ScAI has a strong presence of researchers working in core Machine Learning. The research focuses on several areas, including
Neuro-symbolic reasoning, Deep Generative Models, Graph Neural Networks and Fairness in Machine Learning.
Neuro-Symbolic
Reasoning
This is an emerging area of Machine Learning / Deep Learning whose goal is to combine the power of low-level perceptual reasoning with logical/symbolic reasoning at a higher level. We are actively furthering these research directions for transfer learning across domain sizes in structured domains, neuro-symbolic image manipulation, as well as neuro-symbolic robotic execution given natural language commands.
Deep Generative
Models
This is another important area of ML research at ScAI. The goal in these models is to be able to effectively generate data from an underlying distribution implicitly specified in terms of representative samples. We are also working on the problem of domain transfer in Generative Models: given a generative model learned over a domain with large amounts of data, how do we transfer the learned model to generate samples from another (similar) domain with limited training data.
Graph-Based Machine
Learning
Graph-based approaches with big-data and machine learning are one of the strongest driving forces of the current research frontiers. This decade has witnessed an unprecedented research interest in developing and using graph-based methods for a variety of applications. At ScAI we have a strong presence in this area.
Fairness in Machine
Learning
This is an emerging area in Machine Learning, with focus on having models which are unbiased with respect to specific attribute classes, e.g., gender or race. Our recent work focuses on performing a principled analysis of impact of class skew, class skew given the attribute, and attribute skew on the amount of fairness obtained in the learned model. We are also working on the notion of fairness specifically in the context of ML applied to vision problems.
Faculty Members Currently Involved
