Active Matter JEPA
JEPA-style self-supervised learning on active matter simulation videos to predict future latent states without using physical parameter labels during representation learning.
JEPAself-supervised learningphysical systemslatent predictionlinear probeskNN probes
Detailed page
Links: Report (placeholder URL) ยท Code (placeholder URL)
Hierarchical JEPA / World Models
Exploring hierarchical latent world models where higher-level abstractions are learned from unresolved predictive structure across time.
world modelshierarchyresidual predictionlatent actionsplanning
Detailed page
Links: Slides (placeholder URL)
Cryo-EM and Protein Sequence Representation Learning
Multimodal representation learning between Cryo-EM density maps and protein sequences.
Cryo-EMprotein sequencesmultimodal learningcontrastive learningJEPA
Detailed page
Links: Paper (placeholder URL)
Active Spectral Learning
Bayesian machine learning project on active learning for Gaussian Processes with spectral kernels.
Gaussian ProcessesBayesian MLactive learningspectral kernels
Detailed page
Links: Report (placeholder URL)
ADAS / Applied ML Systems
Perception and machine learning systems for ADAS applications from internship work at Robert Bosch India.
ADAScomputer visionperceptionautonomous drivingapplied ML
Detailed page
Links: Drive (placeholder URL)