The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Mapping Dark Matter with Bayesian Neural Networks w/ Yashar Hezaveh - TWiML Talk #250
Apr 11, 2019
Yashar Hezaveh, an Assistant Professor at the University of Montreal and a research fellow at the Center for Computational Astrophysics, dives into the fascinating intersection of machine learning and astrophysics. He discusses the challenges of gravitational lensing, where light bends due to gravity, and how ML enhances image reconstruction of distant galaxies. Yashar explores simulation-based approaches, the critical role of 'priors' in analysis, and the future potential of integrating deep learning techniques to transform our understanding of the universe.
AI Snips
Chapters
Transcript
Episode notes
Strong Gravitational Lensing Explained
- Strong gravitational lensing distorts distant object images due to intervening object's gravity.
- This creates rings and arcs around the foreground object, like a wine glass distorting a candle flame.
Machine Learning in Lensing Analysis
- Machine learning helps analyze distorted images from strong lensing.
- It predicts the distortion and reconstructs the true background image, like determining the wine glass' shape and the undistorted candle flame.
Traditional Lensing Modeling
- Traditional lens modeling uses maximum likelihood, simulating many "candles" and "lenses".
- Researchers seek simulations matching observations to infer real parameters, but this is computationally expensive.