Program

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

Abstract:

When will my package arrive? At Posten, we use machine learning to estimate delivery times. The current solution is built on a summary of the production flow that starts with the sender, goes on to a first terminal, the shipment then transits to a second terminal, from where it is delivered out to the receiver. However, the data collected during the production flow looks nothing like this, requiring a 'translation-step' to make it concur with the given simplified description. Even though the end solution performs very well, the amount of work that goes into translating the raw data is manual, and has become a very large chore when adapting to each service and their unique properties. Not only is it time-consuming, it also requires a lot of knowledge about business logic.

In this talk we will describe a new approach to modelling delivery time estimates. Inspired by more advanced machine learning algorithms, we have tried to solve this issue by working with the data in its rawer form. The goal is not necessarily to improve the results, but rather to minimize the manual preprocessing step. The new model will pick up local variations in production, and integrate business logic into its predictions. Our aim is to lower significantly the requirements on understanding all the production processes that can occur within the different services, as well as freeing up time that is spent on manual data cleaning.

Speaker(s):

Simon Simonsson

Originally from Stockholm, Sweden. After almost 10 years of experience within software engineering including small applications, web and hardware development; Simon Simonsson graduated with a MSc in Computer Science from KTH in Stockholm, Sweden in 2019. Since then he has focused on more data driven solutions, working as a Data Scientist and part-time research assistant at the Oslo Metropolitan University researching on smart homes for elderly care. Today he works full-time as a Data Scientist in the innovation department at Posten AS working on among other things, making the best possible predictions on delivery times for customers. Interests include machine learning, automation, data visualization and robotics.

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

Abstract:

What does AI have in common with Sustainability? This talk describes how Data and AI technologies are currently being leveraged at DNB to create the ESG Data Hub, with the mission to fuel the transition to a green economy. She will provide concrete examples on how the analysis of ESG data can have massive positive implications on global economy, and will provide concrete examples on why Responsible AI and Sustainability are converging in many unforeseen ways.

Speaker(s):

Aiko Yamashita

Aiko has a background in Informatics with a doctoral degree in experimental software engineering. She has +10 years of experience in data analytics within the logistics, retail and finance industries. She is currently leading the ESG Data Task Force, which has as mission to ensure top-of-class ESG analytics and timely access to ESG data for the Corporate. She believes in responsible use of Data, AI, and Technology, and is a passionate advocate for data literacy in DNB. In addition to being a Data Scientist at DNB, Aiko is an Adjunct Associate Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Science and supervised students.

Type: presentation, 45 min Language: en Tags: Data Science Audience: Beginner

Abstract:

Imagine a world where data is machine readable, connected and contain meaning for both humans and machine. By structuring data as directed graphs, and applying a layer of propositional logic, to explore strange new insight, to seek out new connections and new knowledge, to boldly go where no man has gone before.

Speaker(s):

Veronika Heimsbakk

I adore working with messy, unstructured and chaotic data, and turning them into a wonder of semantic knowledge graphs. Apart from the wonders of semantics and logic, I love making and soldering stuff, LaTeX, and teaching kids to code.

Type: presentation, 45 min Language: en Tags: AI, Ethics, ML Audience: All

Abstract:

Technology is embedded in our lives, and Artificial Intelligence will have an ever greater impact on the development of humanity. As creators of technology, are we prepared for the responsibility that this entails, as we try to create machines that emulate human reasoning and behaviour? Let's find out what Isaac Asimov, one of the greatest writers of Science Fiction, had to say about it.

Speaker(s):

Patty O'Callaghan

Patty is a Venezuelan living in Scotland. Her background education is Computer Science, in the field of Artificial Intelligence. She has 20 years of experience working in Tech and works as a Software Engineer at Charles River Laboratories. Patty is very passionate about inspiring girls into joining Computer Science, which is why she is a volunteer at Code Club Scotland and she is also an advisor at WeAndAI.org, where she hopes to bring discussions of artificial intelligence into her work with young people, and ultimately assist in the construction of an educational framework. She is a Lead @ GDG Glasgow and a member of Google's Women Techmakers.

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

Abstract:

I will be talking about AI in Healthcare: The Hype, Hope and the Hurdles. I will touch upon the endless possibilities AI represents, how AI can be implemented into clinical practice in the future and the ethical challenges with AI in Medicine.

Speaker(s):

Ishita Barua

Ishita Barua is a MD and a PhD Candidate in the Clinical Effectiveness Research Group at the University of Oslo and Oslo University Hospital. Her research focuses on the use of artificial intelligence in colorectal cancer diagnosis and colonoscopy. She is one of the initiators behind and lecturers in the medical elective subject called MED3065: AI, innovation, big data and clinical decision support.

She was awarded the Fulbright Scholarship in 2020, and was a visiting researcher at BIDMC and Harvard Medical School in Boston, USA, in 2021.

She is currently a deputy member of the Norwegian Biotechnology Advisory Board, appointed by the Norwegian government. In 2021, she was named one of top 30 women in Norway changing the field of Artificial Intelligence, by NORA - The Norwegian Artificial Intelligence Research Consortium, and in 2022 she was named one of Norway's leading Tech-Women by Abelia and ODA. She has previously worked as a Resident Doctor in Gastroenterology and Internal Medicine, and served as the President of Young Internists of Norway.

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

Abstract:

Recent improvements in machine learning (ML) have enabled the application of artificial intelligence (AI) in many different areas, resulting in significant achievements in computational vision, speech recognition or protein analysis.

Being amazing, these techniques suffer significant limitations, presenting what some call 'diminishing returns'. In particular, machine learning often only recognizes a pattern it has seen before, 'catastrophic forgetting', overwriting past knowledge with new knowledge, lacking explanation of the train of its thoughts or basic common sense.

For all these reasons, in comparative terms, all of these results correspond to basic human perception hugely empowered with computational resources. Evolution of the artificial intelligence requires additional techniques such as neuro symbolic computations or causal inference.

In this short presentation, after a broader definition of artificial intelligence, causal inference is introduced, explaining its main features and methods, proposing a model for a causal inference engine and describing the main elements of a prototype implemented in Kotlin.

Speaker(s):

Andrés-Leonardo Martínez-Ortiz

Andrés-Leonardo Martínez-Ortiz, better known as almo, holds a PhD on Software, Systems, and Computing and is a member of the Google Machine Learning Site Reliability Engineering team where he leads several programs aiming for reliability & convergence. He is also a member of IEEE, ACM, the Linux Foundation and Computer Society.

