- Profiled a large-scale K-Means implementation on an 800MB dataset using fine-grained instrumentation to isolate runtime bottlenecks, identifying distance computation and point-assignment phases as dominant contributors.
- Redesigned the hotspot computation using data-parallel execution with std::thread, partitioning datapoints across worker threads while minimizing synchronization overhead and preserving algorithmic correctness.
- Evaluated scalability through controlled performance experiments, analyzing workload distribution, cache behavior, and speedup trade-offs to achieve a 2x+ reduction in end-to-end runtime.
/// project
Parallel K-Means Optimization
Stanford CS149 Parallel Computing
- C++
- Machine Learning
- Multi-Threading
- Kernel Profiling