GPU & high-performance computing
CUDA algorithms for challenging mathematical optimization.
At J.P. Morgan Chase’s Global Technology Applied Research group, I optimized algorithms with CUDA to tackle hard mathematical problems. A central example is massively parallel memetic tabu search for low-autocorrelation binary sequences (LABS), combining block- and thread-level parallelism with compact binary representations. Related work also explores GPU-based continuous local search for hybrid SAT.
Algorithm design meets GPU architecture
At J.P. Morgan Chase’s Global Technology Applied Research group, I worked on CUDA optimization for challenging mathematical problems. The key is to adapt both the search algorithm and its data representation to the parallelism and memory hierarchy of a GPU.
Low-autocorrelation binary sequences
LABS is a difficult binary optimization problem. Our work develops a massively parallel implementation of memetic tabu search: GPU blocks and threads explore candidates while compact binary representations support efficient computation and memory use. Candidate evaluation and population updates connect local search with the broader search process.
Related GPU research
A separate collaborative line investigates massively parallel continuous local search for hybrid SAT on GPUs. Both directions explore how parallel hardware can expand the reach of optimization algorithms, while targeting different objectives and search methods.