks can work better together and understand both perspectives.

Speaker(s):

A. Rosa Castillo

Dr. A. Rosa Castillo has been working many years as Software Engineer until she fell in love with Data Science. In 2018 got her first role as an AI Engineer and now is currently working as a Data Scientist. She has completed different Data Science and Deep Learning certifications, and is an active member of the tech community giving talks and organizing events as well.

Besides she is also a WomenTechmakers Ambassador, engaged in building diversity in IT, and a collaborator of DeepLearning.AI testing new DeepLearning courses in Coursera.

Type: presentation, 45 min Language: en Tags: Cloud, Data Science, ML, anomaly detection Audience: All

Abstract:

Time series forecasting is a powerful way to use current or historical data to monitor, clarify, and predict “cause and effect” behaviors which can better inform future decisions. Anomaly detection for a time series can detect critical incidents such as fraudulent behaviors or potential opportunities such as changes in consumer buying patterns. Join this talk, to see the newest features in BigQuery ML in action and simplify your workflow in time forecasting and anomaly detection.

Speaker(s):

Stephanie Wang

Stephanie is a Java engineer in Developer Relations at Google Cloud Platform. Stephanie primarily focuses on building and maintaining the BigQuery Java Client Libraries which help developers interact with BigQuery APIs more easily.

She also works on managing Java dependencies across the 70+ Google Cloud Java Client Libraries to ensure that they work with each other and troubleshoot any dependency problems.

Prior to Google, Stephanie worked as a software engineer at Morgan Stanley and led a team of 10 to build sales and trading applications. Outside work, Stephanie is passionate about animals -- she has an insta-famous dog, @sasha_the_shibapoo and a opinionated semi-aquatic turtle, @bombshell_the_turtle.

Polong Lin

Developer Advocate at GCP focusing on BigQuery ML.

Type: lightning talk, 10 min Language: en Tags: AI, ML, GAN Audience:

Abstract:

Approximating wind flows using Computational Fluid Dynamics (CFD) methods can be time-consuming. Creating a tool for interactively designing prototypes while observing the wind flow change requires simpler models to simulate output faster. Instead of running numerical approximations resulting in detailed calculations, data-driven methods in deep learning might be able to give similar results in a fraction of the time. This paper rephrases the problem from computing 3D flow fields using CFD to a 2D image-to-image translation-based problem on the building footprints to predict the flow field at pedestrian height level. We investigate the use of Generative Adversarial Network (GAN), such as Pix2Pix and CycleGAN representing state-of-the-art for image-to-image translation task in various domains as well as U-Net autoencoder. The models can learn the underlying distribution of a dataset in a data-driven manner, which we argue can help the model learn the underlying Reynolds-averaged Navier–Stokes (RANS) equations from CFD.

We experiment on novel simulated datasets on various three-dimensional bluff-shaped buildings with and without height information. Moreover, we present an extensive qualitative and quantitative evaluation of the generated images for a selection of models and compare their performance with the simulations delivered by CFD. We then show that adding positional data to the input can produce more accurate results by proposing a general framework for injecting such information on the different architectures. Furthermore, we show that the models’ performances improve by applying attention mechanisms and spectral normalization to facilitate stable training. Finally, the paper displays results on a real urban city environment dataset generated from the capital of Norway, which strongly supports the use of neural network-based methods as the underlying model for an interactive design tool.

Speaker(s):

Kristoffer Gjerde

Kristoffer Gjerde is the leader of BRAIN NTNU and a Computer Science major specializing in artificial intelligence at NTNU in Trondheim. He has earlier had internships in Equinor, Cisco, and Cognite where he worked on the remote control and autonomy of Boston Dynamics' Spot robot for use in the industry. Currently, he is writing his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow.

Henrik Høiness

Henrik Høiness has studied Computer Science with a specialization in Artificial Intelligence at NTNU Trondheim and UC Berkeley. His previous experience includes companies such as Yahoo!, Microsoft, Ignite Procurement, and Nabla Flow and research for Multimedia Research Group at UC Berkeley. This year, he wrote his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow. From August, Høiness will begin as Tech Lead at Ignite Procurement.

Type: short talk, 30 min Language: en Tags: Data Science, Research Audience: Beginner

Abstract:

As a data scientist, you may be asked to find the effect of a new initiative that your company trying to test. It could be a new product, a new policy, a marketing campaign or some changes to existing product. The business side gives you access to the market test data. However, you have not participated in its design, you may have or have not a control group and if you have a control group, you may find the participants in the treatment group inherently different from the control group. Can you still detect the causal effect of the new initiative? A difference in difference DID technique can be your solution!

The aforementioned case is one of many situations when randomization is not possible. DiD is a popular approach to evaluate causal effects of programs or policies when there is a lack of random assignments.

In this talk, we are going to discuss:
- Why you need experimental thinking in business world?
- What is the ideal experimental design?
- Intro to DiD technique especially when there is a selection bias!

The talk will be beneficial for those who are working with experiments ( A/B testing data) or are interested to get into this world.

The talk will assume no experience with experiments from before. Familiarity with regression analysis would be helpful but not necessary.

Speaker(s):

Sherif Ahmed

Sherif is a data scientist at Haugaland Kraft AS. He has experience with designing and analyzing social experiments in both lab and field setting.

Type: short talk, 30 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

The rapid rise of Artificial Intelligence (AI) poses fundamental challenges for the creative industry. Although AI technologies are being adopted at an ever faster pace, Design as an academic discipline has so far failed to provide a convincing answer to the opportunities and challenges of AI. As the number of interfaces between humans and information multiplies, so do the amount of design frameworks that are required to support this technology.

When it comes to the Internet of Things (IoT), it’s easy to focus on technological aspects. You can talk about different platforms or discuss which IoT solution might be the best to solve a specific problem. Looking below this layer of technology, it quickly becomes apparent that there are many more aspects that determine the success of the IoT. Not the least of which is the matter of how today’s connected products are designed.

Artificial Intelligence (AI), which was designed initially to replace highly repetitive, manual work, has exceeded expectations to complete tasks involving emotional creativity. A limiting factor of IoT is it adds devices and buttons which makes your life more complicated. Now with AI, you’re able to say things like ‘turn on the lights’ instead of pushing buttons, and it makes life simpler. It is the AI layer of natural language processing that helps IoT improve our lives.

In tapping into technology’s potential, it’s important to remember the end user — us humans. But as more and more experiences are built with ML, it’s clear that UXers still have a lot to learn about how to make users feel in control of the technology, and not the other way round. How do we create experiences that are user friendly and human-centric, while taking advantage of technology?

This talk will discuss some of the guidelines focusing on human-centered approach and can be used as reference by any UX designer to help navigate the new terrain of designing ML-driven products. As ML starts to power more and more products and experiences, let’s step up to our responsibility to stay human-centered, find the unique value for people, and make every experience great.

Speaker(s):

Siddhant Agarwal

Siddhant Agarwal currently works with Open Financial Technologies as 'Developer Relations Lead'. Previously, he has worked as Program Coordinator with the Developer Relations team at Google in India. As part of his role at Google, he had led the Developer Student Clubs program, TensorFlow User Groups, Google Developer Groups, and Google Developer Experts program in India. He also designed and executed the initiative “Build for Digital India” in 2019 launched in collaboration with the Ministry of Electronics & Information Technology, Govt. of India which witnessed 7K+ students participating and building solutions that “Solve for India”. Siddhant is one of ACM’s Distinguished Speakers among 200+ speakers worldwide and 20+ in India. A design thinker at heart, he loves working with startups and helping them scale in UX and improve their designs.

With over 6 years of industry experience, he has found his passion in ed-tech initiatives, community building, design innovation, startup ecosystem and building for the next billion users. For his contributions in community building, he was also nominated as one of the finalists of CMX Community Industry Awards 2021 for 'Community Professional of the Year for a Developer Relations Community' and 'Community Professional of the Year' categories. Being an avid public speaker, Siddhant has spoken at more than 200+ national and international forums impacting more than 70K students/developers/individuals in the last 5+ years.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

Simplifai has taken NLU to a whole new level and built a solution that read, interpret and process Emails and Documents. In the talk Erik will deep dive into the journey of how the solution now being one of the hottest AI product in Norway right now.

Tentative agenda:
- The history of AI development in Simplifai
- Chatbot -> Emailbot -> Documentbot
- Adoption of AI products in Norway, Europe and Globally
- A few key takeaways in developing AI products
-- Business case
-- Data security
-- Change management

Speaker(s):

Erik Leung

Erik leung is a visionary technologist and co-founder of AI company Simplifai who uses NLP technology for business process automation. Erik has 15 years’ experience in developing, designing, delivering and selling technical, and in recent years, AI solutions.

Type: presentation, 45 min Language: en Tags: Cloud, Kubernetes Audience: Intermediate

Abstract:

The rise of microservices, containers, and organisations and customers that expects new features delivered continuously have brought some new and interesting engineering challenges.

While a Platform Engineer is not an entirely new invention it is not just a pretty new name to the 'DevOps Engineer' or the more traditional SysAdmin role. The SysAdmin primarily concern themselves with the underlying infrastructure supporting the application and the 'DevOps Engineer' is often an ad-hock gatekeeper for getting things done on behalf of the developers for everything that involves automation. The Platform Engineer on the other hands build systems that allow teams to build on top of.

These platforms are built specifically for the developers who will be using them as a fully featured product.

Speaker(s):

Hans Kristian Flaatten

Experienced SRE- and Platform Engineering Team Lead with proven track record of leading engineering teams that build and operate Cloud Native and Kubernetes-based application platforms for customers primarily in public government, telecom, banking and insurance sectors.

Principal Consultant and DevOps Practice Lead for TietoEVRY where I drive build culture and competency within DevOps, Site Reliability Engineering (SRE) and Cloud Native practices internally and for our customers.

