|
I am a Computer Sciences undergraduate at the University of Wisconsin–Madison, graduating in May 2027. My research interests are systems for AI, edge inference, operating systems, and AI compilers. With Professor Yiying Zhang at WukLab, UC San Diego, I work on efficient AI inference on edge devices, with a current focus on NPU compilers and runtimes. Previously, I studied filesystem interference with Professor Michael Swift and built LinuxGuard with Professor Remzi Arpaci-Dusseau and Dr. Vinay Banakar at The ADvanced Systems Laboratory (ADSL), UW–Madison. [show email] / CV / Google Scholar / GitHub / LinkedIn |
|
|
My research connects accelerator compilation, memory management, device placement, and systems security. I build runtimes and analysis tools, then evaluate their behavior under controlled workloads. Representative projects are highlighted. |
 | Xuming Huang · advised by Prof. Yiying Zhang Undergraduate Researcher · WukLab, UC San Diego · Jan. 2026–presentDeveloped a JIT NPU runtime that overlaps CPU graph compilation with NPU inference and grows the KV cache with context: 91.98% less foreground compilation time and 33.05% lower average KV memory, while preserving model quality and decoding. Cached sorted compiler dependency vectors to reduce CPU sorting time by 94.2%, and built CPU/GPU/NPU benchmarks to guide device placement. |
 | Xuming Huang · advised by Prof. Michael Swift Honors Course Research Project · UW–Madison · Jan.–May 2026Measured how background filesystem reads interfere with a fixed LLM inference workload: median latency increased 17.1%. Windows PerfMon showed stable working sets and page-fault counts plateauing near 29.5K, supporting shared-resource contention as the explanation rather than working-set thrashing. |
 | Xuming Huang · advised by Prof. Remzi Arpaci-Dusseau and Dr. Vinay Banakar · The ADvanced Systems Laboratory (ADSL) Undergraduate Researcher · UW–Madison · Jan.–Nov. 2025Built an LLM-driven pipeline over 50K+ Linux kernel commits to generate Clang-Tidy analyzers. Clustering and compiler-feedback repair produced 4× more valid checkers at 73% precision. A full-kernel validation harness uncovered 43 long-latent bugs with an average age of 4.7 years. |
 | Xuming Huang · coauthor; advised by Prof. Xing Hu Undergraduate Researcher · USST · Jan. 2024–Sept. 2026Contributed to experimental validation and performance analysis of CASH, which predicts workload–hardware affinity for CPU/GPU placement. In 5,000-task simulations using MIT Supercloud traces, the scheduler achieved 97.1% resource-matching accuracy, 46.5% lower response time, and 36.8% lower modeled energy than Meta-RHDC. |
| View All Works → |
|
I create interactive visualizations and educational tools for computer science concepts. See All Projects → |
 Algorithms Learning Resources Implementations and visualizations across six algorithm topics. |  AI Systems Learning Resources Transformer and GPT implementations from scratch in PyTorch. |  Heap Allocator Memory allocation, block coalescing, and fragmentation analysis. |
|
Deep Learning Architecture Tutorials: |
|
|
|
Swift 5.9, SwiftUI, Firebase, Core Location iOS application providing location-based legal information for tenants across all 50 U.S. states. Features automatic GPS-based state law detection, comprehensive legal database covering tenant rights, traffic laws, employment regulations, and consumer protections. Includes 10+ legal document templates, smart notifications for law changes, and offline law access. Implements freemium model with pro subscription for unlimited features. [GitHub] |
|
Feel free to clone my template Xuming Huang |