Talk: Scalable Structure Learning and Inference via Probabilistic Programming

LIVE STREAM: https://uwmadison.zoom.us/j/98233213828?pwd=dUJGWUJlUTQ1Smk2aUFBajJ6QjBXZz09

Abstract: Probabilistic programming supports probabilistic modeling, learning, and inference by representing sophisticated probabilistic models as computer programs in new programming languages. This talk presents efficient probabilistic programming-based techniques that address two fundamental challenges in scaling and automating structure learning and inference over complex data.

First, we describe scalable structure learning methods that make it possible to automatically synthesize probabilistic programs from scratch by performing Bayesian inference over hierarchies of flexibly structured symbolic program representations, for discovering models of time series data, tabular data, and relational data. Second, we present fast compilers and symbolic analyses that quickly compute exact answers to a broad range of inference queries about these learned programs, which lets us extract interpretable patterns and make accurate predictions.

I will demonstrate how these techniques deliver substantial improvements in runtime, accuracy, robustness, and programmability by drawing on examples from several real-world applications, which include adapting to extreme novelty in economic time series, online forecasting of flu rates given sparse multivariate observations, discovering stochastic motion models of zebrafish hunting, and verifying the fairness of machine learning classifiers.

Bio: Feras Saad is a PhD candidate in Computer Science at MIT, working at the intersection of programming languages, probabilistic AI, and computational statistics. His research focuses on combining ideas from programming and probability to help people and machines make sense of complex phenomena in the world. His work is associated with a collection of popular open-source probabilistic programming systems, which are used by collaborators in industry, academia, and non-profits for practical applications of structure discovery and probabilistic inference. Feras' MEng thesis on probabilistic programming and data science has been recognized with the 1st Place Computer Science Thesis Award at MIT.