Device Platform

• Developed a few-shot topology-aware segmentation model for quantitative analysis of cellular and tissue morphology.

• Implemented Triton kernels for topology-preserving segmentation.

• Developed RL algorithms to learn policies for maximizing information-theoretic measures at scale via submodular maximization.

• Developed multi-modal algorithms for MRI and clinical data to identify biomarkers for Type 1 diabetes.

• Implemented high-performance parallel RL environment executor with C++11 threadpool, pybind11 and thread-safe pre-fetching that is x40 faster than Python implementation for terabyte-scale giga-pixel image data set.

• Researched gradient-free optimization schemes based on differential evolution for distributed neural network training.

• Implemented highly scalable OpenMPI & CUDA optimized algorithms that achieved x350 speed up on multi-GPU systems.

• Developed static code analysis methods based on linear algebraic transforms on LLVM IR for symbolic execution.

• Worked on Hierarchical & Multi-Agent RL for sample-efficient multi-target classification in giga-pixel images.

• Augmented self-supervised semantic segmentation models with Bayesian predictions for heteroscedastic uncertainty quantification.