Type: presentation, 45 min Language: en Tags: Cloud, ML Audience: Beginner

Abstract:

With the exploding interests in Machine Learning technology in the recent years, they have also introduced many aspects of computing that are required to support such a grand vision in what ML is capable of delivering. On the infrastructural layer, we need to handle the high frequency of data ingestion with low latency, and one of the best mechanisms that we can think of leveraging on is streaming. So what is streaming, and what are the different choices we have as a platform? We will learn about a few options, and zoom in to learn more about what a true next-generation, cloud-native streaming platform such as Apache Pulsar is capable of, beyond what the more common messaging platforms that we have these days.

Speaker(s):

Mary Grygleski

Mary is a Java Champion and a passionate Streaming Developer Advocate at DataStax, a leading data management company that champions Open Source software and specializes in Big Data, DB-as-a-service, Streaming, and Cloud-Native systems. She spent 3.5 years previously as a very effective advocate at IBM, focusing on Java, Jakarta EE, OpenJ9, Open Source, Cloud, and Distributed Systems. She transitioned from Unix/C to Java around 2000 and has never looked back since then. She considers herself a polyglot and loves to continue learning new and better ways to solve real-life problems. She is an active tech community builder outside of her day job, and currently the President of the Chicago Java Users Group (CJUG), as well as a co-organizer for several IBM-sponsored meetup groups in the Greater Chicago area.

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

Abstract:

CI/CD tools are multiples and all have their strenghts and weaknesses. Cloud Build is the fully managed CI/CD solution of Google Cloud that comes with a lot with great feature: scalable, serverless, free tiers,...

To leverage its full power, I propose a deep dive in its architecture. The design choice has consequences and requires specific patterns that I present to avoid mistake and loosing time and to speed up the development of pipelines

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

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

Abstract:

Event driven is a new way to process and ingest the data. Serverless is a new paradigm to abstract the technical layer and to focus on what matters: the business processes.
Both combined, you can create a powerful and scalable data pipeline on Google Cloud. You can react to events, trigger the next step with ease and just in time.
The reality isn't so nice, and it quickly becomes a nightmare to understand and to update the pipelines. Workflows is a new product on Google Cloud that orchestrates the processes.
Come to discover what was the serverless data pipelines before and how it's smoother with Workflow!

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

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

Abstract:

Do you want YOUR app easily deployable in Kubernetes or Openshift? Do you want to control how it scale, deploy and update in an automated way? Do you want to have your app in a cloud market? Then Kubernetes Operators are for YOU.

In this session I will show you What's a K8s Operator, How to create it with Java, Quarkus and native compilation, and how to deploy it on OperatorHub, from my own experience creating this production ready Operator for a FOSS application, including the testing GitHub Actions.

Speaker(s):

Jonathan Vila

Java Champion, Organiser at BarcelonaJUG and cofounder of the JBCNConf conference in Barcelona.

Type: short talk, 30 min Language: en Tags: Cloud, Data Science Audience: Advanced

Abstract:

This session will deliver best practices and recommendations for tuning BI Engine for your existing BigQuery workloads for cheaper and faster queries. Learn how we at REEA are orchestrating BI Engine reservations, on a 5TB dataset, considered small for BigQuery but with big cost savings and accelerated queries. We are seeing many presentations for big enterprises, but now we are showcasing how our queries perform better with lower costs. We are going to address the top considerations when to turn on BI Engine, how to use cloud orchestration for making this an automatic process, and combined with BigQuery and Datastudio query complexity that might save precious development time, lower bills, faster queries.

Speaker(s):

Marton Kodok

Speaker at conferences, a Google Developer Expert, champion of the Cloud Innovators program, top user on Stackoverflow, technical software architect at REEA.net, co-founder of IT Mures, life-long learner influencer, voice of local community on digitalization, social responsibility advocate of transparency, frequent publisher.

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

Abstract:

As introduction, we will explain Microsoft's approach from Research to AI embedded products. After a short Product overview, we explain the differences between scenario-based AI, customizable AI Models and Custom Models. The main focus of the talk is about Responsible AI. How it is ensured through design guidelines and tools. Why we use tools like Fairlearn, InterpretML, Error Analysis and Counterfit…and even more important. Why you should use these Open-Source tools to bring your models to the next level.

Speaker(s):

Lars Ruddigkeit

I am Cloud Solution Architect for Data and AI at Microsoft. Meaning I work hands-on on AI projects with large enterprise customers, very often Fortune 500. My focus is 90% AI, 10% Data and always security. FSI is heavily regulated so no fund with AI without security. My Ph.D. is in Chemistry but I use roughly 1% from my studies and Ph.D. times.

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

Abstract:

Ruter uses machine learning to classify and auto-reply to customer complaints, predict arrival times, and forecast passenger numbers for over half a million departures every day. Our first model was deployed in the summer of 2020. Updating it was time-consuming, and monitoring was limited and manual. Today, we have a robust, reproducible, and scalable framework for developing and deploying machine learning models on AWS. The framework maintains a reproducible history of data, models, and predictions. It monitors pipelines for failures and alerts the team of reduced model quality in real-time. Using this framework, data scientists and machine learning engineers can quickly turn a proof of concept into a value-generating application. In this talk, we will present our framework, its associated tech stack, and how we use it to develop, support, and monitor models in production.

Speaker(s):

David Skålid Amundsen

David is an experienced data scientist who is passionate about building robust and trustworthy machine learning solutions. For the past year and a half he has been working for Ruter on establishing just this.

Jawad Saleemi

Jawad Saleemi is an experienced data scientist and is team lead of the machine learning team at Ruter.

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

Abstract:

The CAP theorem is widely known for distributed systems, but it's not the only tradeoff you should be aware of. For datastores there is also the FAB theory and just like with the CAP theorem you can only pick two:
* Fast: Results are real-time or near real-time instead of batch oriented.
* Accurate: Answers are exact and don't have a margin of error.
* Big: You require horizontal scaling and need to distribute your data.

While Fast and Big are relatively easy to understand, Accurate is a bit harder to picture. This talk shows some concrete examples of accuracy tradeoffs Elasticsearch can take for terms aggregations, cardinality aggregations with HyperLogLog++, and the IDF part of full-text search. Or how to trade some speed or the distribution for more accuracy.

Speaker(s):

Philipp Krenn

Philipp lives to demo interesting technology. Having worked as a web, infrastructure, and database engineer for over ten years, Philipp is now a developer advocate and EMEA team lead at Elastic — the company behind the Elastic Stack consisting of Elasticsearch, Kibana, Beats, and Logstash.

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

Abstract:

Production machine learning is already bringing value to companies and organizations across the world, but at what environmental cost? In this talk we will learn how to estimate the carbon emissions for machine learning pipelines. We will also see some practical approaches to reducing the carbon footprint of your machine learning efforts.

Speaker(s):

Magdalena Stenius

Magdalena is a machine learning engineer at Valohai and an active member of the Finnish female and nonbinary programmers community as well as the co-host of Turku.py, her home town's local python meetup. She is passionate about environmental topics and fairness in the context of ML and AI.