Type: short talk, 30 min Language: en Tags: Cloud, Kubernetes Audience: Beginner

Abstract:

This is a Kubernetes 101 talk geared toward data professionals, i will introduce Kubernetes, talk about the basics and what kind of issues it solves, focusing on things that data professionals care about

Speaker(s):

Abdellfetah SGHIOUAR

Google Cloud Engineer with a focus on Serverless, Kubernetes, and DevOps Methodologies. A supporter and contributor to OSS. Podcast host @cloudcareers.dev

Type: lightning talk, 10 min Language: en Tags: ML, Research Audience: All

Abstract:

Respiratory droplets are considered the main transmission mode of the SARS-CoV-2 virus, so social distancing has been widely adopted as a countermeasure. Individuals from different households are expected to keep a physical distance, while most public venues, such as restaurants, theaters and cinemas, are shut down completely resulting in undesirable economic and social effects. Lately, the unmatched development and application of vaccines have paved a path for reopening, but uncertainties surrounding virus mutations and vaccination rates may require to retain basic physical distancing measures for this transition period.

One of the challenges to reopening public venues in a safe and economically feasible manner is seat allocation, i.e. maximizing the number of occupiable seats while respecting distancing requirements between groups. Static layouts can efficiently be predetermined with classic optimization methods, while dynamic layouts offer more flexibility with respect to the arriving group sizes, but are also more difficult to determine.

In our work, we address the dynamic seat allocation problem with reinforcement learning (RL) -- a paradigm in machine learning of training a model via its interaction with an environment. In this lightning talk, we will share our intermediate results and experiences from this problem: first, how to model the problem and the learning environment (formulating state and action spaces, reward function, group size distribution, distancing constraints, group closeness) and, second, suitability, potentials and limitations of both classic and state-of-the-art RL methods (Q-learning, proximal policy optimization, tree search methods).

Speaker(s):

Arturs Berzins

Arturs Berzins is a PhD fellow at SINTEF, Department of Mathematics and Cybernetics and University of Oslo. His PhD project is part of the European GRAPES network and addresses the intersection of machine learning and geometry. He explores aspects of geometric representations, meta-learning and automated architecture search to increase the efficiency, performance and efficiency of machine learning systems and to address applications of geometric nature.

He received his BSc and MSc degrees with distinction at RWTH Aachen University focusing on the combined use of numerical simulations and machine learning methods in mechanical engineering.

Type: lightning talk, 10 min Language: en Tags: Cloud, Software Development Audience: Beginner

Abstract:

Whether you work as a data scientist, ML engineer or software developer, at some point you will need to develop applications – for your customers, products or as a support for the latest and coolest ML model that you have built.

With the introduction of the Cloud, a lot has changed in the way we work with the development and deployment of our applications.

In this talk, we look at 5 things to look for when creating an application in the Cloud age.

Speaker(s):

Rustam Mehmandarov

Passionate computer scientist. Java Champion and Google Developers Expert for Cloud. Public speaker. Ex-leader of JavaZone and Norwegian JUG – javaBin.

Type: short talk, 30 min Language: en Tags: Data Science, Research, History Audience: All

Abstract:

Terms like “data science” and “big data” has gone from being completely unknown, to everyday language, in just a decade. But in science, the phenomena has existed for many centuries. The mathematician Johannes Kepler inherited the vast catalogue of celestial objects from Tycho Brahe, who together formed massive mounts of data, and as result, Keppler’s laws of planetary motion. Even though there are several differences between todays and Brahe’s and Kepler’s approach, they where pioneers in data science - and the first to describe the universe with numbers and algorithms.

Speaker(s):

Vegard Lundby Rekaa

Astronomer, science communicator and data scientist. Pationate about the vastness of space and humanities limitless attempt to discover it’s fundamental properties.

Type: lightning talk, 10 min Language: en Tags: Ethics Audience: All

Abstract:

In a rapidly changing world of globalization and digitalization, computer ethics has become increasingly important. Focus on computer ethics goes beyond protecting personal and commercial information.

What are the responsibilities of companies, developers, and users, and how can we overcome the lack of awareness of computer ethics?

In this talk, I will talk about the importance of computer ethics in our daily life as a tech person and as a user in the digital world, and how we can start taking action today.

Speaker(s):

Tannaz N. Roshandel

Tannaz works as a cloud engineer for UiO, where she is building an open source cloud service for the research and education sector in Norway. She has a background in robotics and is passionate about computer ethics.

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML, Research Audience: All

Abstract:

This is an experience report on DNBs journey into transforming the organisation's culture and stepping up the competence and skills in data and analytics, corporate wise. One important track of this journey is how we are approaching the ethical part of AI, our learnings along the way, and what challenges we foresee as we increasingly leverage AI and Data in our daily business.

Speaker(s):

Aiko Yamashita

Aiko works as a Senior Data Scientist at the Center of Excellence in Advanced Analytics, at DNB. She has a background in Informatics with a PhD in Experimental Software Engineering, and +10 years of experience in Data Analytics for logistics, retail and the finance industries.

In addition to being a Data Scientist at DNB, Aiko is an Adjunct Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Analytics and has supervised students. Aiko is passionate about how innovation and technology can be better aligned to tackle current societal and environmental challenges, and how to develop technology in an ethical, humane, and just way.

Type: short talk, 30 min Language: no Tags: AI, Data Science, NLP, Semantic Technologies, RDF Audience: Intermediate

Abstract:

Da vi skulle modellere informasjon fra klartekst til semantiske kunnskapsgrafer hos Sjøfartsdirektoratet, så vi raskt at vi trengte en automatisk metode for dette. Manuell tolkning av informasjon er avhengig av menneskelig domenekunnskap for å identifisere kontekst, konsepter og relasjoner i teksten. Denne prosessen er grusomt tidkrevende, og inneholder høy risiko for misforståelser og å overse viktig informasjon. Dette foredraget vil ta deg med på turen fra klartekst til kunnskapsgrafer, via språkprosessering. Du vil høre hva vi lærte og se våre resultater.

Speaker(s):

Veronika Heimsbakk

Veronika er senior semantikknerd hos Capgemini i Oslo. Hun har en bakgrunn fra Universitetet i Oslo. Siden den tid jobbet som utvikler og informasjonsarkitet i semantikkprosjekter i inn- og utland i mange forskjellige sektorer. Hun elsker å modellere kunnskap ut fra et virrvarr av kaotiske data. Bortsett fra semantikkens finurlige verden brenner hun for realfagsformidling, LaTeX og fjellet.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML, TensorFlow.js Audience: All

Abstract:

AI can seem mysterious to neophytes, and rote for your classic scientists. In this talk I will go through an adventure to create a TensorFlow.js demo that is exciting for both.

Speaker(s):

Gant Laborde

Gant Laborde is an owner of Infinite Red, mentor, adjunct professor, published author, and award-winning speaker. For 20 years, he has been involved in software development and continues strong today. He is recognized as a Google Developer Expert in Web and Machine Learning, but informally he is an “open sourcerer” and aspires to one day become a mad scientist. He blogs, videos, and maintains popular repositories for the community. Follow Gant’s adventures at https://gantlaborde.com/

Type: short talk, 30 min Language: en Tags: ML, anomaly detection Audience: Beginner

Abstract:

Machine learning algorithms are used for condition monitoring of various types of machinery. Their flexibility enables them to learn complex patterns directly from training data in order to discern undesired operations and provide an early warning about potential equipment malfunction.

However, how to design an interpretable machine learning algorithm without knowing when problematic operations occur or even how they potentially present themselves in the data? In my talk, I will guide you through the process of combining unsupervised and supervised machine learning paradigms, as well as domain knowledge, to construct an algorithm for anomaly detection in maritime propulsion and manoeuvring systems.

Speaker(s):

Katarzyna Michalowska

Researcher in artificial intelligence at SINTEF Digital and a PhD student at University of Oslo in informed machine learning. Main areas of expertise include hybrid models of machine learning and physics, with the focus on making artificial intelligence models explainable and trustable.

Type: short talk, 30 min Language: en Tags: Data Science, ML Audience: Intermediate

Abstract:

Most progressive technology companies optimize their online products and make smart decisions utilizing thousands and thousands of A/B tests or running multi-armed bandits. However, many traditional offline businesses are still relying on simple analytics to make crucial decisions for their businesses: releasing new products, running marketing campaigns, changing their premises layouts etc. All this actions affect the market companies are working in, but how to dissociate the external factors impacting the market? In this session I will try to give examples how with the help of machine learning Coop Norway runs offline A/B tests or measures causal effects in offline retail.

Speaker(s):

Stanislav Polyakov

Stanislav Polyakov leads the Advanced Analytics team at Coop, one of the largest retailers in Norway. Stanislav primarily focuses on making Coop a data driven organization and a digital leader in the retail sector using cutting edge ML solutions.

Prior to Coop, Stanislav worked as a consultant helping many companies across the world to build their AI capabilities. His previous experiences include different Microsoft partners in Europe, Australia, New Zealand or large companies in Nordics, such as Tine SA where he was working on preventing the next butter crisis in Norway using ML.

Stanislav Polyakov has his Ph.D. in applied physics from Norwegian University of Science and Technology (NTNU) in Trondheim. His research was focused on new materials for solar cells and IR-lasers.

Type: short talk, 30 min Language: en Tags: AI, Research Audience: All

Abstract:

This talk will address why and how we should use artificial intelligence to predict the effectiveness of marketing communications, so that advertising entering the market becomes better and more effective.

Today, marketers are fighting for consumer attention in an increasingly fragmented and busy media landscape. It is increasingly important that marketing that goes into the market manages to break through the noise. Demands to prove returns on marketing dollars in both the form of sales and brand effects emphasize the importance. Nevertheless, up until now, few advertisers in Norway invest in measures to predict effectiveness and optimize their marketing campaigns before going live. This is often due to two main elements: restricted time and budget limitations.

Development of artificial intelligence helps to remove those barriers for marketers. Artificial intelligent tools with a large learning database spend only a few minutes analyzing advertisements second by second and can predict with high accuracy how the advertisement would perform in a traditional market research with “real consumers”. This provides a basis for optimization so that advertising is better received in-market and thus become more effective.

