Heterogeneous Computing

Kubism: Disassembling and Reassembling K-Means Clustering for Mobile Heterogeneous Platforms

Kubism disassembles and reassembles the K-means algorithm to better utilize CPU and GPU resources on mobile heterogeneous platforms, achieving up to 2.65× speedup per iteration and an average 1.23× end-to-end improvement on NVIDIA Jetson Orin AGX.

Futurism: Accelerating K-Means Clustering on Embedded Systems via Vector Engines

Futurism combines SIMD-based distance computation, a sample-wise data layout, and dynamic CPU-GPU workload distribution, achieving a 10% end-to-end speedup over prior CPU-GPU heterogeneous implementations on NVIDIA Jetson Orin AGX.