Learning to calibrate battery models using Deep Reinforcement Learning

Learning to calibrate battery models using lyapunov constrained actor-critic RL. Intial results has been presented at NeurIPS 2020 workshop on ML4Eng

Deep-direct Visual Localization using learned feature optimization

Visual Localization using direct image alignment on learned deep visual feature maps. Initial results were presented at CVPR 2020 workshop on VisLocOdomMap

Continual Learning in Image Classification

Semester project with the Digital Pathology team at IBM Research, Zurich. I proposed a knowledge distillation based replay-free class-incremental learning method. This method was inspired from Learning without Forgetting (LwF), but improves accuracy for remembering older tasks using an auxiliary data.

Making robots dance (in simulation)

Using Inverse Kinematics (IK) to build different gaits for the hexapod

Parallel algorithms for Subgraph Isomorphism (OpenMP and MPI)

Parallelization of subpgraph isomorphism solvers- VF2 ang Glasgow