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

Abstract:

When machine learning models are deployed to production, their performance starts degrading. Now that ML models are increasingly becoming mission critical for enterprises and startups alike, root cause analysis and gaining observability into your AI systems is similarly mission critical. However, many organizations struggle to prevent model performance degradation and assure the quality of the data being fed to their ML models, largely because they don't have the tools and organizational knowledge to do so.

In this talk, MLOps Architect Danny D. Leybzon will explain the problems associated with ML models deployed in production, and how many of these problems can be addressed with data monitoring and AI observability best practices. Taking it a step further, the speaker will discuss steps that data scientists and machine learning engineers can take to proactively ensure the performance of their models, rather than reacting to the impacts of performance degradation reported by their customers.

Speaker(s):

Danny D. Leybzon

Danny D. Leybzon has worn many hats, all of them related to data. He studied computational statistics at UCLA and has worked in the data and ML space ever since. In his role as MLOps architect, he has worked to evangelize machine learning best practices, talking on subjects such as distributed deep learning, productionizing machine learning models, automated machine learning, and lately has been talking about AI observability and data logging. When Danny's not researching, practicing, or talking about data science, he's usually doing one of his numerous outside hobbies: rock climbing, backcountry backpacking, skiing, etc.

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

Abstract:

Let's merge domain-specific models with generic learning. With Julia it is easy to combine these different worlds, differential equations on the one hand and (deep) neural networks on the other. The combination is known as scientific machine learning and it brings immense benefits to all types of industries.

Neural differential equations are just one example out of this rich ecosystem. The core idea is to predict a time-series by modelling its derivative with a neural network. We will briefly explain the theory behind neural differential equations, why Julia is especially suited for this, and experience its power for predicting time-series in practice.

Speaker(s):

Stephan Sahm

Stephan Sahm is founder of Jolin.io, an IT consultancy for high-performant data/ml pipelines powered by Julia. He organises the biggest Julia meetup in Germany, the Julia-User-Group-Munich.
With 10 years experience in data science and 5 years in IT consultancy and datastartups, he strives to bring tomorrows best practice into todays business.

His expertise is in big-data and real-time processing, optionally empowered by machine learning. From PoC via MVP to production, all stages and requirements he supported. Industries include automotive, telecommunication, ecommerce, and media, both cloud-based and on-premise. He implemented use cases with time series, natural language processing, recommendation systems and anomaly detection using Python, R, Scala, and Julia.

Type: presentation, 45 min Language: en Tags: AI, Ethics Audience: All

Abstract:

To seize the opportunities AI offers society, while limiting its risks and ensuring citizen protection, we need to bring abstract ethical principles into practice. A group of scientists and tech businesses are working on an AI Trust Label to make this possible. It specifies and operationalizes values like transparency, privacy and fairness, to allow for effective and enforceable ethics guidelines.

In this talk you will get an overview over the recently published standard VDE SPEC 90012 (VCIO based description of systems for AI trustworthiness characterization). It is a framework to describe whether a product adheres to specific values and can be trusted. End consumers, companies and government organizations can use the description to define requirements or to compare different products. The standard is the basis for a future AI Trust Label or certification mechanisms. The presentation will also discuss the opportunities and limitations of operationalizing trust or ethical values for AI systems.

Speaker(s):

Michael Puntschuh

Michael is a freelance researcher, speaker and advocate on human rights in our digitized world. He is a social scientist working on the issues of human rights online and tech governance as well as other societal and legal aspects of digital technologies. His work is currently focused on criteria for the design of algorithmic systems, public innovation, digital and AI ethics and algorithmically mediated discrimination. As part of the AI Ethics Impact Group, he has worked on the development of an AI Trust Label.

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

Abstract:

I'd like to take you through a v short history of the semantic web (failure?), the semantic technologies we have as a result of this effort, the rise of graph databases and knowledge graphs as well as the new data mesh data architecture paradigm and what it might mean for data architecture and in terms of possibilities for data science going forward. For example Gaia-X, EUs New project fordevelopment of a federation of data infrastructure and service providers for Europe with the objective of ensuring a European digital sovereignty. Or identity using graphs and zero knowledge proofs.

Speaker(s):

Julia Flament-Wallin

I'm a developer and lead in the intersection between data science and software engineering.

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

Abstract:

When will my package arrive? At Posten, we use machine learning to estimate delivery times. The current solution is built on a summary of the production flow that starts with the sender, goes on to a first terminal, the shipment then transits to a second terminal, from where it is delivered out to the receiver. However, the data collected during the production flow looks nothing like this, requiring a 'translation-step' to make it concur with the given simplified description. Even though the end solution performs very well, the amount of work that goes into translating the raw data is manual, and has become a very large chore when adapting to each service and their unique properties. Not only is it time-consuming, it also requires a lot of knowledge about business logic.

In this talk we will describe a new approach to modelling delivery time estimates. Inspired by more advanced machine learning algorithms, we have tried to solve this issue by working with the data in its rawer form. The goal is not necessarily to improve the results, but rather to minimize the manual preprocessing step. The new model will pick up local variations in production, and integrate business logic into its predictions. Our aim is to lower significantly the requirements on understanding all the production processes that can occur within the different services, as well as freeing up time that is spent on manual data cleaning.

Speaker(s):

Simon Simonsson

Originally from Stockholm, Sweden. After almost 10 years of experience within software engineering including small applications, web and hardware development; Simon Simonsson graduated with a MSc in Computer Science from KTH in Stockholm, Sweden in 2019. Since then he has focused on more data driven solutions, working as a Data Scientist and part-time research assistant at the Oslo Metropolitan University researching on smart homes for elderly care. Today he works full-time as a Data Scientist in the innovation department at Posten AS working on among other things, making the best possible predictions on delivery times for customers. Interests include machine learning, automation, data visualization and robotics.

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

Abstract:

What does AI have in common with Sustainability? This talk describes how Data and AI technologies are currently being leveraged at DNB to create the ESG Data Hub, with the mission to fuel the transition to a green economy. She will provide concrete examples on how the analysis of ESG data can have massive positive implications on global economy, and will provide concrete examples on why Responsible AI and Sustainability are converging in many unforeseen ways.

Speaker(s):

Aiko Yamashita

Aiko has a background in Informatics with a doctoral degree in experimental software engineering. She has +10 years of experience in data analytics within the logistics, retail and finance industries. She is currently leading the ESG Data Task Force, which has as mission to ensure top-of-class ESG analytics and timely access to ESG data for the Corporate. She believes in responsible use of Data, AI, and Technology, and is a passionate advocate for data literacy in DNB. In addition to being a Data Scientist at DNB, Aiko is an Adjunct Associate Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Science and supervised students.