We address implications for marketers and how to move forward in the industry.

Speaker(s):

Eirik Friberg Ekrann

Eirik is Strategy Director at Kantar Norway and has 15 years of experience helping Norwegian and international companies strategically position themselves in the minds of consumers to ensure growth.

Type: short talk, 30 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

It is always cheaper to retain current customers than to acquire new ones. Decreasing customer churn is therefore crucial for any subscription-based service. It is important to target customers with high risks of churn and to understand which features drive up or down the probability of churn. By understanding the most important features for churn risk for an individual customer, we can target more precise measures to improve customer experience and retention.

In this talk we will tell the story on how we went from customer data to churn predictions. We will go through all the steps of a typical machine learning project, focusing on learning points in each step. We will present the architecture of our resulting churn pipeline and give you examples of how we are going to use the predictions to create business value for TV 2.

Speaker(s):

Astrid Johnsen Tessem

Astrid Johnsen Tessem works as a data scientist at TV 2. Her responsibilities consist of data modelling, predictive analytics, and making data available across the organization. She has previously worked as a developer at Knowit Experience and has a master degree in computer science and machine learning from NTNU.

Silje Hindenes

Silje is a Data Scientist at TV 2 with a background in Mathematics and Statistics from NTNU. At TV 2 Silje is a part of the team responsible for TV 2’s analytics platform, focusing on making data and insights available across the organization

Type: short talk, 30 min Language: en Tags: AI, Cloud, ML Audience: Beginner

Abstract:

I am 14 years old (I will be 15 in May) and I have been coding and building robots since I was 9 years old. I have built 5 robots until now; ic4u and ic4u2- a robot guide dog for blind people, BB4All - a robot that is designed to help teachers and students and to prevent bullying. Hashmet the Android – the Android robot and DO the droid from Starwars.

In my session, I would like to talk about the technologies I have integrated into my robots such as Google Assistant API, Google Maps API, and Google AIY voice kit. My robots have AI capabilities such as voice recognition, image processing, voice commands, and sensor-based automated decision mechanisms. To be able to do that I used Google Cloud, Tensorflow, and Dialogflow platforms and libraries. My motivation to be a part of Confer 2021 is to be able to encourage and show other kids especially girls that technology is fun and its good to be able to use technology for human benefit.

Speaker(s):

Selin Alara Ornek

My name is Selin Alara Örnek, I am 14 years old and I am a student in the 9th grade at Irmak High School, Istanbul. I started to learn to code when I was 9 years old and started building robots when I was 10. I speak English, Turkish, and French. I am learning C++, Python and Java. When I get the chance I really enjoy attending events, talking about my projects, and receiving feedback from all different people. I have 5 robots; ic4u, ic4u2, BB4All, Hashmet the Android and DO. In 2018 I won first prize with my robot ic4u in the hardware category at Coolest Projects International, Dublin. The same year I was also a finalist in the European Youth Awards Open Innovation category with ic4u. In October 2019 I was awarded the “European Digital Girl of the Year, 2019” award in the (11-14year old category) by the Ada Awards in Brussels. In November 2019 I mentioned as one of the world's “Devkid”’s in Google Developers Blog. In May 2020, I also won the Microsoft Youth Tech Star Award in the Women Leaders in Technology Awards. I am also a young member of the New York Academy Of Sciences

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML Audience: All

Abstract:

The goal of this talk is to advocate for a multidispinary approach to AI. Much like how HCI is a multidisplinary approach to Systems desifgn. - To humble Data Scientist in the audience about many aspects of Computer Science and Engineering they probably dont know, and to show how sociology, psychology and other fields are relevant to the training of ML models and AI systems more generally.

Speaker(s):

Julia Flament-Wallin

I'm a mostly back-end Software/Data Engineer with a HCI background and an interest for AI and ethics. I currently wrk in the Norwegian VR PaaS start-up Haptiq.

Type: presentation, 45 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

The process of building an AI looks like is so glamorous but is a long process, and at the end of the day, the tasks related to the AI model are just a 5% or less of the project.

We will see how to start an AI project from zero: defining the objectives, creating the architecture, building the game interfaces, massive data pipelines, defining model strategies, how to parallelize everything, etc.

'The abbey of crime' is an adamant 8-bit game. This game is more complicated than Montezuma Revenge and is a perfect challenge for an AI. Its complexity is about 10^1000 legal moves to solve it.

As AI technology, we will use Reinforcement Learning using Deep Neural Networks and Monte Carlo Tree Search.

The takeaways of this talk will be: understanding all the process involved to create an AI and learning the basics of Reinforcement Learning.

Speaker(s):

Juantomas Garcia

Juantomás leads Sngular's Data Science team and is also Chief Envisioning Officer. Since joining Sngular in 2018, Juantomás has leveraged his extensive experience to harness the potential of new technologies and implement them across the company’s solutions and services.

Juantomás is Google Developer Expert for Cloud and Machine Learning, co-author of the software book La Pastilla Roja and creator of “AbadIA”, the artificial intelligence platform built to solve the popular Spanish game La Abadía del Crimen. He’s an expert on free software technologies and has been a speaker at more than 200 international industry events.

He studied IT engineering at the Universidad Politécnica de Madrid and plays an active role as a tech contributor and mentor to various academic organizations and startups. He regularly organizes Machine Learning Spain and GDG cloud Madrid meetups and is a mentor at Google Launchpad for entrepreneurs.

Type: presentation, 45 min Language: en Tags: AI, Data Science, Ethics, ML, Security Audience: Intermediate

Abstract:

With the constant progress of AI, ML has now become all around us in real life apps instead of just research papers.

Doing so has given rise to a new type of issue that wasn't a problem before: Robustness.

With the nearly unlimited possibilities of the real world and its constant changes, Deep Learning models are frequently subject to tricky data points that cause it to give wrong predictions, ranging from very unfortunate events to people messing with them on purpose.

To mitigate this, robustness in our Neual Networks has become a must. We need to address this issue and test our models extensively before going to production.

In this talk, we will be introducting robustness and give examples of how easily our networks can be fooled, thus shedding light on how we can work around this.

Speaker(s):

Elyes Manai

Elyes is an Artificial Intelligence consultant and teacher as well as a Google Developer Expert in Machine Learning.

He enjoys teaching companies how to use AI/Data to improve their services and make more money.

Currently, he is a mentor at Google For Startups Accelerator and the co-founder of PyData Tunisia.

But most importantly, he is enjoying every part of it.

Type: short talk, 30 min Language: en Tags: ML, Research, GAN Audience: Intermediate

Abstract:

Generative adversarial networks framework immensely attracted the machine learning community attention in the recent years. GANs have succeeded in generating realistic-looking data from noise and have countless applications. On the other hand, Transfer Learning is a smart way that aims to enhance machine learning models in the case of scarce data. The main intuition behind transfer learning is to train models on a plenty of data that is “not” our targeted one but has some similar attributes to it. Then transfer learning in models that contains single networks, such as CNNs, are implemented by sharing the first layers (from the input side) of the model between source and target domains and initialize other layers then retrain the model on our scarce target data. This allows the model to have a good start when learning on the target data and achieve better results than if we trained it from scratch on the target. Now, GANs framework consists of two networks, a generator and a discriminator, and a legit question here is: which network and which part of that network contains the transferable features and hence parameter? Should we fine-tune or freeze the shared parameters? I am trying to answer this specific and direct question by showing imperial results of a series of experiments that I designed based on conditional GANs framework.

Speaker(s):

Awdelrahman M. A. Ahmed

I am a Ph.D research fellow in the department of informatics at the University of Oslo. My current research focuses on machine learning applications in smart grids and energy markets. I obtained MSc in renewable energy from Oldenburg University in Germany. Also I am a Marie-Curie alumnus.

Type: lightning talk, 10 min Language: en Tags: AI, ML, Research Audience: All

Abstract:

Machine learning -- in the form of convolutional neural networks -- has come a long way, from recognizing handwritten digits in the 90s to generating convincing deepfakes and providing highly accurate medical diagnoses based on medical images. In these examples the inputs and outputs are 1D, 2D, or 3D images, or numbers.

But how can a computer generate something more complex, like a shape? In this talk we will investigate one way of doing this, using our method involving implicit spline representations. We will demonstrate how this works on a congenital heart disease data-set, where it is able to reconstruct heart and connecting vessels from CT images with state-of-the-art accuracy.

Speaker(s):

Georg Muntingh

Research Scientist at SINTEF Digital, with a focus on machine learning and geometric modelling.

Type: short talk, 30 min Language: en Tags: AI, ML, Research Audience: Intermediate

Abstract:

An obstacle often encountered in computer vision applications is the time it takes to collect sufficient amounts of data. We'll talk about how Spoor use synthetic (rendered) data to training AI systems for detecting and classifying birds on wind farms with applications to environmental impact studies and operations.

Speaker(s):

Helge Reikerås

Helge is an experienced data scientist and AI engineer with 10+ years of experience. Currently he's the CTO at Oslo startup Spoor.

Type: workshop, 120 min Language: en Tags: AI, Data Science, ML Audience: Beginner

Abstract:

This workshop is divided in two parts. For the TensorFlow portion, the goal is to give the audience the flavor of building neural networks by using Colab notebooks. For the design part, the takeaway is to discover the user impact of the technical decisions when designing and building AI systems.

The goal is to introduce elements from both fields (development and design), so folks can work better together and understand both perspectives.

Speaker(s):

A. Rosa Castillo

Dr. A. Rosa Castillo has been working many years as Software Engineer until she fell in love with Data Science. In 2018 got her first role as an AI Engineer and now is currently working as a Data Scientist. She has completed different Data Science and Deep Learning certifications, and is an active member of the tech community giving talks and organizing events as well.

