Current Research
Domain-Specific Foundation Models
Foundation models allow us to form robust representations with less annotation costs. However, medical images come with their own challenges and properties requiring domain-specific strategies.
Trustworthy ML
Deep learning models offer great promise for improving speed and quality of medical care. However, they are overconfident in cases where humans could quickly realize that they aren't sure.
Observational Supervision
A key challenge with healthcare is the lack of training labels in medical data. Instead, we extract information from passively-collected metadata in clinical practice to supervise our models.
Past Research
Phenotyping Mental Health
Neuropsychological disorders are the No.1 cause of disability in the world. But, diagnosis is hard due to lack of objective tests. Large-scale text narratives instead can aid phenotyping mental health.
Epiretinal Prostheses
An epiretinal prosthesis electrically stimulates cells in the retina (RGCs) to compensate for total or partial blindness due to photoreceptor loss to transmit vision-like messages to the brain.
Thermionic Energy Converters
Thermionic Energy Converters (TECs) are novel devices that convert heat to electricity directly, potentially reaching even 35% in efficiency when micro-fabrication techniques are applied to it.
Digital Multipliers
Designed a novel multiplier based on Vedic techniques. The design achieved an optimum time delay, area occupancy and power consumption of 4.971 ns, 75013 sq.nm and 28.73 mW respectively.