Type: presentation, 45 min Language: en Tags: Data Science Audience: Beginner

Abstract:

Imagine a world where data is machine readable, connected and contain meaning for both humans and machine. By structuring data as directed graphs, and applying a layer of propositional logic, to explore strange new insight, to seek out new connections and new knowledge, to boldly go where no man has gone before.

Speaker(s):

Veronika Heimsbakk

I adore working with messy, unstructured and chaotic data, and turning them into a wonder of semantic knowledge graphs. Apart from the wonders of semantics and logic, I love making and soldering stuff, LaTeX, and teaching kids to code.

Type: presentation, 45 min Language: en Tags: AI, Ethics, ML Audience: All

Abstract:

Technology is embedded in our lives, and Artificial Intelligence will have an ever greater impact on the development of humanity. As creators of technology, are we prepared for the responsibility that this entails, as we try to create machines that emulate human reasoning and behaviour? Let's find out what Isaac Asimov, one of the greatest writers of Science Fiction, had to say about it.

Speaker(s):

Patty O'Callaghan

Patty is a Venezuelan living in Scotland. Her background education is Computer Science, in the field of Artificial Intelligence. She has 20 years of experience working in Tech and works as a Software Engineer at Charles River Laboratories. Patty is very passionate about inspiring girls into joining Computer Science, which is why she is a volunteer at Code Club Scotland and she is also an advisor at WeAndAI.org, where she hopes to bring discussions of artificial intelligence into her work with young people, and ultimately assist in the construction of an educational framework. She is a Lead @ GDG Glasgow and a member of Google's Women Techmakers.

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

Abstract:

I will be talking about AI in Healthcare: The Hype, Hope and the Hurdles. I will touch upon the endless possibilities AI represents, how AI can be implemented into clinical practice in the future and the ethical challenges with AI in Medicine.

Speaker(s):

Ishita Barua

Ishita Barua is a MD and a PhD Candidate in the Clinical Effectiveness Research Group at the University of Oslo and Oslo University Hospital. Her research focuses on the use of artificial intelligence in colorectal cancer diagnosis and colonoscopy. She is one of the initiators behind and lecturers in the medical elective subject called MED3065: AI, innovation, big data and clinical decision support.

She was awarded the Fulbright Scholarship in 2020, and was a visiting researcher at BIDMC and Harvard Medical School in Boston, USA, in 2021.

She is currently a deputy member of the Norwegian Biotechnology Advisory Board, appointed by the Norwegian government. In 2021, she was named one of top 30 women in Norway changing the field of Artificial Intelligence, by NORA - The Norwegian Artificial Intelligence Research Consortium, and in 2022 she was named one of Norway's leading Tech-Women by Abelia and ODA. She has previously worked as a Resident Doctor in Gastroenterology and Internal Medicine, and served as the President of Young Internists of Norway.

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

Abstract:

Recent improvements in machine learning (ML) have enabled the application of artificial intelligence (AI) in many different areas, resulting in significant achievements in computational vision, speech recognition or protein analysis.

Being amazing, these techniques suffer significant limitations, presenting what some call 'diminishing returns'. In particular, machine learning often only recognizes a pattern it has seen before, 'catastrophic forgetting', overwriting past knowledge with new knowledge, lacking explanation of the train of its thoughts or basic common sense.

For all these reasons, in comparative terms, all of these results correspond to basic human perception hugely empowered with computational resources. Evolution of the artificial intelligence requires additional techniques such as neuro symbolic computations or causal inference.

In this short presentation, after a broader definition of artificial intelligence, causal inference is introduced, explaining its main features and methods, proposing a model for a causal inference engine and describing the main elements of a prototype implemented in Kotlin.

Speaker(s):

Andrés-Leonardo Martínez-Ortiz

Andrés-Leonardo Martínez-Ortiz, better known as almo, holds a PhD on Software, Systems, and Computing and is a member of the Google Machine Learning Site Reliability Engineering team where he leads several programs aiming for reliability & convergence. He is also a member of IEEE, ACM, the Linux Foundation and Computer Society.

Type: presentation, 45 min Language: en Tags: Cloud, ML Audience: Beginner

Abstract:

With the exploding interests in Machine Learning technology in the recent years, they have also introduced many aspects of computing that are required to support such a grand vision in what ML is capable of delivering. On the infrastructural layer, we need to handle the high frequency of data ingestion with low latency, and one of the best mechanisms that we can think of leveraging on is streaming. So what is streaming, and what are the different choices we have as a platform? We will learn about a few options, and zoom in to learn more about what a true next-generation, cloud-native streaming platform such as Apache Pulsar is capable of, beyond what the more common messaging platforms that we have these days.

Speaker(s):

Mary Grygleski

Mary is a Java Champion and a passionate Streaming Developer Advocate at DataStax, a leading data management company that champions Open Source software and specializes in Big Data, DB-as-a-service, Streaming, and Cloud-Native systems. She spent 3.5 years previously as a very effective advocate at IBM, focusing on Java, Jakarta EE, OpenJ9, Open Source, Cloud, and Distributed Systems. She transitioned from Unix/C to Java around 2000 and has never looked back since then. She considers herself a polyglot and loves to continue learning new and better ways to solve real-life problems. She is an active tech community builder outside of her day job, and currently the President of the Chicago Java Users Group (CJUG), as well as a co-organizer for several IBM-sponsored meetup groups in the Greater Chicago area.

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

Abstract:

CI/CD tools are multiples and all have their strenghts and weaknesses. Cloud Build is the fully managed CI/CD solution of Google Cloud that comes with a lot with great feature: scalable, serverless, free tiers,...

To leverage its full power, I propose a deep dive in its architecture. The design choice has consequences and requires specific patterns that I present to avoid mistake and loosing time and to speed up the development of pipelines

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

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

Abstract:

Event driven is a new way to process and ingest the data. Serverless is a new paradigm to abstract the technical layer and to focus on what matters: the business processes.
Both combined, you can create a powerful and scalable data pipeline on Google Cloud. You can react to events, trigger the next step with ease and just in time.
The reality isn't so nice, and it quickly becomes a nightmare to understand and to update the pipelines. Workflows is a new product on Google Cloud that orchestrates the processes.
Come to discover what was the serverless data pipelines before and how it's smoother with Workflow!

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

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

Abstract:

Do you want YOUR app easily deployable in Kubernetes or Openshift? Do you want to control how it scale, deploy and update in an automated way? Do you want to have your app in a cloud market? Then Kubernetes Operators are for YOU.

In this session I will show you What's a K8s Operator, How to create it with Java, Quarkus and native compilation, and how to deploy it on OperatorHub, from my own experience creating this production ready Operator for a FOSS application, including the testing GitHub Actions.

Speaker(s):

Jonathan Vila

Java Champion, Organiser at BarcelonaJUG and cofounder of the JBCNConf conference in Barcelona.

