Topic

• Can large language models reason about medical questions?

Summary

• Large language models like GPT-3 or PaLM have acquired human-like language understanding capabilities. While previous methods required expensive task-specific fine-tuning, large language models can solve new tasks only based on a text instructions. During this seminar, I will present how to apply GPT-3 to answer complex medical questions (e.g., USMLE) requiring both strong reasoning skills and expert knowledge. I will cover Chain-of-Thought prompting: a technique that allows triggering reasoning-like behaviour in language models, and will discuss augmenting GPT-3 with external systems like search engines.

Speaker

• Valentin Liévin, PhD, Machine Learning Researcher, raffle.ai

Time

• February 20, 14:30-15:30 (CET)

Location

• Steno Diabetes Center Aarhus (Verdensrummet, Palle Juul-Jensens Boulevard 11, Entrance A) or via Zoom

Topic

• Generation and use of realistic and privacy-preserving synthetic data

Summary

• A large amount of personal data is highly valuable for the research and scientific community. However, these data are often not accessible or require a lengthy request process because of privacy concerns and legal restrictions. Synthetic data is one of the most promising solutions to this issue.

  Chang will present her work on generating realistic and privacy-preserving synthetic tabular data using conditional GANs (Generative Adversarial Network) models combining differential privacy techniques. She will talk about how synthetic data can help researchers to build up analysis models before obtaining the access to the real data and for education and training purpose to enhance students analysis skills without taking risks of using real personal data. Finally, she will open the discussion on the legal-ethical aspect of generating and use of synthetic data in different domain and its future research direction.

Speaker

• Chang Sun, PhD candidate, Institute of Data Science, Maastricht University, NL

Time

• November 28, 14:30-15:30 (CET)

Location

• Steno Diabetes Center Aarhus (Skillerummet, Palle Juul-Jensens Boulevard 11, Entrance A) or via Zoom

Topic

• Algorithmic Fairness & Bias in Machine Learning

Summary

• Algorithms completely surround us and are becoming an organic part of our lives. In entering this era, we rely heavily on machine learning (ML) algorithms. In many applications, ML algorithms were designed with the intent of removing preexisting human biases and generate fair decisions, avoiding biased human judgements. Unfortunately though, societal inequities and biases from other sources taint the data that algorithms use, further propagating biases into decisions for the future. During this seminar I will introduce the fundamentals of algorithmic fairness, the most common biases in machine learning, and give a couple of pointers on how to avoid bias in designing ML algorithms.

Speaker

• Tibor V. Varga, Assistant Professor, Section of Epidemiology, University of Copenhagen, Denmark

Time

• October 28, 14:00-15:00 (CET)

Location

• Steno Diabetes Center Aarhus (Skillerummet, Palle Juul-Jensens Boulevard 11, Entrance A) or via Zoom

Topic

• Causes of Outcome Learning: a causal inference inspired machine learning approach (preliminary title)

Speaker

• Andreas Rieckmann, Assistant Professor, Section of Epidemiology, University of Copenhagen, Denmark

Time

• June 20, 14:00-15:00 (CET)

Location

• Steno Diabetes Center Aarhus (Hedeager 3) or via Zoom

Topic

• The regulation of innovation: safe or stifling?

Summary

• By now, we are all familiar with the promised potential of the application of machine learning techniques to healthcare. Currently, healthcare ML models are essentially regulated under provisions that were developed for an earlier age of slowly-updated, physical medical devices. This can slow down the development and deployment of standard ML models, let alone models that will learn from data in real time or near real time. In this talk, I will examine the current and expected regulatory landscape for ML in health, comparing approaches from the EU and the US.

Speaker

• Matthew Fenech, Co-founder & Chief Medical Officer, Una Health, Germany

Time

• May 30, 14:00-15:00 (CET)

Location

• Steno Diabetes Center Aarhus (Hedeager 3) or via Zoom

Topic

• Synthetic data for health research

Speakers

• Lora Frayling, Data scientist, Health Data Insight CIC, UK
• Vajira Thambawita, Research scientist, SimulaMet, Norway
• Henning Langberg, Head of innovation, Rigshospitalet, Denmark

Time

• March 14, 14:00-16:00 (CET)

Location

• Steno Diabetes Center Aarhus (Hedeager 3) or via Zoom

Slides

• Please contact Adam Hulman.

Topic

• The framing of machine learning risk prediction models

Speaker

• Simon Meyer Lauritsen, Head of Life Science Lab, Enversion

Time

• February 21, 14:00-15:00 (CET)

Location

• Steno Diabetes Center Aarhus (Hedeager 3) or via Zoom

Topic

• Clinical applications of transfer learning

Speaker

• Adam Hulman, Senior data scientist, Steno Diabetes Center Aarhus

Time

• January 10, 15:00-16:00 (CET)

Location

• Zoom