Pratyay Dutta
I specialize in computer vision, deep learning, and physics-based modeling. My research focuses on designing algorithms that learn effectively from data and bridge physical principles with neural networks.
I aim to push the boundaries of robust and interpretable AI by collaborating with interdisciplinary teams and contributing solutions that impact real-world vision tasks, from person re-identification to event-based perception.
"Stay Hungry. Stay Foolish"
Academic Background
Doctor of Philosophy (Ph.D.)
Bachelor of Engineering
Primary, Middle and High School
Publications
SAKE: Structure Aware Kernel Experts for Event-Based Object Detection
SAKE replaces heavy AI calculations with a heat-flow mechanism that mimics how information spreads along the edges of objects. This physics-inspired approach allows the model to "see" clearly in low light or fast motion with much less computing power.
POANet: Parts-based Occlusion Aware Network for Person Re-ID
A parts-based occlusion-aware network for robust person re-identification, addressing partial visibility through structured part-level feature learning and attention-guided reconstruction.
Latent Diffusion-Guided Feature Inpainting for Occluded Person Re-Identification With Hybrid Re-Ranking
Our De-Occluder (DDO) uses AI to "fill in" hidden parts of a person's image, allowing for accurate identification even when someone is partially blocked. Combined with a smart re-ranking system, it achieves state-of-the-art results in identifying people through dense crowds.
Hierarchical Vaccine Allocation based on Epidemiological and Behavioral Considerations
An optimized hierarchical vaccine distribution strategy incorporating epidemiological and behavioral factors with clustering-based zone allocation for improved vaccine redistribution across demographics.
Generalizable Multi-Vaccine Distribution Strategy based on Demographic and Behavioral Heterogeneity
A generalizable multi-vaccine distribution strategy using linear optimization and clustering algorithms to address demographic and behavioral heterogeneity across geographic zones.
Experience
Present
Research Assistant
University of California, Riverside
Conducting research under Dr. Bir Bhanu on event-based vision with anisotropic heat diffusion, virtual cloth try-off using Stable Diffusion, and latent diffusion-guided feature inpainting for occluded person re-identification.
Aug 2022
Research Intern
Dalhousie University, Canada
Built a predictive modeling pipeline on Lichess chess data and trained a PPO-based RL agent capable of defeating amateur players below a target FIDE rating, under Dr. Yannick Marchand.
Nov 2021
Research Intern
Virginia Commonwealth University, USA
Optimized multi-vaccine distribution strategies using linear programming and clustering; work resulted in publications at IEEE BIBM 2021 and IEEE/ACM TCBB 2022, under Dr. Preetam Ghosh.
Awards & Honors
Dean's Fellowship
University of California, Riverside
Awarded for outstanding research acumen upon admission. Acceptance rate: 15%.
MITACS Fellowship
MITACS Canada
Funded summer research internship at Dalhousie University, Canada. Acceptance rate: 18%.
Bishop Thoburn Award — Student of the Year
Calcutta Boys' School
Recognized for outstanding achievement in academics, sports, and extracurricular activities. Awarded to 1 in 120 students.
Capabilities
Featured Work
Talk To Segment
Edit only what you mean: segment by text, rewrite by prompt, keep everything else intact.
Remember To Play
When memory enters the Q-network, the agent stops guessing and starts planning.
ImpulseForge
Physics that feels honest: impulse-level collision logic, friction, and MuJoCo cross-checks in one loop.
EyeWitness
Cross-camera identity matching with transformer attention that remembers the whole person, not just local patches.
TriStrand
Three DNA strands, one aligned story: 3D dynamic programming at near-C speed with Python ergonomics.
Get in Touch
Have a question, collaboration idea, or just want to say hello?
Fill in the form and I'll get back to you.