Type: short talk, 30 min Language: en Tags: Cloud, Data Science Audience: Advanced

Abstract:

This session will deliver best practices and recommendations for tuning BI Engine for your existing BigQuery workloads for cheaper and faster queries. Learn how we at REEA are orchestrating BI Engine reservations, on a 5TB dataset, considered small for BigQuery but with big cost savings and accelerated queries. We are seeing many presentations for big enterprises, but now we are showcasing how our queries perform better with lower costs. We are going to address the top considerations when to turn on BI Engine, how to use cloud orchestration for making this an automatic process, and combined with BigQuery and Datastudio query complexity that might save precious development time, lower bills, faster queries.

Speaker(s):

Marton Kodok

Speaker at conferences, a Google Developer Expert, champion of the Cloud Innovators program, top user on Stackoverflow, technical software architect at REEA.net, co-founder of IT Mures, life-long learner influencer, voice of local community on digitalization, social responsibility advocate of transparency, frequent publisher.

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

Abstract:

As introduction, we will explain Microsoft's approach from Research to AI embedded products. After a short Product overview, we explain the differences between scenario-based AI, customizable AI Models and Custom Models. The main focus of the talk is about Responsible AI. How it is ensured through design guidelines and tools. Why we use tools like Fairlearn, InterpretML, Error Analysis and Counterfit…and even more important. Why you should use these Open-Source tools to bring your models to the next level.

Speaker(s):

Lars Ruddigkeit

I am Cloud Solution Architect for Data and AI at Microsoft. Meaning I work hands-on on AI projects with large enterprise customers, very often Fortune 500. My focus is 90% AI, 10% Data and always security. FSI is heavily regulated so no fund with AI without security. My Ph.D. is in Chemistry but I use roughly 1% from my studies and Ph.D. times.

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

Abstract:

Ruter uses machine learning to classify and auto-reply to customer complaints, predict arrival times, and forecast passenger numbers for over half a million departures every day. Our first model was deployed in the summer of 2020. Updating it was time-consuming, and monitoring was limited and manual. Today, we have a robust, reproducible, and scalable framework for developing and deploying machine learning models on AWS. The framework maintains a reproducible history of data, models, and predictions. It monitors pipelines for failures and alerts the team of reduced model quality in real-time. Using this framework, data scientists and machine learning engineers can quickly turn a proof of concept into a value-generating application. In this talk, we will present our framework, its associated tech stack, and how we use it to develop, support, and monitor models in production.

Speaker(s):

David Skålid Amundsen

David is an experienced data scientist who is passionate about building robust and trustworthy machine learning solutions. For the past year and a half he has been working for Ruter on establishing just this.

Jawad Saleemi

Jawad Saleemi is an experienced data scientist and is team lead of the machine learning team at Ruter.

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

Abstract:

The CAP theorem is widely known for distributed systems, but it's not the only tradeoff you should be aware of. For datastores there is also the FAB theory and just like with the CAP theorem you can only pick two:
* Fast: Results are real-time or near real-time instead of batch oriented.
* Accurate: Answers are exact and don't have a margin of error.
* Big: You require horizontal scaling and need to distribute your data.

While Fast and Big are relatively easy to understand, Accurate is a bit harder to picture. This talk shows some concrete examples of accuracy tradeoffs Elasticsearch can take for terms aggregations, cardinality aggregations with HyperLogLog++, and the IDF part of full-text search. Or how to trade some speed or the distribution for more accuracy.

Speaker(s):

Philipp Krenn

Philipp lives to demo interesting technology. Having worked as a web, infrastructure, and database engineer for over ten years, Philipp is now a developer advocate and EMEA team lead at Elastic — the company behind the Elastic Stack consisting of Elasticsearch, Kibana, Beats, and Logstash.

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

Abstract:

Production machine learning is already bringing value to companies and organizations across the world, but at what environmental cost? In this talk we will learn how to estimate the carbon emissions for machine learning pipelines. We will also see some practical approaches to reducing the carbon footprint of your machine learning efforts.

Speaker(s):

Magdalena Stenius

Magdalena is a machine learning engineer at Valohai and an active member of the Finnish female and nonbinary programmers community as well as the co-host of Turku.py, her home town's local python meetup. She is passionate about environmental topics and fairness in the context of ML and AI.

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

Abstract:

When machine learning models are deployed to production, their performance starts degrading. Now that ML models are increasingly becoming mission critical for enterprises and startups alike, root cause analysis and gaining observability into your AI systems is similarly mission critical. However, many organizations struggle to prevent model performance degradation and assure the quality of the data being fed to their ML models, largely because they don't have the tools and organizational knowledge to do so.

In this talk, MLOps Architect Danny D. Leybzon will explain the problems associated with ML models deployed in production, and how many of these problems can be addressed with data monitoring and AI observability best practices. Taking it a step further, the speaker will discuss steps that data scientists and machine learning engineers can take to proactively ensure the performance of their models, rather than reacting to the impacts of performance degradation reported by their customers.

Speaker(s):

Danny D. Leybzon

Danny D. Leybzon has worn many hats, all of them related to data. He studied computational statistics at UCLA and has worked in the data and ML space ever since. In his role as MLOps architect, he has worked to evangelize machine learning best practices, talking on subjects such as distributed deep learning, productionizing machine learning models, automated machine learning, and lately has been talking about AI observability and data logging. When Danny's not researching, practicing, or talking about data science, he's usually doing one of his numerous outside hobbies: rock climbing, backcountry backpacking, skiing, etc.

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

Abstract:

Let's merge domain-specific models with generic learning. With Julia it is easy to combine these different worlds, differential equations on the one hand and (deep) neural networks on the other. The combination is known as scientific machine learning and it brings immense benefits to all types of industries.

Neural differential equations are just one example out of this rich ecosystem. The core idea is to predict a time-series by modelling its derivative with a neural network. We will briefly explain the theory behind neural differential equations, why Julia is especially suited for this, and experience its power for predicting time-series in practice.

Speaker(s):

Stephan Sahm

Stephan Sahm is founder of Jolin.io, an IT consultancy for high-performant data/ml pipelines powered by Julia. He organises the biggest Julia meetup in Germany, the Julia-User-Group-Munich.
With 10 years experience in data science and 5 years in IT consultancy and datastartups, he strives to bring tomorrows best practice into todays business.

His expertise is in big-data and real-time processing, optionally empowered by machine learning. From PoC via MVP to production, all stages and requirements he supported. Industries include automotive, telecommunication, ecommerce, and media, both cloud-based and on-premise. He implemented use cases with time series, natural language processing, recommendation systems and anomaly detection using Python, R, Scala, and Julia.