Besides she is also a WomenTechmakers Ambassador, engaged in building diversity in IT, and a collaborator of DeepLearning.AI testing new DeepLearning courses in Coursera.

Type: presentation, 45 min Language: en Tags: Cloud, Data Science, ML, anomaly detection Audience: All

Abstract:

Time series forecasting is a powerful way to use current or historical data to monitor, clarify, and predict “cause and effect” behaviors which can better inform future decisions. Anomaly detection for a time series can detect critical incidents such as fraudulent behaviors or potential opportunities such as changes in consumer buying patterns. Join this talk, to see the newest features in BigQuery ML in action and simplify your workflow in time forecasting and anomaly detection.

Speaker(s):

Stephanie Wang

Stephanie is a Java engineer in Developer Relations at Google Cloud Platform. Stephanie primarily focuses on building and maintaining the BigQuery Java Client Libraries which help developers interact with BigQuery APIs more easily.

She also works on managing Java dependencies across the 70+ Google Cloud Java Client Libraries to ensure that they work with each other and troubleshoot any dependency problems.

Prior to Google, Stephanie worked as a software engineer at Morgan Stanley and led a team of 10 to build sales and trading applications. Outside work, Stephanie is passionate about animals -- she has an insta-famous dog, @sasha_the_shibapoo and a opinionated semi-aquatic turtle, @bombshell_the_turtle.

Polong Lin

Developer Advocate at GCP focusing on BigQuery ML.

Type: lightning talk, 10 min Language: en Tags: AI, ML, GAN Audience:

Abstract:

Approximating wind flows using Computational Fluid Dynamics (CFD) methods can be time-consuming. Creating a tool for interactively designing prototypes while observing the wind flow change requires simpler models to simulate output faster. Instead of running numerical approximations resulting in detailed calculations, data-driven methods in deep learning might be able to give similar results in a fraction of the time. This paper rephrases the problem from computing 3D flow fields using CFD to a 2D image-to-image translation-based problem on the building footprints to predict the flow field at pedestrian height level. We investigate the use of Generative Adversarial Network (GAN), such as Pix2Pix and CycleGAN representing state-of-the-art for image-to-image translation task in various domains as well as U-Net autoencoder. The models can learn the underlying distribution of a dataset in a data-driven manner, which we argue can help the model learn the underlying Reynolds-averaged Navier–Stokes (RANS) equations from CFD.

We experiment on novel simulated datasets on various three-dimensional bluff-shaped buildings with and without height information. Moreover, we present an extensive qualitative and quantitative evaluation of the generated images for a selection of models and compare their performance with the simulations delivered by CFD. We then show that adding positional data to the input can produce more accurate results by proposing a general framework for injecting such information on the different architectures. Furthermore, we show that the models’ performances improve by applying attention mechanisms and spectral normalization to facilitate stable training. Finally, the paper displays results on a real urban city environment dataset generated from the capital of Norway, which strongly supports the use of neural network-based methods as the underlying model for an interactive design tool.

Speaker(s):

Kristoffer Gjerde

Kristoffer Gjerde is the leader of BRAIN NTNU and a Computer Science major specializing in artificial intelligence at NTNU in Trondheim. He has earlier had internships in Equinor, Cisco, and Cognite where he worked on the remote control and autonomy of Boston Dynamics' Spot robot for use in the industry. Currently, he is writing his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow.

Henrik Høiness

Henrik Høiness has studied Computer Science with a specialization in Artificial Intelligence at NTNU Trondheim and UC Berkeley. His previous experience includes companies such as Yahoo!, Microsoft, Ignite Procurement, and Nabla Flow and research for Multimedia Research Group at UC Berkeley. This year, he wrote his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow. From August, Høiness will begin as Tech Lead at Ignite Procurement.

Type: short talk, 30 min Language: en Tags: Data Science, Research Audience: Beginner

Abstract:

As a data scientist, you may be asked to find the effect of a new initiative that your company trying to test. It could be a new product, a new policy, a marketing campaign or some changes to existing product. The business side gives you access to the market test data. However, you have not participated in its design, you may have or have not a control group and if you have a control group, you may find the participants in the treatment group inherently different from the control group. Can you still detect the causal effect of the new initiative? A difference in difference DID technique can be your solution!

The aforementioned case is one of many situations when randomization is not possible. DiD is a popular approach to evaluate causal effects of programs or policies when there is a lack of random assignments.

In this talk, we are going to discuss:
- Why you need experimental thinking in business world?
- What is the ideal experimental design?
- Intro to DiD technique especially when there is a selection bias!

The talk will be beneficial for those who are working with experiments ( A/B testing data) or are interested to get into this world.

The talk will assume no experience with experiments from before. Familiarity with regression analysis would be helpful but not necessary.

Speaker(s):

Sherif Ahmed

Sherif is a data scientist at Haugaland Kraft AS. He has experience with designing and analyzing social experiments in both lab and field setting.

Type: short talk, 30 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

The rapid rise of Artificial Intelligence (AI) poses fundamental challenges for the creative industry. Although AI technologies are being adopted at an ever faster pace, Design as an academic discipline has so far failed to provide a convincing answer to the opportunities and challenges of AI. As the number of interfaces between humans and information multiplies, so do the amount of design frameworks that are required to support this technology.

When it comes to the Internet of Things (IoT), it’s easy to focus on technological aspects. You can talk about different platforms or discuss which IoT solution might be the best to solve a specific problem. Looking below this layer of technology, it quickly becomes apparent that there are many more aspects that determine the success of the IoT. Not the least of which is the matter of how today’s connected products are designed.

Artificial Intelligence (AI), which was designed initially to replace highly repetitive, manual work, has exceeded expectations to complete tasks involving emotional creativity. A limiting factor of IoT is it adds devices and buttons which makes your life more complicated. Now with AI, you’re able to say things like ‘turn on the lights’ instead of pushing buttons, and it makes life simpler. It is the AI layer of natural language processing that helps IoT improve our lives.

In tapping into technology’s potential, it’s important to remember the end user — us humans. But as more and more experiences are built with ML, it’s clear that UXers still have a lot to learn about how to make users feel in control of the technology, and not the other way round. How do we create experiences that are user friendly and human-centric, while taking advantage of technology?

This talk will discuss some of the guidelines focusing on human-centered approach and can be used as reference by any UX designer to help navigate the new terrain of designing ML-driven products. As ML starts to power more and more products and experiences, let’s step up to our responsibility to stay human-centered, find the unique value for people, and make every experience great.

Speaker(s):

Siddhant Agarwal

Siddhant Agarwal currently works with Open Financial Technologies as 'Developer Relations Lead'. Previously, he has worked as Program Coordinator with the Developer Relations team at Google in India. As part of his role at Google, he had led the Developer Student Clubs program, TensorFlow User Groups, Google Developer Groups, and Google Developer Experts program in India. He also designed and executed the initiative “Build for Digital India” in 2019 launched in collaboration with the Ministry of Electronics & Information Technology, Govt. of India which witnessed 7K+ students participating and building solutions that “Solve for India”. Siddhant is one of ACM’s Distinguished Speakers among 200+ speakers worldwide and 20+ in India. A design thinker at heart, he loves working with startups and helping them scale in UX and improve their designs.

With over 6 years of industry experience, he has found his passion in ed-tech initiatives, community building, design innovation, startup ecosystem and building for the next billion users. For his contributions in community building, he was also nominated as one of the finalists of CMX Community Industry Awards 2021 for 'Community Professional of the Year for a Developer Relations Community' and 'Community Professional of the Year' categories. Being an avid public speaker, Siddhant has spoken at more than 200+ national and international forums impacting more than 70K students/developers/individuals in the last 5+ years.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

Simplifai has taken NLU to a whole new level and built a solution that read, interpret and process Emails and Documents. In the talk Erik will deep dive into the journey of how the solution now being one of the hottest AI product in Norway right now.

Tentative agenda:
- The history of AI development in Simplifai
- Chatbot -> Emailbot -> Documentbot
- Adoption of AI products in Norway, Europe and Globally
- A few key takeaways in developing AI products
-- Business case
-- Data security
-- Change management

Speaker(s):

Erik Leung

Erik leung is a visionary technologist and co-founder of AI company Simplifai who uses NLP technology for business process automation. Erik has 15 years’ experience in developing, designing, delivering and selling technical, and in recent years, AI solutions.

Type: presentation, 45 min Language: en Tags: Cloud, Kubernetes Audience: Intermediate

Abstract:

The rise of microservices, containers, and organisations and customers that expects new features delivered continuously have brought some new and interesting engineering challenges.

While a Platform Engineer is not an entirely new invention it is not just a pretty new name to the 'DevOps Engineer' or the more traditional SysAdmin role. The SysAdmin primarily concern themselves with the underlying infrastructure supporting the application and the 'DevOps Engineer' is often an ad-hock gatekeeper for getting things done on behalf of the developers for everything that involves automation. The Platform Engineer on the other hands build systems that allow teams to build on top of.

These platforms are built specifically for the developers who will be using them as a fully featured product.

Speaker(s):

Hans Kristian Flaatten

Experienced SRE- and Platform Engineering Team Lead with proven track record of leading engineering teams that build and operate Cloud Native and Kubernetes-based application platforms for customers primarily in public government, telecom, banking and insurance sectors.

Principal Consultant and DevOps Practice Lead for TietoEVRY where I drive build culture and competency within DevOps, Site Reliability Engineering (SRE) and Cloud Native practices internally and for our customers.

Type: short talk, 30 min Language: en Tags: Cloud, Kubernetes Audience: Beginner

Abstract:

This is a Kubernetes 101 talk geared toward data professionals, i will introduce Kubernetes, talk about the basics and what kind of issues it solves, focusing on things that data professionals care about

Speaker(s):

Abdellfetah SGHIOUAR

Google Cloud Engineer with a focus on Serverless, Kubernetes, and DevOps Methodologies. A supporter and contributor to OSS. Podcast host @cloudcareers.dev

Type: lightning talk, 10 min Language: en Tags: ML, Research Audience: All

Abstract:

Respiratory droplets are considered the main transmission mode of the SARS-CoV-2 virus, so social distancing has been widely adopted as a countermeasure. Individuals from different households are expected to keep a physical distance, while most public venues, such as restaurants, theaters and cinemas, are shut down completely resulting in undesirable economic and social effects. Lately, the unmatched development and application of vaccines have paved a path for reopening, but uncertainties surrounding virus mutations and vaccination rates may require to retain basic physical distancing measures for this transition period.

