From Entropy to Artistry: on Thermodynamics and Generative AI

Published on ● Video Link: https://www.youtube.com/watch?v=RYqiAZgLmBg



Duration: 54:53
176 views
5


Stephan Mandt (UC Irvine)
https://simons.berkeley.edu/talks/stephan-mandt-uc-irvine-2024-06-10
AI≡Science: Strengthening the Bond Between the Sciences and Artificial Intelligence

Denoising diffusion models have led to a series of breakthroughs in image and video generation. In this talk, I will explore some of the connections between diffusion models and physics. Rooted in non-equilibrium thermodynamics, diffusion models enable a variety of extensions by lifting them into augmented spaces, encompassing position, momentum, and potentially additional auxiliary variables. This viewpoint gives rise to a “complete recipe” for constructing invertible diffusion processes, as well as new samplers that significantly reduce the number of sampling steps at test time. Additionally, thermodynamic processes offer a natural playground for generative AI. I will demonstrate how video diffusion models can effectively downscale precipitation patterns to finer scales, capturing extreme event statistics and local geographical patterns.




Other Videos By Simons Institute for the Theory of Computing


2024-06-21Some thoughts on ML-based protein engineering
2024-06-21Machine learning models of differential gene expression
2024-06-21Open challenges in AI for molecular design: representation, experimental alignment, and...
2024-06-21Illuminating protein space with generative models
2024-06-21Symmetries in Machine Learning for Materials Science
2024-06-21Large ML potentials for chemistry: generalization, inductive biases, and cancellation of errors
2024-06-21Deep learning and numerical methods intersections for improving molecular and fluid dynamics
2024-06-21ML gradients in Molecular Simulations
2024-06-21Generalizable sampling of conformational ensembles with latent space dynamics
2024-06-21The State of Protein Structure Prediction and Friends
2024-06-21From Entropy to Artistry: on Thermodynamics and Generative AI
2024-06-21Explorations in Exploration: Deep Learning meets Value of Information for Sequential...
2024-06-21Error Embraced: Making Trustworthy Scientific Decisions with Imperfect Predictions
2024-06-21Debugging genomic profiling experiments and predictive models with interpretation tools
2024-06-21Harnessing the properties of equivariant neural networks to understand and design materials
2024-06-20Audio-visual self-supervised baby learning
2024-06-18The Platonic Representation Hypothesis
2024-06-18Robot learning, with inspiration from child development
2024-06-18Lightning Talk
2024-06-18The Role of Prior Data in Rapid Learning of Motor Skills
2024-06-18Pathway to Robot Intelligence



Tags:
Simons Institute
theoretical computer science
UC Berkeley
Computer Science
Theory of Computation
Theory of Computing
AI≡Science: Strengthening the Bond Between the Sciences and Artificial Intelligence
Stephan Mandt