Type: presentation, 45 min Language: en Tags: AI, Ethics Audience: All

Abstract:

To seize the opportunities AI offers society, while limiting its risks and ensuring citizen protection, we need to bring abstract ethical principles into practice. A group of scientists and tech businesses are working on an AI Trust Label to make this possible. It specifies and operationalizes values like transparency, privacy and fairness, to allow for effective and enforceable ethics guidelines.

In this talk you will get an overview over the recently published standard VDE SPEC 90012 (VCIO based description of systems for AI trustworthiness characterization). It is a framework to describe whether a product adheres to specific values and can be trusted. End consumers, companies and government organizations can use the description to define requirements or to compare different products. The standard is the basis for a future AI Trust Label or certification mechanisms. The presentation will also discuss the opportunities and limitations of operationalizing trust or ethical values for AI systems.

Speaker(s):

Michael Puntschuh

Michael is a freelance researcher, speaker and advocate on human rights in our digitized world. He is a social scientist working on the issues of human rights online and tech governance as well as other societal and legal aspects of digital technologies. His work is currently focused on criteria for the design of algorithmic systems, public innovation, digital and AI ethics and algorithmically mediated discrimination. As part of the AI Ethics Impact Group, he has worked on the development of an AI Trust Label.

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

Abstract:

I'd like to take you through a v short history of the semantic web (failure?), the semantic technologies we have as a result of this effort, the rise of graph databases and knowledge graphs as well as the new data mesh data architecture paradigm and what it might mean for data architecture and in terms of possibilities for data science going forward. For example Gaia-X, EUs New project fordevelopment of a federation of data infrastructure and service providers for Europe with the objective of ensuring a European digital sovereignty. Or identity using graphs and zero knowledge proofs.

Speaker(s):

Julia Flament-Wallin

I'm a developer and lead in the intersection between data science and software engineering.

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

Abstract:

Do you want YOUR app easily deployable in Kubernetes or Openshift? Do you want to control how it scale, deploy and update in an automated way? Do you want to have your app in a cloud market? Then Kubernetes Operators are for YOU.

In this session I will show you What's a K8s Operator, How to create it with Java, Quarkus and native compilation, and how to deploy it on OperatorHub, from my own experience creating this production ready Operator for a FOSS application, including the testing GitHub Actions.

Speaker(s):

Jonathan Vila

Java Champion, Organiser at BarcelonaJUG and cofounder of the JBCNConf conference in Barcelona.

Type: presentation, 45 min Language: en Tags: AI, Ethics, ML Audience: All

Abstract:

Technology is embedded in our lives, and Artificial Intelligence will have an ever greater impact on the development of humanity. As creators of technology, are we prepared for the responsibility that this entails, as we try to create machines that emulate human reasoning and behaviour? Let's find out what Isaac Asimov, one of the greatest writers of Science Fiction, had to say about it.

Speaker(s):

Patty O'Callaghan

Patty is a Venezuelan living in Scotland. Her background education is Computer Science, in the field of Artificial Intelligence. She has 20 years of experience working in Tech and works as a Software Engineer at Charles River Laboratories. Patty is very passionate about inspiring girls into joining Computer Science, which is why she is a volunteer at Code Club Scotland and she is also an advisor at WeAndAI.org, where she hopes to bring discussions of artificial intelligence into her work with young people, and ultimately assist in the construction of an educational framework. She is a Lead @ GDG Glasgow and a member of Google's Women Techmakers.

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

Abstract:

Production machine learning is already bringing value to companies and organizations across the world, but at what environmental cost? In this talk we will learn how to estimate the carbon emissions for machine learning pipelines. We will also see some practical approaches to reducing the carbon footprint of your machine learning efforts.

Speaker(s):

Magdalena Stenius

Magdalena is a machine learning engineer at Valohai and an active member of the Finnish female and nonbinary programmers community as well as the co-host of Turku.py, her home town's local python meetup. She is passionate about environmental topics and fairness in the context of ML and AI.

Type: presentation, 45 min Language: en Tags: Data Science Audience: Beginner

Abstract:

Imagine a world where data is machine readable, connected and contain meaning for both humans and machine. By structuring data as directed graphs, and applying a layer of propositional logic, to explore strange new insight, to seek out new connections and new knowledge, to boldly go where no man has gone before.

Speaker(s):

Veronika Heimsbakk

I adore working with messy, unstructured and chaotic data, and turning them into a wonder of semantic knowledge graphs. Apart from the wonders of semantics and logic, I love making and soldering stuff, LaTeX, and teaching kids to code.

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

Abstract:

I will be talking about AI in Healthcare: The Hype, Hope and the Hurdles. I will touch upon the endless possibilities AI represents, how AI can be implemented into clinical practice in the future and the ethical challenges with AI in Medicine.

Speaker(s):

Ishita Barua

Ishita Barua is a MD and a PhD Candidate in the Clinical Effectiveness Research Group at the University of Oslo and Oslo University Hospital. Her research focuses on the use of artificial intelligence in colorectal cancer diagnosis and colonoscopy. She is one of the initiators behind and lecturers in the medical elective subject called MED3065: AI, innovation, big data and clinical decision support.

She was awarded the Fulbright Scholarship in 2020, and was a visiting researcher at BIDMC and Harvard Medical School in Boston, USA, in 2021.

She is currently a deputy member of the Norwegian Biotechnology Advisory Board, appointed by the Norwegian government. In 2021, she was named one of top 30 women in Norway changing the field of Artificial Intelligence, by NORA - The Norwegian Artificial Intelligence Research Consortium, and in 2022 she was named one of Norway's leading Tech-Women by Abelia and ODA. She has previously worked as a Resident Doctor in Gastroenterology and Internal Medicine, and served as the President of Young Internists of Norway.

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

Abstract:

When will my package arrive? At Posten, we use machine learning to estimate delivery times. The current solution is built on a summary of the production flow that starts with the sender, goes on to a first terminal, the shipment then transits to a second terminal, from where it is delivered out to the receiver. However, the data collected during the production flow looks nothing like this, requiring a 'translation-step' to make it concur with the given simplified description. Even though the end solution performs very well, the amount of work that goes into translating the raw data is manual, and has become a very large chore when adapting to each service and their unique properties. Not only is it time-consuming, it also requires a lot of knowledge about business logic.

In this talk we will describe a new approach to modelling delivery time estimates. Inspired by more advanced machine learning algorithms, we have tried to solve this issue by working with the data in its rawer form. The goal is not necessarily to improve the results, but rather to minimize the manual preprocessing step. The new model will pick up local variations in production, and integrate business logic into its predictions. Our aim is to lower significantly the requirements on understanding all the production processes that can occur within the different services, as well as freeing up time that is spent on manual data cleaning.

