Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

AGPhil: Arbeitsgruppe Philosophie der Physik

AGPhil 6: Semi-Classical Gravity 2

AGPhil 6.3: Vortrag

Dienstag, 19. März 2024, 12:30–13:00, PTB SR AvHB

Black boxes in black hole imaging — •Juliusz Doboszewski^1,2 and Elder Jamee^3,2 — ¹Lichtenberg Group for History and Philosophy of Physics, University of Bonn — ²Black Hole Initiative, Harvard University — ³Tufts University

Machine learning methods are increasingly adapted to various problems in black hole imaging. Examples include the 2023 M87* image based on PRIMO (a dictionary-learning algorithm), alpha-DPI (a deep learning framework for, among others, posterior estimation of black hole parameters), and machine learning-based denoisers (suggested as a plug-in component within more conventional imaging algorithms). As a result, issues related to the notion of epistemic opacity also become relevant to black hole imaging. In this talk, I will first argue that at least one problematic form of opacity is already present in black hole imaging: GRMHD simulations of some (e.g. SgrA*; but not all, e.g. M87*) sources are opaque to some extent. This form of opacity signals limitations of the current understanding of the source*s models. However, there are also forms of opacity (including opacity resulting from the use of a deep neural network) which can remain entirely unproblematic when seen as a part of a broader inferential framework. I will propose six conditions under which that can plausibly the case, and discuss how opaque methods can be useful in the context of the next generation Event Horizon Telescope.

Keywords: black hole; VLBI; opacity; deep learning; Event Horizon Telescope