Runhua Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Runhua Zhang has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
Hongxu Jiang
4 shared publications- Hongxu Jiang
Beihang University
4 shared publications Chenhui Zhu
2 shared publications- Chenhui Zhu
School of Computer Science and Engineering, Beihang University, Beijing 100191, China
2 shared publications - Fangzheng Tian
School of Computer Science and Engineering, Beihang University, Beijing 100191, China
2 shared publications - Jinkun Geng
Stanford University
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 2863-2871
ACS: Adaptive Color Space Learning for Appearance-Preserving VIS-NIR Image Fusion
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 2863-2871
Open AccessACS: Adaptive Color Space Learning for Appearance-Preserving VIS-NIR Image Fusion
- Open Access
- 19 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (20) , 4826
Large Language Model Compression for Autoregressive Text Generation: Methods, Evaluation, and Deployment Trade-Offs
Compression changes both the information retained by a large language model and the resources required to complete a request. This narrative review examines quantization, pruning and sparsity, knowledge distillation, low-rank factorization, and token, context, and key–value cache compression for autoregressive text generation, principally with decoder-only LLMs. Its central contribution is a synthesis of changes relative to each study’s baseline, connecting compression mechanisms to capability and deployment choices. Four numerical tables distinguish quality changes at stated compression budgets, distillation gains over same-size students, and paired or independent deployment measurements. Reporting is appraised by source identity, comparison conditions, measurement scope, artefacts, and uncertainty. The reviewed cases illustrate how capacity constraints change the feasible precision choices, how allocating a shared budget changes different task scores, and how compact-model construction and request compression address complementary resource demands. A deployment-validation sequence turns these findings into candidate checks for a target workload. Remaining questions concern transfer across capabilities, delayed use of compressed information, interactions between compression stages, and complete-task cost, including agent interactions.
- 19 October 2026
- Published by MDPI AG in Electronics
- Vol. 15 (20) , 4826
Open AccessLarge Language Model Compression for Autoregressive Text Generation: Methods, Evaluation, and Deployment Trade-Offs
Compression changes both the information retained by a large language model and the resources required to complete a request. This narrative review examines quantization, pruning and sparsity, knowledge distillation, low-rank factorization, and token, context, and key–value cache compression for autoregressive text generation, principally with decoder-only LLMs. Its central contribution is a synthesis of changes relative to each study’s baseline, connecting compression mechanisms to capability and deployment choices. Four numerical tables distinguish quality changes at stated compression budgets, distillation gains over same-size students, and paired or independent deployment measurements. Reporting is appraised by source identity, comparison conditions, measurement scope, artefacts, and uncertainty. The reviewed cases illustrate how capacity constraints change the feasible precision choices, how allocating a shared budget changes different task scores, and how compact-model construction and request compression address complementary resource demands. A deployment-validation sequence turns these findings into candidate checks for a target workload. Remaining questions concern transfer across capabilities, delayed use of compressed information, interactions between compression stages, and complete-task cost, including agent interactions.
- 3 October 2026
- Published by Elsevier in Journal of Systems Architecture
- Vol. 181
Multi-dimensional dataflow optimization for efficient unstructured sparse convolutional neural network inference
- 3 October 2026
- Published by Elsevier in Journal of Systems Architecture
- Vol. 181
Multi-dimensional dataflow optimization for efficient unstructured sparse convolutional neural network inference
- 3 October 2026
- Published by Elsevier in Journal of Systems Architecture
- Vol. 181
Accelerating DNN inference on vector accelerators through joint vectorization and scheduling
- 3 October 2026
- Published by Elsevier in Journal of Systems Architecture
- Vol. 181
Accelerating DNN inference on vector accelerators through joint vectorization and scheduling
- 7 September 2026
- Published by Elsevier in Neurocomputing
- Vol. 707
RotSQ: Joint quantization and sparsification of large language models in rotated space
- 7 September 2026
- Published by Elsevier in Neurocomputing
- Vol. 707
RotSQ: Joint quantization and sparsification of large language models in rotated space
- Open Access
- 16 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 487-498
JOVS: Joint Optimization of Vectorization and Scheduling for DNN on AI DSPs
- 16 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 487-498
Open AccessJOVS: Joint Optimization of Vectorization and Scheduling for DNN on AI DSPs
- Top Cited PapersOpen Access
- 27 March 2025
- Published by MDPI AG in Electronics
- Vol. 14 (7) , 1345
Optimization Methods, Challenges, and Opportunities for Edge Inference: A Comprehensive Survey
Artificial intelligence (AI) continues to enhance production efficiency across various fields of society. Considering real-time requirements and privacy issues, edge inference (EI) is shifting from cloud scenarios to edge scenarios. As intelligent models grow in complexity and size, EI encounters significant challenges. To address these, existing research works have optimized EI from four aspects (model design, model compression, compilation toolchain, and collaborative inference) to ensure the advantages of edge intelligence. However, current works lack a comprehensive classification and discussion of existing research results. Thus, we conduct a comprehensive survey on their state-of-the-art research. Specifically, we first review the background and motivation of EI, then analyze the key issues, characteristics, and technologies of each direction. Finally, we analyze future development trends. This paper can help researchers quickly sort out the different directions of EI optimization and important related work. We hope it can bring inspiration to the researchers in these communities and motivate more follow-up works.
- 27 March 2025
- Published by MDPI AG in Electronics
- Vol. 14 (7) , 1345
Top Cited PapersOpen AccessOptimization Methods, Challenges, and Opportunities for Edge Inference: A Comprehensive Survey
Artificial intelligence (AI) continues to enhance production efficiency across various fields of society. Considering real-time requirements and privacy issues, edge inference (EI) is shifting from cloud scenarios to edge scenarios. As intelligent models grow in complexity and size, EI encounters significant challenges. To address these, existing research works have optimized EI from four aspects (model design, model compression, compilation toolchain, and collaborative inference) to ensure the advantages of edge intelligence. However, current works lack a comprehensive classification and discussion of existing research results. Thus, we conduct a comprehensive survey on their state-of-the-art research. Specifically, we first review the background and motivation of EI, then analyze the key issues, characteristics, and technologies of each direction. Finally, we analyze future development trends. This paper can help researchers quickly sort out the different directions of EI optimization and important related work. We hope it can bring inspiration to the researchers in these communities and motivate more follow-up works.
- Open Access
FlexPie: Accelerate Distributed Inference on Edge Devices with Flexible Combinatorial Optimization[Technical Report]
The rapid advancement of deep learning has catalyzed the development of novel IoT applications, which often deploy pre-trained deep neural network (DNN) models across multiple edge devices for collaborative inference.
Open AccessFlexPie: Accelerate Distributed Inference on Edge Devices with Flexible Combinatorial Optimization[Technical Report]
The rapid advancement of deep learning has catalyzed the development of novel IoT applications, which often deploy pre-trained deep neural network (DNN) models across multiple edge devices for collaborative inference.
- 20 May 2024
- Published by Elsevier in Journal of Systems Architecture
- Vol. 152
A high-performance dataflow-centric optimization framework for deep learning inference on the edge
- 20 May 2024
- Published by Elsevier in Journal of Systems Architecture
- Vol. 152
A high-performance dataflow-centric optimization framework for deep learning inference on the edge
- 6 April 2023
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 79 (13) , 14313-14341
LOCP: Latency-optimized channel pruning for CNN inference acceleration on GPUs
- 6 April 2023
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 79 (13) , 14313-14341
LOCP: Latency-optimized channel pruning for CNN inference acceleration on GPUs