Speaker(s):

Simon Simonsson

Originally from Stockholm, Sweden. After almost 10 years of experience within software engineering including small applications, web and hardware development; Simon Simonsson graduated with a MSc in Computer Science from KTH in Stockholm, Sweden in 2019. Since then he has focused on more data driven solutions, working as a Data Scientist and part-time research assistant at the Oslo Metropolitan University researching on smart homes for elderly care. Today he works full-time as a Data Scientist in the innovation department at Posten AS working on among other things, making the best possible predictions on delivery times for customers. Interests include machine learning, automation, data visualization and robotics.

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

Abstract:

As introduction, we will explain Microsoft's approach from Research to AI embedded products. After a short Product overview, we explain the differences between scenario-based AI, customizable AI Models and Custom Models. The main focus of the talk is about Responsible AI. How it is ensured through design guidelines and tools. Why we use tools like Fairlearn, InterpretML, Error Analysis and Counterfit…and even more important. Why you should use these Open-Source tools to bring your models to the next level.

Speaker(s):

Lars Ruddigkeit

I am Cloud Solution Architect for Data and AI at Microsoft. Meaning I work hands-on on AI projects with large enterprise customers, very often Fortune 500. My focus is 90% AI, 10% Data and always security. FSI is heavily regulated so no fund with AI without security. My Ph.D. is in Chemistry but I use roughly 1% from my studies and Ph.D. times.

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

Abstract:

Event driven is a new way to process and ingest the data. Serverless is a new paradigm to abstract the technical layer and to focus on what matters: the business processes.
Both combined, you can create a powerful and scalable data pipeline on Google Cloud. You can react to events, trigger the next step with ease and just in time.
The reality isn't so nice, and it quickly becomes a nightmare to understand and to update the pipelines. Workflows is a new product on Google Cloud that orchestrates the processes.
Come to discover what was the serverless data pipelines before and how it's smoother with Workflow!

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

Type: presentation, 45 min Language: en Tags: AI, Ethics Audience: All

Abstract:

To seize the opportunities AI offers society, while limiting its risks and ensuring citizen protection, we need to bring abstract ethical principles into practice. A group of scientists and tech businesses are working on an AI Trust Label to make this possible. It specifies and operationalizes values like transparency, privacy and fairness, to allow for effective and enforceable ethics guidelines.

In this talk you will get an overview over the recently published standard VDE SPEC 90012 (VCIO based description of systems for AI trustworthiness characterization). It is a framework to describe whether a product adheres to specific values and can be trusted. End consumers, companies and government organizations can use the description to define requirements or to compare different products. The standard is the basis for a future AI Trust Label or certification mechanisms. The presentation will also discuss the opportunities and limitations of operationalizing trust or ethical values for AI systems.

Speaker(s):

Michael Puntschuh

Michael is a freelance researcher, speaker and advocate on human rights in our digitized world. He is a social scientist working on the issues of human rights online and tech governance as well as other societal and legal aspects of digital technologies. His work is currently focused on criteria for the design of algorithmic systems, public innovation, digital and AI ethics and algorithmically mediated discrimination. As part of the AI Ethics Impact Group, he has worked on the development of an AI Trust Label.

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

Abstract:

What does AI have in common with Sustainability? This talk describes how Data and AI technologies are currently being leveraged at DNB to create the ESG Data Hub, with the mission to fuel the transition to a green economy. She will provide concrete examples on how the analysis of ESG data can have massive positive implications on global economy, and will provide concrete examples on why Responsible AI and Sustainability are converging in many unforeseen ways.

Speaker(s):

Aiko Yamashita

Aiko has a background in Informatics with a doctoral degree in experimental software engineering. She has +10 years of experience in data analytics within the logistics, retail and finance industries. She is currently leading the ESG Data Task Force, which has as mission to ensure top-of-class ESG analytics and timely access to ESG data for the Corporate. She believes in responsible use of Data, AI, and Technology, and is a passionate advocate for data literacy in DNB. In addition to being a Data Scientist at DNB, Aiko is an Adjunct Associate Professor at OsloMet – Oslo Metropolitan University, where she has taught courses in Data Science and supervised students.

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

Abstract:

Let's merge domain-specific models with generic learning. With Julia it is easy to combine these different worlds, differential equations on the one hand and (deep) neural networks on the other. The combination is known as scientific machine learning and it brings immense benefits to all types of industries.

Neural differential equations are just one example out of this rich ecosystem. The core idea is to predict a time-series by modelling its derivative with a neural network. We will briefly explain the theory behind neural differential equations, why Julia is especially suited for this, and experience its power for predicting time-series in practice.

Speaker(s):

Stephan Sahm

Stephan Sahm is founder of Jolin.io, an IT consultancy for high-performant data/ml pipelines powered by Julia. He organises the biggest Julia meetup in Germany, the Julia-User-Group-Munich.
With 10 years experience in data science and 5 years in IT consultancy and datastartups, he strives to bring tomorrows best practice into todays business.

His expertise is in big-data and real-time processing, optionally empowered by machine learning. From PoC via MVP to production, all stages and requirements he supported. Industries include automotive, telecommunication, ecommerce, and media, both cloud-based and on-premise. He implemented use cases with time series, natural language processing, recommendation systems and anomaly detection using Python, R, Scala, and Julia.

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

Abstract:

Recent improvements in machine learning (ML) have enabled the application of artificial intelligence (AI) in many different areas, resulting in significant achievements in computational vision, speech recognition or protein analysis.

Being amazing, these techniques suffer significant limitations, presenting what some call 'diminishing returns'. In particular, machine learning often only recognizes a pattern it has seen before, 'catastrophic forgetting', overwriting past knowledge with new knowledge, lacking explanation of the train of its thoughts or basic common sense.

For all these reasons, in comparative terms, all of these results correspond to basic human perception hugely empowered with computational resources. Evolution of the artificial intelligence requires additional techniques such as neuro symbolic computations or causal inference.

In this short presentation, after a broader definition of artificial intelligence, causal inference is introduced, explaining its main features and methods, proposing a model for a causal inference engine and describing the main elements of a prototype implemented in Kotlin.

Speaker(s):

Andrés-Leonardo Martínez-Ortiz

Andrés-Leonardo Martínez-Ortiz, better known as almo, holds a PhD on Software, Systems, and Computing and is a member of the Google Machine Learning Site Reliability Engineering team where he leads several programs aiming for reliability & convergence. He is also a member of IEEE, ACM, the Linux Foundation and Computer Society.

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

Abstract:

I'd like to take you through a v short history of the semantic web (failure?), the semantic technologies we have as a result of this effort, the rise of graph databases and knowledge graphs as well as the new data mesh data architecture paradigm and what it might mean for data architecture and in terms of possibilities for data science going forward. For example Gaia-X, EUs New project fordevelopment of a federation of data infrastructure and service providers for Europe with the objective of ensuring a European digital sovereignty. Or identity using graphs and zero knowledge proofs.

