Over the last few decades, the two central fields of AI, namely, machine learning and symbolic reasoning (e.g., SAT/SMT/CP solvers), have developed relatively independently. Researchers have considered a variety of ways to bringing these fields together, under the broad umbrella of neuro-symbolic computing. In our research, we are working on a variety of neuro-symbolic methods that enable greater control over the learning process of machine learning models by combining machine learning with symbolic or logical reasoning. We are applying these methods to Math, Physics, and other practical applications. Concurrently, we are working towards solver-based testing, analysis, and verification (TAV) methods aimed at making neural networks reliable and trust worthy. Some of the projects we are working on include:
Click here to see how machine learning can help logical reasoning.