Portfolio item number 1
Short description of portfolio item number 1
Posts by Collection
portfolio
Portfolio item number 2
Short description of portfolio item number 2
publications
Liberator: A Data Reuse Framework for Out-of-Memory Graph Computing on GPUs
Published in IEEE Transactions on Parallel and Distributed Systems (TPDS), 2023
Liberator is a data reuse framework that enables efficient out-of-memory graph computing on GPUs by intelligently managing data transfers between host and device memory.
S. Li, R. Tang, J. Zhu, Z. Zhao, X. Gong, W. Wang, J. Zhang, P.-C. Yew. "Liberator: A Data Reuse Framework for Out-of-Memory Graph Computing on GPUs." IEEE Transactions on Parallel and Distributed Systems (TPDS), 34(6): 1954-1967, 2023.
Download Paper | Code
OneGraph: A Cross-Architecture Framework for Large-Scale Graph Computing on GPUs Based on oneAPI
Published in CCF Transactions on High Performance Computing (CCF-THPC), 2024
OneGraph is a cross-architecture graph computing framework built on oneAPI that enables portable and efficient large-scale graph processing across different GPU architectures.
S. Li, J. Zhu, J. Han, Y. Peng, Z. Wang, X. Gong, G. Wang, J. Zhang, X. Wang. "OneGraph: A Cross-Architecture Framework for Large-Scale Graph Computing on GPUs Based on oneAPI." CCF Transactions on High Performance Computing (CCF-THPC), 6(2): 179-191, 2024.
Download Paper
DR-CircuitGNN: Training Acceleration of Heterogeneous Circuit Graph Neural Network on GPUs
Published in ACM International Conference on Supercomputing (ICS), 2025
DR-CircuitGNN accelerates the training of heterogeneous circuit graph neural networks on GPUs through novel data reuse strategies and GPU-optimized computation kernels.
Y. Luo, S. Li, J. Tao, K. G. Thorat, X. Xie, H. Peng, N. Xu, C. Ding, S. Huang. "DR-CircuitGNN: Training Acceleration of Heterogeneous Circuit Graph Neural Network on GPUs." In Proceedings of the 39th ACM International Conference on Supercomputing (ICS '25), 2025.
Download Paper
CudaForge: An Agent Framework with Hardware Feedback for CUDA Kernel Optimization
Published in arXiv preprint, 2025
CudaForge is an agentic framework that automatically optimizes CUDA kernels using LLM-based agents with hardware profiling feedback, achieving significant speedups over hand-tuned baselines.
Z. Zhang, R. Wang, S. Li, Y. Luo, M. Hong, C. Ding. "CudaForge: An Agent Framework with Hardware Feedback for CUDA Kernel Optimization." arXiv preprint arXiv:2511.01884, 2025.
Download Paper
XuanJia: A Comprehensive Virtualization-Based Code Obfuscator for Binary Protection
Published in arXiv preprint, 2026
We present XuanJia, a comprehensive virtualization-based code obfuscator that leverages virtual machine protection to safeguard binary programs against reverse engineering and tampering attacks.
X. Zou, X. Gong, J. Zhang, S. Li, P.-C. Yew. "XuanJia: A Comprehensive Virtualization-Based Code Obfuscator for Binary Protection." arXiv preprint arXiv:2601.10261, 2026.
Download Paper
StitchCUDA: An Automated Multi-Agents End-to-End GPU Programming Framework with Rubric-based Agentic Reinforcement Learning
Published in International Conference on Machine Learning (ICML 2026), 2026
Accepted to ICML 2026. StitchCUDA is a multi-agent framework for end-to-end GPU program optimization using rubric-based agentic reinforcement learning, achieving ~1.72x speedup over multi-agent baselines and ~2.73x over RL model baselines on KernelBench.
S. Li, Z. Zhang, W. Chen, Y. Luo, M. Hong, C. Ding. "StitchCUDA: An Automated Multi-Agents End-to-End GPU Programming Framework with Rubric-based Agentic Reinforcement Learning." International Conference on Machine Learning (ICML), 2026.
Download Paper | Code
FPTC: A Fast Parallel Transform-based Codec for Efficient Asymmetric Signal Compression
Published in arXiv preprint, 2026
FPTC is a high-throughput asymmetric signal codec pairing a lightweight sequential encoder with a massively parallel GPU decoder, outperforming existing frameworks in compression ratio across biomedical, seismic, power-grid, and meteorological signals.
B. Mechels, R. Billmeyer, A. Chen, S. Li, C. Ding. "FPTC: A Fast Parallel Transform-based Codec for Efficient Asymmetric Signal Compression." arXiv preprint arXiv:2605.01086, 2026.
Download Paper
CUDAHercules: Benchmarking Hardware-Aware Expert-level CUDA Optimization for LLMs
Published in arXiv preprint, 2026
CUDAHercules is a benchmark that evaluates LLM-generated CUDA against end-to-end human-expert SOTA systems, spanning single kernels, module-level operators, full applications, and unsolved challenge tasks across Ampere, Hopper, and Blackwell GPUs.
S. Li, Z. Zhang, G. Sun, Y. Luo, W. Chen, Y. Wang, M. Hong, C. Ding. "CUDAHercules: Benchmarking Hardware-Aware Expert-level CUDA Optimization for LLMs." arXiv preprint arXiv:2605.08467, 2026.
Download Paper
CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging
Published in arXiv preprint, 2026
CUDABeaver is a benchmark for CUDA debugging built from real failing workspaces produced during LLM-based CUDA generation, evaluating whether a fixer truly repairs failing CUDA code or merely finds a slower test-passing replacement across 213 tasks and seven frontier LLMs.
S. Li*, H. Chen*, M. Fazzini, C. Ding. "CUDABeaver: Benchmarking LLM-Based Automated CUDA Debugging." arXiv preprint arXiv:2605.08455, 2026.
Download Paper | Code
GSR-GNN: Training Acceleration and Memory-Saving Framework of Deep GNNs on Circuit Graph
Published in ACM/IEEE Design Automation Conference (DAC), 2026
GSR-GNN enables training GNNs with up to hundreds of layers on circuit graphs while reducing both compute and memory overhead, achieving up to 87.2% peak memory reduction and over 30x training speedup.
Y. Luo, S. Li, Y. Feng, V. Kancharla, S. Huang, C. Ding. "GSR-GNN: Training Acceleration and Memory-Saving Framework of Deep GNNs on Circuit Graph." In Proceedings of the 63rd ACM/IEEE Design Automation Conference (DAC '26), 2026.
Download Paper
SparseDitto: Customizing GPU Kernels for Different Sparsity Patterns with LLM-Based Agentic System
Published in arXiv preprint, 2026
SparseDitto is an LLM-based agentic system that constructs a customized GPU kernel for each matrix, operator, and target GPU, supporting SpMV, SpMM, and SpGEMM, achieving a geometric-mean speedup of 2.68x over cuSPARSE on an RTX PRO 6000 and 2.79x on an H200.
S. Li, G. Sun, J. Tang, Y. Wang, M. Hong, C. Ding. "SparseDitto: Customizing GPU Kernels for Different Sparsity Patterns with LLM-Based Agentic System." arXiv preprint arXiv:2608.05033, 2026.
Download Paper
talks
Talk 1 on Relevant Topic in Your Field
Published:
This is a description of your talk, which is a markdown files that can be all markdown-ified like any other post. Yay markdown!
Conference Proceeding talk 3 on Relevant Topic in Your Field
Published:
This is a description of your conference proceedings talk, note the different field in type. You can put anything in this field.
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.