Speaker(s):

Julia Flament-Wallin

I'm a developer and lead in the intersection between data science and software engineering.

Type: short talk, 30 min Language: en Tags: Cloud, Data Science Audience: Advanced

Abstract:

This session will deliver best practices and recommendations for tuning BI Engine for your existing BigQuery workloads for cheaper and faster queries. Learn how we at REEA are orchestrating BI Engine reservations, on a 5TB dataset, considered small for BigQuery but with big cost savings and accelerated queries. We are seeing many presentations for big enterprises, but now we are showcasing how our queries perform better with lower costs. We are going to address the top considerations when to turn on BI Engine, how to use cloud orchestration for making this an automatic process, and combined with BigQuery and Datastudio query complexity that might save precious development time, lower bills, faster queries.

Speaker(s):

Marton Kodok

Speaker at conferences, a Google Developer Expert, champion of the Cloud Innovators program, top user on Stackoverflow, technical software architect at REEA.net, co-founder of IT Mures, life-long learner influencer, voice of local community on digitalization, social responsibility advocate of transparency, frequent publisher.

Type: presentation, 45 min Language: en Tags: Cloud, ML Audience: Beginner

Abstract:

With the exploding interests in Machine Learning technology in the recent years, they have also introduced many aspects of computing that are required to support such a grand vision in what ML is capable of delivering. On the infrastructural layer, we need to handle the high frequency of data ingestion with low latency, and one of the best mechanisms that we can think of leveraging on is streaming. So what is streaming, and what are the different choices we have as a platform? We will learn about a few options, and zoom in to learn more about what a true next-generation, cloud-native streaming platform such as Apache Pulsar is capable of, beyond what the more common messaging platforms that we have these days.

Speaker(s):

Mary Grygleski

Mary is a Java Champion and a passionate Streaming Developer Advocate at DataStax, a leading data management company that champions Open Source software and specializes in Big Data, DB-as-a-service, Streaming, and Cloud-Native systems. She spent 3.5 years previously as a very effective advocate at IBM, focusing on Java, Jakarta EE, OpenJ9, Open Source, Cloud, and Distributed Systems. She transitioned from Unix/C to Java around 2000 and has never looked back since then. She considers herself a polyglot and loves to continue learning new and better ways to solve real-life problems. She is an active tech community builder outside of her day job, and currently the President of the Chicago Java Users Group (CJUG), as well as a co-organizer for several IBM-sponsored meetup groups in the Greater Chicago area.

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

Abstract:

CI/CD tools are multiples and all have their strenghts and weaknesses. Cloud Build is the fully managed CI/CD solution of Google Cloud that comes with a lot with great feature: scalable, serverless, free tiers,...

To leverage its full power, I propose a deep dive in its architecture. The design choice has consequences and requires specific patterns that I present to avoid mistake and loosing time and to speed up the development of pipelines

Speaker(s):

Guillaume Blaquiere

Guillaume is a Google Developer Expert on Cloud Platform since 2019 and works at Carrefour as Group Data Architect.

Java developer for more than 15 years, and despite positions of responsibilities, he has always kept his wish to create, develop, discover and test new solutions, especially in the Cloud, the machine learning or Python and Go language.

Innovation addict and Google Cloud 3x certified, writer and speaker in his free time, he's fascinated by the serverless solution and all the 'usual' problems that it solves.

More generally, he likes helping people stuck on Google Cloud; you can find him on Stack Overflow (guillaume-blaquiere), Medium (@guillaume-blaquiere) and Twitter (@gblaquiere).

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

Abstract:

Ruter uses machine learning to classify and auto-reply to customer complaints, predict arrival times, and forecast passenger numbers for over half a million departures every day. Our first model was deployed in the summer of 2020. Updating it was time-consuming, and monitoring was limited and manual. Today, we have a robust, reproducible, and scalable framework for developing and deploying machine learning models on AWS. The framework maintains a reproducible history of data, models, and predictions. It monitors pipelines for failures and alerts the team of reduced model quality in real-time. Using this framework, data scientists and machine learning engineers can quickly turn a proof of concept into a value-generating application. In this talk, we will present our framework, its associated tech stack, and how we use it to develop, support, and monitor models in production.

Speaker(s):

David Skålid Amundsen

David is an experienced data scientist who is passionate about building robust and trustworthy machine learning solutions. For the past year and a half he has been working for Ruter on establishing just this.

Jawad Saleemi

Jawad Saleemi is an experienced data scientist and is team lead of the machine learning team at Ruter.

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

Abstract:

When machine learning models are deployed to production, their performance starts degrading. Now that ML models are increasingly becoming mission critical for enterprises and startups alike, root cause analysis and gaining observability into your AI systems is similarly mission critical. However, many organizations struggle to prevent model performance degradation and assure the quality of the data being fed to their ML models, largely because they don't have the tools and organizational knowledge to do so.

In this talk, MLOps Architect Danny D. Leybzon will explain the problems associated with ML models deployed in production, and how many of these problems can be addressed with data monitoring and AI observability best practices. Taking it a step further, the speaker will discuss steps that data scientists and machine learning engineers can take to proactively ensure the performance of their models, rather than reacting to the impacts of performance degradation reported by their customers.

Speaker(s):

Danny D. Leybzon

Danny D. Leybzon has worn many hats, all of them related to data. He studied computational statistics at UCLA and has worked in the data and ML space ever since. In his role as MLOps architect, he has worked to evangelize machine learning best practices, talking on subjects such as distributed deep learning, productionizing machine learning models, automated machine learning, and lately has been talking about AI observability and data logging. When Danny's not researching, practicing, or talking about data science, he's usually doing one of his numerous outside hobbies: rock climbing, backcountry backpacking, skiing, etc.

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

Abstract:

The CAP theorem is widely known for distributed systems, but it's not the only tradeoff you should be aware of. For datastores there is also the FAB theory and just like with the CAP theorem you can only pick two:
* Fast: Results are real-time or near real-time instead of batch oriented.
* Accurate: Answers are exact and don't have a margin of error.
* Big: You require horizontal scaling and need to distribute your data.

While Fast and Big are relatively easy to understand, Accurate is a bit harder to picture. This talk shows some concrete examples of accuracy tradeoffs Elasticsearch can take for terms aggregations, cardinality aggregations with HyperLogLog++, and the IDF part of full-text search. Or how to trade some speed or the distribution for more accuracy.

Speaker(s):

Philipp Krenn

Philipp lives to demo interesting technology. Having worked as a web, infrastructure, and database engineer for over ten years, Philipp is now a developer advocate and EMEA team lead at Elastic — the company behind the Elastic Stack consisting of Elasticsearch, Kibana, Beats, and Logstash.