One of the challenges to reopening public venues in a safe and economically feasible manner is seat allocation, i.e. maximizing the number of occupiable seats while respecting distancing requirements between groups. Static layouts can efficiently be predetermined with classic optimization methods, while dynamic layouts offer more flexibility with respect to the arriving group sizes, but are also more difficult to determine.

In our work, we address the dynamic seat allocation problem with reinforcement learning (RL) -- a paradigm in machine learning of training a model via its interaction with an environment. In this lightning talk, we will share our intermediate results and experiences from this problem: first, how to model the problem and the learning environment (formulating state and action spaces, reward function, group size distribution, distancing constraints, group closeness) and, second, suitability, potentials and limitations of both classic and state-of-the-art RL methods (Q-learning, proximal policy optimization, tree search methods).

Speaker(s):

Arturs Berzins

Arturs Berzins is a PhD fellow at SINTEF, Department of Mathematics and Cybernetics and University of Oslo. His PhD project is part of the European GRAPES network and addresses the intersection of machine learning and geometry. He explores aspects of geometric representations, meta-learning and automated architecture search to increase the efficiency, performance and efficiency of machine learning systems and to address applications of geometric nature.

He received his BSc and MSc degrees with distinction at RWTH Aachen University focusing on the combined use of numerical simulations and machine learning methods in mechanical engineering.

Type: lightning talk, 10 min Language: en Tags: Cloud, Software Development Audience: Beginner

Abstract:

Whether you work as a data scientist, ML engineer or software developer, at some point you will need to develop applications – for your customers, products or as a support for the latest and coolest ML model that you have built.

With the introduction of the Cloud, a lot has changed in the way we work with the development and deployment of our applications.

In this talk, we look at 5 things to look for when creating an application in the Cloud age.

Speaker(s):

Rustam Mehmandarov

Passionate computer scientist. Java Champion and Google Developers Expert for Cloud. Public speaker. Ex-leader of JavaZone and Norwegian JUG – javaBin.

Type: short talk, 30 min Language: en Tags: Data Science, Research, History Audience: All

Abstract:

Terms like “data science” and “big data” has gone from being completely unknown, to everyday language, in just a decade. But in science, the phenomena has existed for many centuries. The mathematician Johannes Kepler inherited the vast catalogue of celestial objects from Tycho Brahe, who together formed massive mounts of data, and as result, Keppler’s laws of planetary motion. Even though there are several differences between todays and Brahe’s and Kepler’s approach, they where pioneers in data science - and the first to describe the universe with numbers and algorithms.

Speaker(s):

Vegard Lundby Rekaa

Astronomer, science communicator and data scientist. Pationate about the vastness of space and humanities limitless attempt to discover it’s fundamental properties.

Type: lightning talk, 10 min Language: en Tags: Ethics Audience: All

Abstract:

In a rapidly changing world of globalization and digitalization, computer ethics has become increasingly important. Focus on computer ethics goes beyond protecting personal and commercial information.

What are the responsibilities of companies, developers, and users, and how can we overcome the lack of awareness of computer ethics?

In this talk, I will talk about the importance of computer ethics in our daily life as a tech person and as a user in the digital world, and how we can start taking action today.

Speaker(s):

Tannaz N. Roshandel

Tannaz works as a cloud engineer for UiO, where she is building an open source cloud service for the research and education sector in Norway. She has a background in robotics and is passionate about computer ethics.

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML, Research Audience: All

Abstract:

This is an experience report on DNBs journey into transforming the organisation's culture and stepping up the competence and skills in data and analytics, corporate wise. One important track of this journey is how we are approaching the ethical part of AI, our learnings along the way, and what challenges we foresee as we increasingly leverage AI and Data in our daily business.

Speaker(s):

Aiko Yamashita

Aiko works as a Senior Data Scientist at the Center of Excellence in Advanced Analytics, at DNB. She has a background in Informatics with a PhD in Experimental Software Engineering, and +10 years of experience in Data Analytics for logistics, retail and the finance industries.

In addition to being a Data Scientist at DNB, Aiko is an Adjunct Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Analytics and has supervised students. Aiko is passionate about how innovation and technology can be better aligned to tackle current societal and environmental challenges, and how to develop technology in an ethical, humane, and just way.

Type: short talk, 30 min Language: no Tags: AI, Data Science, NLP, Semantic Technologies, RDF Audience: Intermediate

Abstract:

Da vi skulle modellere informasjon fra klartekst til semantiske kunnskapsgrafer hos Sjøfartsdirektoratet, så vi raskt at vi trengte en automatisk metode for dette. Manuell tolkning av informasjon er avhengig av menneskelig domenekunnskap for å identifisere kontekst, konsepter og relasjoner i teksten. Denne prosessen er grusomt tidkrevende, og inneholder høy risiko for misforståelser og å overse viktig informasjon. Dette foredraget vil ta deg med på turen fra klartekst til kunnskapsgrafer, via språkprosessering. Du vil høre hva vi lærte og se våre resultater.

Speaker(s):

Veronika Heimsbakk

Veronika er senior semantikknerd hos Capgemini i Oslo. Hun har en bakgrunn fra Universitetet i Oslo. Siden den tid jobbet som utvikler og informasjonsarkitet i semantikkprosjekter i inn- og utland i mange forskjellige sektorer. Hun elsker å modellere kunnskap ut fra et virrvarr av kaotiske data. Bortsett fra semantikkens finurlige verden brenner hun for realfagsformidling, LaTeX og fjellet.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML, TensorFlow.js Audience: All

Abstract:

AI can seem mysterious to neophytes, and rote for your classic scientists. In this talk I will go through an adventure to create a TensorFlow.js demo that is exciting for both.

Speaker(s):

Gant Laborde

Gant Laborde is an owner of Infinite Red, mentor, adjunct professor, published author, and award-winning speaker. For 20 years, he has been involved in software development and continues strong today. He is recognized as a Google Developer Expert in Web and Machine Learning, but informally he is an “open sourcerer” and aspires to one day become a mad scientist. He blogs, videos, and maintains popular repositories for the community. Follow Gant’s adventures at https://gantlaborde.com/

Type: short talk, 30 min Language: en Tags: ML, anomaly detection Audience: Beginner

Abstract:

Machine learning algorithms are used for condition monitoring of various types of machinery. Their flexibility enables them to learn complex patterns directly from training data in order to discern undesired operations and provide an early warning about potential equipment malfunction.

However, how to design an interpretable machine learning algorithm without knowing when problematic operations occur or even how they potentially present themselves in the data? In my talk, I will guide you through the process of combining unsupervised and supervised machine learning paradigms, as well as domain knowledge, to construct an algorithm for anomaly detection in maritime propulsion and manoeuvring systems.

Speaker(s):

Katarzyna Michalowska

Researcher in artificial intelligence at SINTEF Digital and a PhD student at University of Oslo in informed machine learning. Main areas of expertise include hybrid models of machine learning and physics, with the focus on making artificial intelligence models explainable and trustable.

Type: short talk, 30 min Language: en Tags: Data Science, ML Audience: Intermediate

Abstract:

Most progressive technology companies optimize their online products and make smart decisions utilizing thousands and thousands of A/B tests or running multi-armed bandits. However, many traditional offline businesses are still relying on simple analytics to make crucial decisions for their businesses: releasing new products, running marketing campaigns, changing their premises layouts etc. All this actions affect the market companies are working in, but how to dissociate the external factors impacting the market? In this session I will try to give examples how with the help of machine learning Coop Norway runs offline A/B tests or measures causal effects in offline retail.

Speaker(s):

Stanislav Polyakov

Stanislav Polyakov leads the Advanced Analytics team at Coop, one of the largest retailers in Norway. Stanislav primarily focuses on making Coop a data driven organization and a digital leader in the retail sector using cutting edge ML solutions.

Prior to Coop, Stanislav worked as a consultant helping many companies across the world to build their AI capabilities. His previous experiences include different Microsoft partners in Europe, Australia, New Zealand or large companies in Nordics, such as Tine SA where he was working on preventing the next butter crisis in Norway using ML.

Stanislav Polyakov has his Ph.D. in applied physics from Norwegian University of Science and Technology (NTNU) in Trondheim. His research was focused on new materials for solar cells and IR-lasers.

Type: short talk, 30 min Language: en Tags: AI, Research Audience: All

Abstract:

This talk will address why and how we should use artificial intelligence to predict the effectiveness of marketing communications, so that advertising entering the market becomes better and more effective.

Today, marketers are fighting for consumer attention in an increasingly fragmented and busy media landscape. It is increasingly important that marketing that goes into the market manages to break through the noise. Demands to prove returns on marketing dollars in both the form of sales and brand effects emphasize the importance. Nevertheless, up until now, few advertisers in Norway invest in measures to predict effectiveness and optimize their marketing campaigns before going live. This is often due to two main elements: restricted time and budget limitations.

Development of artificial intelligence helps to remove those barriers for marketers. Artificial intelligent tools with a large learning database spend only a few minutes analyzing advertisements second by second and can predict with high accuracy how the advertisement would perform in a traditional market research with “real consumers”. This provides a basis for optimization so that advertising is better received in-market and thus become more effective.

We address implications for marketers and how to move forward in the industry.

Speaker(s):

Eirik Friberg Ekrann

Eirik is Strategy Director at Kantar Norway and has 15 years of experience helping Norwegian and international companies strategically position themselves in the minds of consumers to ensure growth.

