1. Health and Medicine
Computational methods have the potential to significantly reduce the cost and time required to find drugs with a lower probability of attrition. While many Computer-Aided-Design methods have been used for structure-based-design, such methods have typically failed in predicting physiochemical properties resulting in high attrition rates. Our proposal is based on utilizing advances in Machine Learning, or more specifically in Generative Models, to develop methods that will allow rapid discovery of molecule with desired binding, ADME and toxicological profiles.
The laboratory is also focused on the use of machine learning for processing multi-modal images to serve as an accurate though low cost, diagnostic aid to pathologists and radiologists.
We are committed to using advances in Computer Science to improve the quality of life of persons with disability (PwDs).
One of our efforts is based on haptic rendering of an image to help those with limited or no visibility understand their surroundings. Specifically, we use a cane mounted camera or head mounted camera and two haptic wristbands to render visual scenes. The camera continually acquires images; advanced resource sensitive machine learning methods identify objects, and haptic feedback is used to “haptically render” an image.
3. Information Curation
Our effort is directed towards using natural language and advanced machine learning approaches to automatically construct a knowledge graph from publicly available information.
We also use a design thinking motivated approach to create an interface that makes it easy for individuals with varying levels of computer literacy to be able to discover, explore and rapidly understand a larger picture of a given topic.
4. Core Machine Learning
This category of our work focuses on the core aspects of machine learning. It includes work that involves advanced methods of transfer learning, privacy-preserving machine learning, federated learning, creativity, Fairness, Accountability and Transparency (FAT), and Explainable AI.
We are particularly interested in black-box methods that allow discovery of bias in empirical models, and in situ and ex-situ methods that allow human understanding of machine learning models.
5. Experiential Pedagogy
A goal of the laboratory is to make its discoveries and methods widely accessible to students globally. We adopt a novel pedagogy and believe that “doing” leads to long term retention and deeper understanding of concepts. Thus, instead of a linear approach to learning in which topics are arranged by subject, we adopt a non-linear approach in which students learn all technologies and methods that are required to achieve a specific goal. We hope to be able to periodically release self-contained modules that describe the overall journey of each completed project in our laboratory. The modules holistically describe the conceptualization, design, implementation, testing, and usability aspects of each projects, and where applicable, includes Jupyter notebooks and other aids to allow students to recreate many aspects of the project.