Research

My philosphy of research in deep learning (and sometimes ML) is to reduce/remove its black box nature. The output of my research work has potential applications in Healthcare and Biometrics. Therefore, the need for explainability is of paramount importance.

Affective Computing

Considering this, my PhD thesis is two-fold: Modelling the complex affect data by combining representation learning techniques and Flow-based generative models, and developing new methods to predict/classify different affective components. Currently, the existing work is limited to structured modalities like images but I intend to extend it to unstructured data like continuous signals and tabular data.

Healthcare

I am also working on a collaborative project with the Social Behaviour department at USF to develop techniques for automatic recognition of PTSD in children. This study include using computer vision tools such as Openface to analyze videos of participants with potential trauma interacting with a psychiatrist and develop an algorithm to succesfully predict PTSD. The primary work is to explain why the ML/DL model thinks a participant does or does not have PTSD.

Biometrics

A new venue in my research includes a collaborative work under NSF wherein continuous authentication (CA) on mobile devices using unstructured modalities like physiological signals, EEG etc. Plenty of importance is given to practical implementation of CA considering the ethical aspects of the data being collected.