Type: presentation, 45 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

The process of building an AI looks like is so glamorous but is a long process, and at the end of the day, the tasks related to the AI model are just a 5% or less of the project.

We will see how to start an AI project from zero: defining the objectives, creating the architecture, building the game interfaces, massive data pipelines, defining model strategies, how to parallelize everything, etc.

'The abbey of crime' is an adamant 8-bit game. This game is more complicated than Montezuma Revenge and is a perfect challenge for an AI. Its complexity is about 10^1000 legal moves to solve it.

As AI technology, we will use Reinforcement Learning using Deep Neural Networks and Monte Carlo Tree Search.

The takeaways of this talk will be: understanding all the process involved to create an AI and learning the basics of Reinforcement Learning.

Speaker(s):

Juantomas Garcia

Juantomás leads Sngular's Data Science team and is also Chief Envisioning Officer. Since joining Sngular in 2018, Juantomás has leveraged his extensive experience to harness the potential of new technologies and implement them across the company’s solutions and services.

Juantomás is Google Developer Expert for Cloud and Machine Learning, co-author of the software book La Pastilla Roja and creator of “AbadIA”, the artificial intelligence platform built to solve the popular Spanish game La Abadía del Crimen. He’s an expert on free software technologies and has been a speaker at more than 200 international industry events.

He studied IT engineering at the Universidad Politécnica de Madrid and plays an active role as a tech contributor and mentor to various academic organizations and startups. He regularly organizes Machine Learning Spain and GDG cloud Madrid meetups and is a mentor at Google Launchpad for entrepreneurs.

Type: short talk, 30 min Language: en Tags: AI, Cloud, Data Science, ML Audience: Intermediate

Abstract:

It is always cheaper to retain current customers than to acquire new ones. Decreasing customer churn is therefore crucial for any subscription-based service. It is important to target customers with high risks of churn and to understand which features drive up or down the probability of churn. By understanding the most important features for churn risk for an individual customer, we can target more precise measures to improve customer experience and retention.

In this talk we will tell the story on how we went from customer data to churn predictions. We will go through all the steps of a typical machine learning project, focusing on learning points in each step. We will present the architecture of our resulting churn pipeline and give you examples of how we are going to use the predictions to create business value for TV 2.

Speaker(s):

Astrid Johnsen Tessem

Astrid Johnsen Tessem works as a data scientist at TV 2. Her responsibilities consist of data modelling, predictive analytics, and making data available across the organization. She has previously worked as a developer at Knowit Experience and has a master degree in computer science and machine learning from NTNU.

Silje Hindenes

Silje is a Data Scientist at TV 2 with a background in Mathematics and Statistics from NTNU. At TV 2 Silje is a part of the team responsible for TV 2’s analytics platform, focusing on making data and insights available across the organization

Type: short talk, 30 min Language: en Tags: ML, anomaly detection Audience: Beginner

Abstract:

Machine learning algorithms are used for condition monitoring of various types of machinery. Their flexibility enables them to learn complex patterns directly from training data in order to discern undesired operations and provide an early warning about potential equipment malfunction.

However, how to design an interpretable machine learning algorithm without knowing when problematic operations occur or even how they potentially present themselves in the data? In my talk, I will guide you through the process of combining unsupervised and supervised machine learning paradigms, as well as domain knowledge, to construct an algorithm for anomaly detection in maritime propulsion and manoeuvring systems.

Speaker(s):

Katarzyna Michalowska

Researcher in artificial intelligence at SINTEF Digital and a PhD student at University of Oslo in informed machine learning. Main areas of expertise include hybrid models of machine learning and physics, with the focus on making artificial intelligence models explainable and trustable.

Type: short talk, 30 min Language: en Tags: Cloud, Kubernetes Audience: Beginner

Abstract:

This is a Kubernetes 101 talk geared toward data professionals, i will introduce Kubernetes, talk about the basics and what kind of issues it solves, focusing on things that data professionals care about

Speaker(s):

Abdellfetah SGHIOUAR

Google Cloud Engineer with a focus on Serverless, Kubernetes, and DevOps Methodologies. A supporter and contributor to OSS. Podcast host @cloudcareers.dev

Type: short talk, 30 min Language: no Tags: AI, Data Science, NLP, Semantic Technologies, RDF Audience: Intermediate

Abstract:

Da vi skulle modellere informasjon fra klartekst til semantiske kunnskapsgrafer hos Sjøfartsdirektoratet, så vi raskt at vi trengte en automatisk metode for dette. Manuell tolkning av informasjon er avhengig av menneskelig domenekunnskap for å identifisere kontekst, konsepter og relasjoner i teksten. Denne prosessen er grusomt tidkrevende, og inneholder høy risiko for misforståelser og å overse viktig informasjon. Dette foredraget vil ta deg med på turen fra klartekst til kunnskapsgrafer, via språkprosessering. Du vil høre hva vi lærte og se våre resultater.

Speaker(s):

Veronika Heimsbakk

Veronika er senior semantikknerd hos Capgemini i Oslo. Hun har en bakgrunn fra Universitetet i Oslo. Siden den tid jobbet som utvikler og informasjonsarkitet i semantikkprosjekter i inn- og utland i mange forskjellige sektorer. Hun elsker å modellere kunnskap ut fra et virrvarr av kaotiske data. Bortsett fra semantikkens finurlige verden brenner hun for realfagsformidling, LaTeX og fjellet.

Type: short talk, 30 min Language: en Tags: AI, Cloud, ML Audience: Beginner

Abstract:

I am 14 years old (I will be 15 in May) and I have been coding and building robots since I was 9 years old. I have built 5 robots until now; ic4u and ic4u2- a robot guide dog for blind people, BB4All - a robot that is designed to help teachers and students and to prevent bullying. Hashmet the Android – the Android robot and DO the droid from Starwars.

In my session, I would like to talk about the technologies I have integrated into my robots such as Google Assistant API, Google Maps API, and Google AIY voice kit. My robots have AI capabilities such as voice recognition, image processing, voice commands, and sensor-based automated decision mechanisms. To be able to do that I used Google Cloud, Tensorflow, and Dialogflow platforms and libraries. My motivation to be a part of Confer 2021 is to be able to encourage and show other kids especially girls that technology is fun and its good to be able to use technology for human benefit.

Speaker(s):

Selin Alara Ornek

My name is Selin Alara Örnek, I am 14 years old and I am a student in the 9th grade at Irmak High School, Istanbul. I started to learn to code when I was 9 years old and started building robots when I was 10. I speak English, Turkish, and French. I am learning C++, Python and Java. When I get the chance I really enjoy attending events, talking about my projects, and receiving feedback from all different people. I have 5 robots; ic4u, ic4u2, BB4All, Hashmet the Android and DO. In 2018 I won first prize with my robot ic4u in the hardware category at Coolest Projects International, Dublin. The same year I was also a finalist in the European Youth Awards Open Innovation category with ic4u. In October 2019 I was awarded the “European Digital Girl of the Year, 2019” award in the (11-14year old category) by the Ada Awards in Brussels. In November 2019 I mentioned as one of the world's “Devkid”’s in Google Developers Blog. In May 2020, I also won the Microsoft Youth Tech Star Award in the Women Leaders in Technology Awards. I am also a young member of the New York Academy Of Sciences

Type: lightning talk, 10 min Language: en Tags: AI, ML, GAN Audience:

Abstract:

Approximating wind flows using Computational Fluid Dynamics (CFD) methods can be time-consuming. Creating a tool for interactively designing prototypes while observing the wind flow change requires simpler models to simulate output faster. Instead of running numerical approximations resulting in detailed calculations, data-driven methods in deep learning might be able to give similar results in a fraction of the time. This paper rephrases the problem from computing 3D flow fields using CFD to a 2D image-to-image translation-based problem on the building footprints to predict the flow field at pedestrian height level. We investigate the use of Generative Adversarial Network (GAN), such as Pix2Pix and CycleGAN representing state-of-the-art for image-to-image translation task in various domains as well as U-Net autoencoder. The models can learn the underlying distribution of a dataset in a data-driven manner, which we argue can help the model learn the underlying Reynolds-averaged Navier–Stokes (RANS) equations from CFD.

We experiment on novel simulated datasets on various three-dimensional bluff-shaped buildings with and without height information. Moreover, we present an extensive qualitative and quantitative evaluation of the generated images for a selection of models and compare their performance with the simulations delivered by CFD. We then show that adding positional data to the input can produce more accurate results by proposing a general framework for injecting such information on the different architectures. Furthermore, we show that the models’ performances improve by applying attention mechanisms and spectral normalization to facilitate stable training. Finally, the paper displays results on a real urban city environment dataset generated from the capital of Norway, which strongly supports the use of neural network-based methods as the underlying model for an interactive design tool.

Speaker(s):

Kristoffer Gjerde

Kristoffer Gjerde is the leader of BRAIN NTNU and a Computer Science major specializing in artificial intelligence at NTNU in Trondheim. He has earlier had internships in Equinor, Cisco, and Cognite where he worked on the remote control and autonomy of Boston Dynamics' Spot robot for use in the industry. Currently, he is writing his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow.

Henrik Høiness

Henrik Høiness has studied Computer Science with a specialization in Artificial Intelligence at NTNU Trondheim and UC Berkeley. His previous experience includes companies such as Yahoo!, Microsoft, Ignite Procurement, and Nabla Flow and research for Multimedia Research Group at UC Berkeley. This year, he wrote his master thesis on Computational Fluid Dynamics (CFD) simulations and wind flow assessment using Generative Adversarial Networks in collaboration with Norwegian Open AI Lab and Nabla Flow. From August, Høiness will begin as Tech Lead at Ignite Procurement.

Type: lightning talk, 10 min Language: en Tags: Ethics Audience: All

Abstract:

In a rapidly changing world of globalization and digitalization, computer ethics has become increasingly important. Focus on computer ethics goes beyond protecting personal and commercial information.

What are the responsibilities of companies, developers, and users, and how can we overcome the lack of awareness of computer ethics?

