Machine Learning: Prediction Without Explanation?
-
type of event:
Workshop
- place:ITAS, Karlstr. 11, 76133 Karlsruhe
-
date:
17.02.20 - 18.02.20
Description
"Machine Learning: Prediction Without Explanation?" is a 2-day workshop taking place from 17 to 18 February 2020 at the Karlsruhe Institute for Technology (KIT), Germany. It aims to bring together philosophers of science and scholars from various fields using Machine Learning techniques, to reflect on the changing face of science in the light of Machine Learning's constantly growing use. This workshop is organized by the project “The Impact of Computer Simulations and Machine Learning on the Epistemic Status of LHC Data” within the interdisciplinary, DFG/FWF-funded research unit “Epistemology of the LHC”.
Over the last decades, Machine Learning techniques have gained prominence in various areas of science. However, Machine Learning largely aim at predictions and does not seem to provide explanations for these, at least not in the same sense as predictions from theories or models do. Depending on the area of application, explanations may be desired or even necessary though. In this workshop, we want to address the complex of questions regarding scientific explanation that arise from this observation. These include, but are not restricted to:
- Will future science favor prediction above explanation?
- What methods are available to use Machine Learning results for explanations?
- What is the nature of these explanations?
- Does machine learning introduce a shift from the classical scientific explanation towards a statistical interpretation of explanation?
Schedule
17.02.2020 | |
---|---|
12:00 – 12:30 | Welcome |
12:30 – 13:45 | Stefan Hinz – Automatic understanding of large scale imagery - from semantic networks to deep learning (and back?) |
13:45 – 14:30 | Tom Sterkenburg - On explaining the success of machine learning methods |
14:30 – 14:45 | Coffee break |
14:45 – 15:30 | Timo Freiesleben - Counterfactual Explanations & Adversarial Examples |
15:30 – 16:45 | Annette Zimmermann - Opacity, Explainability, and Justification in Machine Learning |
16:45 – 17:30 | Florian Boge & Paul Grünke – Machine Learning Opacity and Explanations in High Energy Physics |
18.02.2020 | |
---|---|
09:00 – 10:15 | Erwin Zehe, Uwe Ehret & Ralf Lorenz – Machine learning in environmental sciences - beyond causality or just brute force? |
10:15 – 11:00 | Thomas Grote - Evidence, Uncertainty and the Integration of Machine Learning into Medical Practice |
11:00 – 12:15 | Johannes Lenhard - The History of Mathematization and a New Culture of Prediction |
12:15 – 13:15 | Lunch break |
13:15 – 14:00 | Sergey Titov - Statistical relevance explanation models and modern methods of interpretable machine learning |
14:00 – 14:45 | Maël Pégny – The Relations Between Scientific and Pedagogical Explainability: Lessons from Algorithmic Aids to Decision-Making |
14:45 – 15:15 | Coffee break |
15:15 – 16:30 | Andreas Kaminski – The kind of reasons and the type of explanation in technoscience |
16:30 – 17:00 | Final discussion |
Organization & Contact
This workshop is organized by the project “The Impact of Computer Simulations and Machine Learning on the Epistemic Status of LHC Data” within the interdisciplinary, DFG/FWF-funded research unit “Epistemology of the LHC”. For further information, please contact the organizers:
- Paul Grünke (KIT) - paul.gruenke∂kit.edu
- Rafaela Hillerbrand (KIT) - rafaela.hillerbrand∂kit.edu