Feb 16, 2023 at 6:00PM Slides
Abstract
I will talk about neurosymbolic programming, an emerging subarea of AI that bridges the fields of deep learning and program synthesis. Like in classical machine learning, the goal here is to learn functions from data. However, these functions are represented as programs that use both neural modules and symbolic primitives and are induced using a combination of symbolic search and gradient-based optimization. Because neurosymbolic approaches have explicit access to symbolic truths about the world, they can often be learned in a more robust and data-efficient way than deep neural networks. Neurosymbolic representations are also, in many cases, easier to interpret and analyze than neural nets. There are other benefits too: programs can sometimes naturally represent long-horizon, procedural tasks that are hard to perform using deep networks, and compositional programming abstractions can be a natural way of reusing learned modules across learning tasks.
In the talk, I will illustrate the main research themes in this area that my lab has been pursuing, using concrete applications in behavioral neuroscience. I will conclude with a discussion of some of the open technical challenges in the field.
Bio
Swarat Chaudhuri is an associate professor of computer science at the University of Texas at Austin, USA. He studies problems at the interface of programming languages, logic and formal methods, and machine learning. Through a combination of programming language abstractions, statistical learning, search, and automated mathematical reasoning, he and his research group hope to build a new class of intelligent systems that are reliable, secure, and transparent by construction and can perform complex tasks that are beyond the scope of contemporary AI.