Mao Lin (林茂)
Graduate Student at UC Merced
Researching ML/LLM systems, heterogeneous computer architectures, high-performance computing, and program analysis and optimization.
Education
Research Areas
Experience
Working on hardware/software co-design for MoE models on Samsung's AI accelerator.
Optimized PyTorch memory management for distributed LLM training, reducing memory usage by 10% to 30% on models including GPT-2 and Whisper.
Analyzed production Go services and fixed more than 50 data race issues.
Built GPU profiling and floating-point analysis tooling that found critical overflow issues in DOE applications.
Open Source Software
A profiling and analysis framework for various accelerator applications.
Tooling for guiding memory optimization in GPU-accelerated applications.
Publications
HybridGen: Efficient LLM Generative Inference via CPU-GPU Hybrid Computing
arXiv preprint, April 2026 (arXiv 2026)
PASTA: A Modular Program Analysis Tool Framework for Accelerators
The 23rd ACM/IEEE International Symposium on Code Generation and Optimization, Jan 31–Feb 4, 2026, Sydney, Australia (CGO '26)
Forest: Access-aware GPU UVM Management
The 52nd Annual International Symposium on Computer Architecture, Jun 21–25, 2025, Tokyo, Japan (ISCA '25)
Understanding Oversubscribed Memory Management for Deep Learning Training
The 5th Workshop on Machine Learning and Systems, Mar 30–Apr 3, 2025, Rotterdam, Netherlands (EuroMLSys '25)
DrGPUM: Guiding Memory Optimization for GPU-accelerated Applications
The 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Mar 25–29, 2023, Vancouver, BC, Canada (ASPLOS '23)
A Comprehensive Memory Management Framework for CPU-FPGA Heterogenous SoCs
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2022 (TCAD '22)
Poster: Squeezing GPU Memory Usage in PyTorch
PyTorch Conference '22, Dec 2022, New Orleans, LA, USA (PyTorch Conference '22)
Talks & Presentations
Professional Services
Artifact Evaluation Committee
Reviewer