New Programming Constructs for Probabilistic AI
Explore novel programming constructs for probabilistic AI that automate and hide complex mathematical and numerical details in this conference talk from Strange Loop. Learn about a new conceptual framework that replaces arcane mathematical objects with more accessible code, enabling programmers without advanced mathematical training to engage in state-of-the-art AI programming. Discover how to write stochastic simulators that produce imaginary data sets and create metaprograms to analyze simulator code alongside real-world data, inverting the simulator to infer explanatory events. Witness practical applications using Gen, a general-purpose probabilistic programming system, including inferring 3D structure from images and finding hidden compositional structure in time series data for improved forecasts. Gain insights from Marco Cusumano-Towner, creator of Gen and PhD student at MIT, as he demonstrates how these constructs make powerful AI approaches more feasible and accessible.