Several grand challenges like climate change and pandemics emerge from the complex interactions of millions of individual decisions. While LLMs and AI agents excel at individual behavior, they can't model these intricate societal dynamics effectively. Enter Large Population Models (lpm.media.mit.edu): a new AI paradigm, simulating millions of interacting agents simultaneously and capturing collective behaviors at societal scale.
AgentTorch, an open-source framework created by MIT PhD candidate Ayush Chopra under the supervision of Prof. Ramesh Raskar, makes building and running these massive simulations accessible. It's optimized for GPUs, allowing efficient simulation of entire cities or countries. AgentTorch has been used worldwide in various scenarios, most notably in optimizing vaccine distribution strategies.
In this talk, Vedant Kulkarni - a co-maintainer of the project and a student at IIITH - will dive into the architecture behind the project, the underlying technology of LPMs, and explore how you can get started contributing to the open source project.