Bit Blasting Probabilistic Programs
Explore a groundbreaking 20-minute video presentation from PLDI 2024 that introduces a novel approach to handling hybrid probabilistic programs. Delve into the concept of bit blasting, a discretization technique that represents continuous distributions using binary representations, enabling efficient inference on hybrid programs. Learn how researchers from UCLA and Northeastern University developed HyBit, a probabilistic programming system that combines bit blasting with discrete probabilistic inference. Discover how this method outperforms existing sampling-based and symbolic inference approaches, offering improved accuracy and efficiency for many common continuous distributions. Gain insights into the theoretical foundations and practical applications of this innovative technique in the field of probabilistic programming languages.