Xijie Huang (Owen) ι»„ζ‚‰εˆ

I am a PhD Student at Hong Kong University of Science and Technology (HKUST), where I work on efficient deep learning (algorithm-hardware co-design) and human-centric computer vision. I am a member of Vision and System Design Lab (VSDL), advised by Prof. Tim Kwang-Ting CHENG. I am currently conducting a research internship at Microsoft Research Asia (MSRA). Previously, I work with Prof. Zhiqiang Shen from MBZUAI, Prof.Cewu Lu from SJTU, and Prof.Mani Srivastava from UCLA.

CV  /  Google Scholar  /  Twitter  /  Zhihu (ηŸ₯乎)  /  Github

profile photo
Research

My research interests lie in the general area of artificial intelligence, particularly in efficient deep learning and human-centric computer vision. More concretely, My research interests focus on quantization, algorithm-hardware co-design, human-object interation, and healthcare. Representative papers are highlighted.

SDQ: Stochastic Differentiable Quantization with Mixed Precision
Xijie Huang, Zhiqiang Shen, Shichao Li, Zechun Liu, Xianghong Hu, Jeffry Wicaksana, Eric Xing, Kwang-Ting Cheng
ICML, 2022   (Spotlight)
Paper / Talk / Slides

A novel stochastic quantization framework to learn the optimal mixed precision quantization strategy.

Fewer is More: Boosting LLM Reasoning with Reinforced Context Pruning
Xijie Huang, Li Lyna Zhang, Kwang-Ting Cheng, Fan Yang, Mao Yang
In Submission
Paper

A novel approach to push the boundaries of few-shot CoT learning to improve LLM math reasoning capabilities. We propose a coarse-to-fine pruner as a plug-and-play module for LLMs, which first identifies crucial CoT examples from a large batch and then further prunes unimportant tokens.

Efficient Quantization-aware Training with Adaptive Coreset Selection
Xijie Huang, Zechun Liu, Shih-Yang Liu, Kwang-Ting Cheng
In Submission
Paper/Code

A new angle through the coreset selection to improve the training efficiency of quantization-aware training. Our method can achieve an accuracy of 68.39% of 4-bit quantized ResNet-18 on the ImageNet-1K dataset with only a 10% subset, which has an absolute gain of 4.24% compared to the previous SoTA.

Variation-aware Vision Transformer Quantization
Xijie Huang, Zhiqiang Shen, Kwang-Ting Cheng
In Submission
Paper/Code

An analysis of the underlying difficulty of ViT quantization in the view of variation. A multi-crop knowledge distillation-based quantization method is proposed.

LLM-FP4: 4-Bit Floating-Point Quantized Transformers
Shih-Yang Liu, Zechun Liu, Xijie Huang, Pingcheng Dong, Kwang-Ting Cheng
EMNLP 2023
Paper/Code

A per-channel activation quantization scheme with additional scaling factors that can be reparameterized as exponential biases of weights, incurring a negligible cost. Our method can quantize both weights and activations in the Bert model to only 4-bit and achieves an average GLUE score of 80.07, which is only 3.66 lower than the full-precision model, significantly outperforming the previous state-of-the-art method that had a gap of 11.48.

A Tiny Accelerator for Mixed-bit Sparse CNN based on Efficient Fetch Method of SIMO SPad
Xianghong Hu, Xuejiao Liu, Yu Liu, Haowei Zhang, Xijie Huang, Xihao Guan, Luhong Liang, Chi Ying Tsui, Xiaomeng Xiong, Kwang-Ting Cheng
IEEE Transactions on Circuits and Systems II: Express Briefs (TCAS-II) 2023
Paper

A tiny accelerator for mixed-bit sparse CNNs featuring a novel scheme of single vector-based compressed sparse filter (CSF) method and single input multiple output scratch pad (SIMO SPad) to effectively compress weight and fetch the needed input activation.

VideoPro: A Visual Analytics Approach for Interactive Video Programming
Jianben He, Xingbo Wang, Kam Kwai Wong, Xijie Huang, Changjian Chen, Zixin Chen, Fengjie Wang, Min Zhu, Huamin Qu
IEEE Transactions on Visualization and Computer Graphics (VIS) 2023
Paper

A visual analytics approach to support flexible and scalable video data programming for model steering with reduced human effort.

