Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AGPhil: Arbeitsgruppe Philosophie der Physik
AGPhil 5: Symposium: Epistemology of Big Data in Physics III
AGPhil 5.2: Vortrag
Donnerstag, 16. März 2017, 18:00–18:30, GW2 B2900
The automated discovery of physical laws — •Nico Formanek1 and Ryan Reece2 — 1Höchstleistungsrechenzentrum Stuttgart (HLRS) — 2University of California, Santa Cruz (UCSC)
In the recent past there have been several attempts to automatically infer known and new laws of physics from large empirical data sets. Machine learning methods are employed to some success in solid-state physics and materials science to predict electronic properties (Schütt et al.; Physical Review B 89, 205118 2014) but there is also the far reaching claim of Schmidt and Lipson concerning the inference of free-form natural laws from experimental data (Schmidt and Lipson; Science 324, 3 2009). Both points give rise to important philosophical questions: How do physical laws derived in such a way differ from humanly generated ones? How do they methodically differ from classical statistical correlations? What is the role of the physicist in studying those laws -- is she becoming a mere interpreter of machine generated knowledge? The answer to these questions is by no means clear and depends on our preconceived notion of physical law. In this talk I point out how computer inferred physical laws pose a challenge to some traditional views of natural laws and how this affects the answer to the questions above.
N.B: This talk is complementary to the talk titled 'Machine learning and realism' which looks at automation in science from a physicist's perspective.