- Talk 1: Interpretable Machine Intelligence using Coalition Game Theory
Speaker:
Ayan Paul, Fellow @ Desy & Senior Scientist @ Humboldt University zu Berlin
Abstract:
I will highlight the importance of interpretability of machine intelligence in physics and economics. I will showcase the use of Shapley values derived from Coalition Game Theory to study global model properties and discuss two specific examples. The first will look into how complex correlations can be studied with gradient boosted machines and Shapley values. The second will address the delicate question of causality and where we stand with exploring causal connections using machine intelligence. The talk will focus on results from recent research work.
- Talk 2: To predict or not to predict? Insights regarding COVID-19 measures via machine learning interpretability tools
Speaker:
Dana Jomar & Ingo Nader, Data Scientists at IT-PS (IT Power Services GmbH)
Abstract:
Political measures like school closings (also called non-pharmaceutical interventions in the literature) have been used by almost all country in the world to fight the outbreak of the COVID-19 pandemic. The effectiveness of these measures was discussed heatedly already in the beginning of the pandemic, so we wanted to form our own opinion based on open data. We will showcase the use of machine learning and model interpretability tools like accumulated local effect (ALE) plots to gain insights about the effectiveness and the time it takes for different measures to show effects on the spread of COVID-19.
- question & answer session