Automated Vision-Based Wellness Analysis for Elderly Care Centers
Xijie Huang, Jeffry Wicaksana, Shichao Li, Kwang-Ting Cheng
AAAI Workshop on Health Intelligence, 2022
Paper

An automatic, vision-based system for monitoring and analyzing the physical and mental well-being of senior citizens.

FedMix: Mixed Supervised Federated Learning for Medical Image Segmentation
Jeffry Wicaksana, Zengqiang Yan, Dong Zhang, Xijie Huang, Huimin Wu, Xin Yang, Kwang-Ting Cheng
IEEE Transactions on Medical Imaging (TMI), 2022
Paper/Code

A label-agnostic unified federated learning framework, named FedMix, for medical image segmentation based on mixed image labels. In FedMix, each client updates the federated model by integrating and effectively making use of all available labeled data ranging from strong pixel-level labels, weak bounding box labels, to weakest image-level class labels.

Transferable Interactiveness Prior for Human-Object Interaction Detection
Yong-Lu Li, Siyuan Zhou, Xijie Huang, Liang Xu, Ze Ma, Hao-Shu Fang, Cewu Lu
TPAMI 2021/CVPR 2019
Paper (TPAMI version) / Paper (CVPR version) / Code

A transferable knowledge learner and can be cooperated with any HOI detection models to achieve desirable results. TIN outperforms state-of-the-art HOI detection results by a great margin, verifying its efficacy and flexibility.

PaStaNet: Toward Human Activity Knowledge Engine
Yong-Lu Li, Liang Xu, Xinpeng Liu, Xijie Huang, Mingyang Chen, Shiyi Wang, Hao-Shu Fang, Cewu Lu
CVPR 2020
Paper/ Code

A large-scale knowledge base PaStaNet, which contains 7M+ PaSta annotations. And two corresponding models are proposed: first, we design a model named Activity2Vec to extract PaSta features, which aim to be general representations for various activities.

Latent Fingerprint Image Enhancement based on progressive generative adversarial network
Xijie Huang, Peng Qian, Manhua Liu
CVPRW 2020
Paper

A latent fingerprint enhancement method based on the progressive generative adversarial network (GAN).

HAKE: Human Activity Knowledge Engine
Yong-Lu Li, Liang Xu, Xinpeng Liu, Xijie Huang, Ze Ma, Hao-Shu Fang, Cewu Lu
Preprint
Paper/ Project webpage

A large-scale Human Activity Knowledge Engine (HAKE) based on the human body part states to promote the activity understanding.

NeuronInspect: Detecting Trojan Backdoors in Deep Neural Networks via VisualInterpretability
Xijie Huang, Moustafa Alzantot, Mani.Srivastava
Preprint
Paper

A framework to detect trojan backdoors in DNNs via output explanation techniques

Misc
Reviewer, ICLR 2024, EMNLP 2023, NeurIPS 2023, ICCV 2023, CVPR 2023, WACV 2022-2023, AAAI 2022-2024, ECCV 2022
Top 10% Reviewer, ICML 2022
Program Committee, ICCV 2023 Workshop on Low-Bit Quantized Neural Networks
Technical Program Committee, ICCV 2023 Workshop on Resource Efficient Deep Learning for Computer Vision
cs188 COMP 2211 (Exploring Artificial Intelligence), Lecture: Professor Desmond Tsoi
COMP 5421 (Computer Vision), Lecture: Professor Dan Xu
COMP 1021 (Introduction to Computer Science), Lecturer: Professor David Rossitor
cs188 National Scholarship (Top 2% students in SJTU), 2017
A Class Scholarship (Top 2% students in SJTU), 2017
CSST Scholarship (USD $5,343 from UCLA), 2019
RongChang Academic Scholarship (Top 20 in SJTU), 2019
RedBird Scholarship, Postgraduate Studentship, 2020-2022
AAAI-22 Student Scholarship, 2022

Thanks to Barron's website template .