13 papers
GeGS-PCR: Effective and Robust 3D Point Cloud Registration with Two-Stage Color-Enhanced Geometric-3DGS Fusion
Jiayi Tian, Haiduo Huang, Tian Xia +2
We address the challenge of point cloud registration using color information, where traditional methods relying solely on geometric features often struggle in low-overlap and incom…
SelecTKD: Selective Token-Weighted Knowledge Distillation for LLMs
Haiduo Huang, Jiangcheng Song, Yadong Zhang +1
Knowledge distillation (KD) is a standard route to compress Large Language Models (LLMs) into compact students, yet most pipelines uniformly apply token-wise loss regardless of tea…
FastEagle: Cascaded Drafting for Accelerating Speculative Decoding
Haiduo Huang, Jiangcheng Song, Wenzhe Zhao +1
Speculative decoding accelerates generation by drafting candidates and verifying them in parallel, yet state-of-the-art drafters (e.g., EAGLE) still require N sequential passes to…
SpecVLM: Fast Speculative Decoding in Vision-Language Models
Haiduo Huang, Fuwei Yang, Zhenhua Liu +4
Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems cons…
NMS: Efficient Edge DNN Training via Near-Memory Sampling on Manifolds
Boran Zhao, Haiduo Huang, Qiwei Dang +3
Training deep neural networks (DNNs) on edge devices has attracted increasing attention due to its potential to address challenges related to domain adaptation and privacy preserva…
DeepKD: A Deeply Decoupled and Denoised Knowledge Distillation Trainer
Haiduo Huang, Jiangcheng Song, Yadong Zhang +1
Recent advances in knowledge distillation have emphasized the importance of decoupling different knowledge components. While existing methods utilize momentum mechanisms to separat…
Partial Convolution Meets Visual Attention
Haiduo Huang, Fuwei Yang, Dong Li +5
Designing an efficient and effective neural network has remained a prominent topic in computer vision research. Depthwise onvolution (DWConv) is widely used in efficient CNNs or Vi…
KernelDNA: Dynamic Kernel Sharing via Decoupled Naive Adapters
Haiduo Huang, Yadong Zhang, Yinghui Xu +1
Dynamic convolution enhances model capacity by adaptively combining multiple kernels, yet faces critical trade-offs: prior works either (1) incur significant parameter overhead by…
Gumiho: A Hybrid Architecture to Prioritize Early Tokens in Speculative Decoding
Jinze Li, Yixing Xu, Haiduo Huang +4
Speculative decoding (SPD) aims to accelerate the auto-regressive token generation process of a target Large Language Model (LLM). Some approaches employ a draft model with multipl…
Jakiro: Boosting Speculative Decoding with Decoupled Multi-Head via MoE
Haiduo Huang, Fuwei Yang, Zhenhua Liu +7
Speculative decoding (SD) accelerates large language model inference by using a smaller draft model to predict multiple tokens, which are then verified in parallel by the larger ta…
Partial Channel Network: Compute Fewer, Perform Better
Haiduo Huang, Tian Xia, Wenzhe zhao +1
Designing a module or mechanism that enables a network to maintain low parameters and FLOPs without sacrificing accuracy and throughput remains a challenge. To address this challen…
Nearly Lossless Adaptive Bit Switching
Haiduo Huang, Zhenhua Liu, Tian Xia +2
Model quantization is widely applied for compressing and accelerating deep neural networks (DNNs). However, conventional Quantization-Aware Training (QAT) focuses on training DNNs…
FTP: A Fine-grained Token-wise Pruner for Large Language Models via Token Routing
Zekai Li, Jintu Zheng, Ji Liu +9
Recently, large language models (LLMs) have demonstrated superior performance across various tasks by adhering to scaling laws, which significantly increase model size. However, th…