In this talk, I will talk about the importance of computer ethics in our daily life as a tech person and as a user in the digital world, and how we can start taking action today.

Speaker(s):

Tannaz N. Roshandel

Tannaz works as a cloud engineer for UiO, where she is building an open source cloud service for the research and education sector in Norway. She has a background in robotics and is passionate about computer ethics.

Type: lightning talk, 10 min Language: en Tags: AI, ML, Research Audience: All

Abstract:

Machine learning -- in the form of convolutional neural networks -- has come a long way, from recognizing handwritten digits in the 90s to generating convincing deepfakes and providing highly accurate medical diagnoses based on medical images. In these examples the inputs and outputs are 1D, 2D, or 3D images, or numbers.

But how can a computer generate something more complex, like a shape? In this talk we will investigate one way of doing this, using our method involving implicit spline representations. We will demonstrate how this works on a congenital heart disease data-set, where it is able to reconstruct heart and connecting vessels from CT images with state-of-the-art accuracy.

Speaker(s):

Georg Muntingh

Research Scientist at SINTEF Digital, with a focus on machine learning and geometric modelling.

Type: workshop, 120 min Language: en Tags: AI, Data Science, ML Audience: Beginner

Abstract:

This workshop is divided in two parts. For the TensorFlow portion, the goal is to give the audience the flavor of building neural networks by using Colab notebooks. For the design part, the takeaway is to discover the user impact of the technical decisions when designing and building AI systems.

The goal is to introduce elements from both fields (development and design), so folks can work better together and understand both perspectives.

Speaker(s):

A. Rosa Castillo

Dr. A. Rosa Castillo has been working many years as Software Engineer until she fell in love with Data Science. In 2018 got her first role as an AI Engineer and now is currently working as a Data Scientist. She has completed different Data Science and Deep Learning certifications, and is an active member of the tech community giving talks and organizing events as well.

Besides she is also a WomenTechmakers Ambassador, engaged in building diversity in IT, and a collaborator of DeepLearning.AI testing new DeepLearning courses in Coursera.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

Simplifai has taken NLU to a whole new level and built a solution that read, interpret and process Emails and Documents. In the talk Erik will deep dive into the journey of how the solution now being one of the hottest AI product in Norway right now.

Tentative agenda:
- The history of AI development in Simplifai
- Chatbot -> Emailbot -> Documentbot
- Adoption of AI products in Norway, Europe and Globally
- A few key takeaways in developing AI products
-- Business case
-- Data security
-- Change management

Speaker(s):

Erik Leung

Erik leung is a visionary technologist and co-founder of AI company Simplifai who uses NLP technology for business process automation. Erik has 15 years’ experience in developing, designing, delivering and selling technical, and in recent years, AI solutions.

Type: presentation, 45 min Language: en Tags: Cloud, Data Science, ML, anomaly detection Audience: All

Abstract:

Time series forecasting is a powerful way to use current or historical data to monitor, clarify, and predict “cause and effect” behaviors which can better inform future decisions. Anomaly detection for a time series can detect critical incidents such as fraudulent behaviors or potential opportunities such as changes in consumer buying patterns. Join this talk, to see the newest features in BigQuery ML in action and simplify your workflow in time forecasting and anomaly detection.

Speaker(s):

Stephanie Wang

Stephanie is a Java engineer in Developer Relations at Google Cloud Platform. Stephanie primarily focuses on building and maintaining the BigQuery Java Client Libraries which help developers interact with BigQuery APIs more easily.

She also works on managing Java dependencies across the 70+ Google Cloud Java Client Libraries to ensure that they work with each other and troubleshoot any dependency problems.

Prior to Google, Stephanie worked as a software engineer at Morgan Stanley and led a team of 10 to build sales and trading applications. Outside work, Stephanie is passionate about animals -- she has an insta-famous dog, @sasha_the_shibapoo and a opinionated semi-aquatic turtle, @bombshell_the_turtle.

Polong Lin

Developer Advocate at GCP focusing on BigQuery ML.

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML Audience: All

Abstract:

The goal of this talk is to advocate for a multidispinary approach to AI. Much like how HCI is a multidisplinary approach to Systems desifgn. - To humble Data Scientist in the audience about many aspects of Computer Science and Engineering they probably dont know, and to show how sociology, psychology and other fields are relevant to the training of ML models and AI systems more generally.

Speaker(s):

Julia Flament-Wallin

I'm a mostly back-end Software/Data Engineer with a HCI background and an interest for AI and ethics. I currently wrk in the Norwegian VR PaaS start-up Haptiq.

Type: presentation, 45 min Language: en Tags: Cloud, Kubernetes Audience: Intermediate

Abstract:

The rise of microservices, containers, and organisations and customers that expects new features delivered continuously have brought some new and interesting engineering challenges.

While a Platform Engineer is not an entirely new invention it is not just a pretty new name to the 'DevOps Engineer' or the more traditional SysAdmin role. The SysAdmin primarily concern themselves with the underlying infrastructure supporting the application and the 'DevOps Engineer' is often an ad-hock gatekeeper for getting things done on behalf of the developers for everything that involves automation. The Platform Engineer on the other hands build systems that allow teams to build on top of.

These platforms are built specifically for the developers who will be using them as a fully featured product.

Speaker(s):

Hans Kristian Flaatten

Experienced SRE- and Platform Engineering Team Lead with proven track record of leading engineering teams that build and operate Cloud Native and Kubernetes-based application platforms for customers primarily in public government, telecom, banking and insurance sectors.

Principal Consultant and DevOps Practice Lead for TietoEVRY where I drive build culture and competency within DevOps, Site Reliability Engineering (SRE) and Cloud Native practices internally and for our customers.

Type: lightning talk, 10 min Language: en Tags: ML, Research Audience: All

Abstract:

Respiratory droplets are considered the main transmission mode of the SARS-CoV-2 virus, so social distancing has been widely adopted as a countermeasure. Individuals from different households are expected to keep a physical distance, while most public venues, such as restaurants, theaters and cinemas, are shut down completely resulting in undesirable economic and social effects. Lately, the unmatched development and application of vaccines have paved a path for reopening, but uncertainties surrounding virus mutations and vaccination rates may require to retain basic physical distancing measures for this transition period.

One of the challenges to reopening public venues in a safe and economically feasible manner is seat allocation, i.e. maximizing the number of occupiable seats while respecting distancing requirements between groups. Static layouts can efficiently be predetermined with classic optimization methods, while dynamic layouts offer more flexibility with respect to the arriving group sizes, but are also more difficult to determine.

In our work, we address the dynamic seat allocation problem with reinforcement learning (RL) -- a paradigm in machine learning of training a model via its interaction with an environment. In this lightning talk, we will share our intermediate results and experiences from this problem: first, how to model the problem and the learning environment (formulating state and action spaces, reward function, group size distribution, distancing constraints, group closeness) and, second, suitability, potentials and limitations of both classic and state-of-the-art RL methods (Q-learning, proximal policy optimization, tree search methods).

Speaker(s):

Arturs Berzins

Arturs Berzins is a PhD fellow at SINTEF, Department of Mathematics and Cybernetics and University of Oslo. His PhD project is part of the European GRAPES network and addresses the intersection of machine learning and geometry. He explores aspects of geometric representations, meta-learning and automated architecture search to increase the efficiency, performance and efficiency of machine learning systems and to address applications of geometric nature.

He received his BSc and MSc degrees with distinction at RWTH Aachen University focusing on the combined use of numerical simulations and machine learning methods in mechanical engineering.

Type: short talk, 30 min Language: en Tags: ML, Research, GAN Audience: Intermediate

Abstract:

Generative adversarial networks framework immensely attracted the machine learning community attention in the recent years. GANs have succeeded in generating realistic-looking data from noise and have countless applications. On the other hand, Transfer Learning is a smart way that aims to enhance machine learning models in the case of scarce data. The main intuition behind transfer learning is to train models on a plenty of data that is “not” our targeted one but has some similar attributes to it. Then transfer learning in models that contains single networks, such as CNNs, are implemented by sharing the first layers (from the input side) of the model between source and target domains and initialize other layers then retrain the model on our scarce target data. This allows the model to have a good start when learning on the target data and achieve better results than if we trained it from scratch on the target. Now, GANs framework consists of two networks, a generator and a discriminator, and a legit question here is: which network and which part of that network contains the transferable features and hence parameter? Should we fine-tune or freeze the shared parameters? I am trying to answer this specific and direct question by showing imperial results of a series of experiments that I designed based on conditional GANs framework.

Speaker(s):

Awdelrahman M. A. Ahmed

I am a Ph.D research fellow in the department of informatics at the University of Oslo. My current research focuses on machine learning applications in smart grids and energy markets. I obtained MSc in renewable energy from Oldenburg University in Germany. Also I am a Marie-Curie alumnus.

Type: short talk, 30 min Language: en Tags: AI, Data Science, Ethics, ML, Research Audience: All

Abstract:

This is an experience report on DNBs journey into transforming the organisation's culture and stepping up the competence and skills in data and analytics, corporate wise. One important track of this journey is how we are approaching the ethical part of AI, our learnings along the way, and what challenges we foresee as we increasingly leverage AI and Data in our daily business.

Speaker(s):

Aiko Yamashita

Aiko works as a Senior Data Scientist at the Center of Excellence in Advanced Analytics, at DNB. She has a background in Informatics with a PhD in Experimental Software Engineering, and +10 years of experience in Data Analytics for logistics, retail and the finance industries.

In addition to being a Data Scientist at DNB, Aiko is an Adjunct Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Analytics and has supervised students. Aiko is passionate about how innovation and technology can be better aligned to tackle current societal and environmental challenges, and how to develop technology in an ethical, humane, and just way.

Type: lightning talk, 10 min Language: en Tags: Cloud, Software Development Audience: Beginner

Abstract:

Whether you work as a data scientist, ML engineer or software developer, at some point you will need to develop applications – for your customers, products or as a support for the latest and coolest ML model that you have built.

