Graph Analytics

Algorithmic Focus

My graph analytics research concentrates on exposing parallelism and eliminating redundant computations in large, irregular workloads. I design techniques that stabilize vertex values, compress traversal work, and adaptively prioritize updates so that the most impactful edges are processed first.

Hardware-Aware Implementations

The frameworks balance computation across heterogeneous targets—including CPUs, GPUs, and custom accelerators—by matching graph partitions to each platform’s strengths. Compiler-guided transformations manage synchronization, tiling, and memory coalescing to keep pipelines full even for billion-edge graphs.

Impact

• Predictable execution time and memory usage on massive, dynamic graphs.
• Improved resilience and accuracy for streaming and evolving analytics.
• A reusable set of formulations that integrate seamlessly with my accelerator and compiler projects.