Scholar
Hongwu Peng
Google Scholar ID: 9P2qtQoAAAAJ
Researcher @ Adobe Research. Ph.D. from University of Connecticut
Citations & Impact
Citations
1,996
H-index
21
i10-index
45
Publications
20
Co-authors
30
list available
Publications
Resume (English only)
Academic Achievements
- Paper 'LoT: Regularization for Improving Model Generalization' accepted by NeurIPS 2024, effective across LLM fine-tuning, game RL, small LM pretraining, and image classification.
- Paper 'MaxK-GNN: Accelerating GNN Training via MaxK Nonlinearity & GPU Kernel Co-design' accepted by ASPLOS 2024.
- Paper 'LinGCN: Accelerating GCN Private Inference under Homomorphic Encryption' accepted by NeurIPS 2023.
- Released Medusa: an easy-to-use framework that accelerates LLM generation using multiple lightweight decoding heads without draft models.
- Paper 'AQ2PNN: Adaptive Quantization for Private Inference under Multi-Party Computation' accepted by MICRO 2023.
- Paper 'Accel-GCN: Graph Learning Acceleration on GPUs' accepted by ICCAD 2023.
- Paper 'AutoReP: ReLU Replacement for Fast Private Inference under MPC' accepted by ICCV 2023.
- Paper 'PASNet: NAS for Private Inference Acceleration under MPC' accepted by DAC 2023.
- Paper on 'GNN Sparsification' accepted by ICCD 2023.
- Paper on 'Transformer Model Acceleration' accepted by DAC 2022.
- Awarded Synchrony Fellowship by UConn CSE in 2024.
- Awarded GE Fellowship by UConn School of Engineering in 2023.
- Awarded Predoctoral Fellowship by UConn CSE in 2023.
- Awarded Taylor L. Booth Predoctoral Fellowship (top 1 scholarly achievement) by UConn CSE in 2022.