With the introduction of the Cloud, a lot has changed in the way we work with the development and deployment of our applications.

In this talk, we look at 5 things to look for when creating an application in the Cloud age.

Speaker(s):

Rustam Mehmandarov

Passionate computer scientist. Java Champion and Google Developers Expert for Cloud. Public speaker. Ex-leader of JavaZone and Norwegian JUG – javaBin.

Type: short talk, 30 min Language: en Tags: AI, Data Science, ML Audience: Intermediate

Abstract:

The rapid rise of Artificial Intelligence (AI) poses fundamental challenges for the creative industry. Although AI technologies are being adopted at an ever faster pace, Design as an academic discipline has so far failed to provide a convincing answer to the opportunities and challenges of AI. As the number of interfaces between humans and information multiplies, so do the amount of design frameworks that are required to support this technology.

When it comes to the Internet of Things (IoT), it’s easy to focus on technological aspects. You can talk about different platforms or discuss which IoT solution might be the best to solve a specific problem. Looking below this layer of technology, it quickly becomes apparent that there are many more aspects that determine the success of the IoT. Not the least of which is the matter of how today’s connected products are designed.

Artificial Intelligence (AI), which was designed initially to replace highly repetitive, manual work, has exceeded expectations to complete tasks involving emotional creativity. A limiting factor of IoT is it adds devices and buttons which makes your life more complicated. Now with AI, you’re able to say things like ‘turn on the lights’ instead of pushing buttons, and it makes life simpler. It is the AI layer of natural language processing that helps IoT improve our lives.

In tapping into technology’s potential, it’s important to remember the end user — us humans. But as more and more experiences are built with ML, it’s clear that UXers still have a lot to learn about how to make users feel in control of the technology, and not the other way round. How do we create experiences that are user friendly and human-centric, while taking advantage of technology?

This talk will discuss some of the guidelines focusing on human-centered approach and can be used as reference by any UX designer to help navigate the new terrain of designing ML-driven products. As ML starts to power more and more products and experiences, let’s step up to our responsibility to stay human-centered, find the unique value for people, and make every experience great.

Speaker(s):

Siddhant Agarwal

Siddhant Agarwal currently works with Open Financial Technologies as 'Developer Relations Lead'. Previously, he has worked as Program Coordinator with the Developer Relations team at Google in India. As part of his role at Google, he had led the Developer Student Clubs program, TensorFlow User Groups, Google Developer Groups, and Google Developer Experts program in India. He also designed and executed the initiative “Build for Digital India” in 2019 launched in collaboration with the Ministry of Electronics & Information Technology, Govt. of India which witnessed 7K+ students participating and building solutions that “Solve for India”. Siddhant is one of ACM’s Distinguished Speakers among 200+ speakers worldwide and 20+ in India. A design thinker at heart, he loves working with startups and helping them scale in UX and improve their designs.

With over 6 years of industry experience, he has found his passion in ed-tech initiatives, community building, design innovation, startup ecosystem and building for the next billion users. For his contributions in community building, he was also nominated as one of the finalists of CMX Community Industry Awards 2021 for 'Community Professional of the Year for a Developer Relations Community' and 'Community Professional of the Year' categories. Being an avid public speaker, Siddhant has spoken at more than 200+ national and international forums impacting more than 70K students/developers/individuals in the last 5+ years.

Type: presentation, 45 min Language: en Tags: AI, Data Science, Ethics, ML, Security Audience: Intermediate

Abstract:

With the constant progress of AI, ML has now become all around us in real life apps instead of just research papers.

Doing so has given rise to a new type of issue that wasn't a problem before: Robustness.

With the nearly unlimited possibilities of the real world and its constant changes, Deep Learning models are frequently subject to tricky data points that cause it to give wrong predictions, ranging from very unfortunate events to people messing with them on purpose.

To mitigate this, robustness in our Neual Networks has become a must. We need to address this issue and test our models extensively before going to production.

In this talk, we will be introducting robustness and give examples of how easily our networks can be fooled, thus shedding light on how we can work around this.

Speaker(s):

Elyes Manai

Elyes is an Artificial Intelligence consultant and teacher as well as a Google Developer Expert in Machine Learning.

He enjoys teaching companies how to use AI/Data to improve their services and make more money.

Currently, he is a mentor at Google For Startups Accelerator and the co-founder of PyData Tunisia.

But most importantly, he is enjoying every part of it.

Type: short talk, 30 min Language: en Tags: Data Science, Research, History Audience: All

Abstract:

Terms like “data science” and “big data” has gone from being completely unknown, to everyday language, in just a decade. But in science, the phenomena has existed for many centuries. The mathematician Johannes Kepler inherited the vast catalogue of celestial objects from Tycho Brahe, who together formed massive mounts of data, and as result, Keppler’s laws of planetary motion. Even though there are several differences between todays and Brahe’s and Kepler’s approach, they where pioneers in data science - and the first to describe the universe with numbers and algorithms.

Speaker(s):

Vegard Lundby Rekaa

Astronomer, science communicator and data scientist. Pationate about the vastness of space and humanities limitless attempt to discover it’s fundamental properties.

Type: presentation, 45 min Language: en Tags: AI, Data Science, ML, TensorFlow.js Audience: All

Abstract:

AI can seem mysterious to neophytes, and rote for your classic scientists. In this talk I will go through an adventure to create a TensorFlow.js demo that is exciting for both.

Speaker(s):

Gant Laborde

Gant Laborde is an owner of Infinite Red, mentor, adjunct professor, published author, and award-winning speaker. For 20 years, he has been involved in software development and continues strong today. He is recognized as a Google Developer Expert in Web and Machine Learning, but informally he is an “open sourcerer” and aspires to one day become a mad scientist. He blogs, videos, and maintains popular repositories for the community. Follow Gant’s adventures at https://gantlaborde.com/

Type: short talk, 30 min Language: en Tags: Data Science, Research Audience: Beginner

Abstract:

As a data scientist, you may be asked to find the effect of a new initiative that your company trying to test. It could be a new product, a new policy, a marketing campaign or some changes to existing product. The business side gives you access to the market test data. However, you have not participated in its design, you may have or have not a control group and if you have a control group, you may find the participants in the treatment group inherently different from the control group. Can you still detect the causal effect of the new initiative? A difference in difference DID technique can be your solution!

The aforementioned case is one of many situations when randomization is not possible. DiD is a popular approach to evaluate causal effects of programs or policies when there is a lack of random assignments.

In this talk, we are going to discuss:
- Why you need experimental thinking in business world?
- What is the ideal experimental design?
- Intro to DiD technique especially when there is a selection bias!

The talk will be beneficial for those who are working with experiments ( A/B testing data) or are interested to get into this world.

The talk will assume no experience with experiments from before. Familiarity with regression analysis would be helpful but not necessary.

Speaker(s):

Sherif Ahmed

Sherif is a data scientist at Haugaland Kraft AS. He has experience with designing and analyzing social experiments in both lab and field setting.

Type: short talk, 30 min Language: en Tags: Data Science, ML Audience: Intermediate

Abstract:

Most progressive technology companies optimize their online products and make smart decisions utilizing thousands and thousands of A/B tests or running multi-armed bandits. However, many traditional offline businesses are still relying on simple analytics to make crucial decisions for their businesses: releasing new products, running marketing campaigns, changing their premises layouts etc. All this actions affect the market companies are working in, but how to dissociate the external factors impacting the market? In this session I will try to give examples how with the help of machine learning Coop Norway runs offline A/B tests or measures causal effects in offline retail.

Speaker(s):

Stanislav Polyakov

Stanislav Polyakov leads the Advanced Analytics team at Coop, one of the largest retailers in Norway. Stanislav primarily focuses on making Coop a data driven organization and a digital leader in the retail sector using cutting edge ML solutions.

Prior to Coop, Stanislav worked as a consultant helping many companies across the world to build their AI capabilities. His previous experiences include different Microsoft partners in Europe, Australia, New Zealand or large companies in Nordics, such as Tine SA where he was working on preventing the next butter crisis in Norway using ML.

Stanislav Polyakov has his Ph.D. in applied physics from Norwegian University of Science and Technology (NTNU) in Trondheim. His research was focused on new materials for solar cells and IR-lasers.

Type: short talk, 30 min Language: en Tags: AI, ML, Research Audience: Intermediate

Abstract:

An obstacle often encountered in computer vision applications is the time it takes to collect sufficient amounts of data. We'll talk about how Spoor use synthetic (rendered) data to training AI systems for detecting and classifying birds on wind farms with applications to environmental impact studies and operations.

Speaker(s):

Helge Reikerås

Helge is an experienced data scientist and AI engineer with 10+ years of experience. Currently he's the CTO at Oslo startup Spoor.

Type: short talk, 30 min Language: en Tags: AI, Research Audience: All

Abstract:

This talk will address why and how we should use artificial intelligence to predict the effectiveness of marketing communications, so that advertising entering the market becomes better and more effective.

Today, marketers are fighting for consumer attention in an increasingly fragmented and busy media landscape. It is increasingly important that marketing that goes into the market manages to break through the noise. Demands to prove returns on marketing dollars in both the form of sales and brand effects emphasize the importance. Nevertheless, up until now, few advertisers in Norway invest in measures to predict effectiveness and optimize their marketing campaigns before going live. This is often due to two main elements: restricted time and budget limitations.

Development of artificial intelligence helps to remove those barriers for marketers. Artificial intelligent tools with a large learning database spend only a few minutes analyzing advertisements second by second and can predict with high accuracy how the advertisement would perform in a traditional market research with “real consumers”. This provides a basis for optimization so that advertising is better received in-market and thus become more effective.

We address implications for marketers and how to move forward in the industry.

Speaker(s):

Eirik Friberg Ekrann

Eirik is Strategy Director at Kantar Norway and has 15 years of experience helping Norwegian and international companies strategically position themselves in the minds of consumers to ensure growth.