Haotong Qin

I am a PhD student (2019.09-) at the State Key Laboratory of Software Development Environment and Shen Yuan Honors College, Beihang University, supervised by Prof. Wei Li and Prof. Xianglong Liu. And I am a visiting PhD student (2022.10-) at the Computer Vision Laboratory, ETH Zürich, supervised by Prof. Fisher Yu. I obtained my BSc degree in Computer Science and Engineering (Summa Cum Laude) from Beihang University (2015.09-2019.07). I interned at ByteDance AI Lab, Tencent WeiXin Group, and Microsoft Research Asia in 2019-2022.

Email: qinhaotong@buaa.edu.cn / qinhaotong@gmail.com

Google Scholar / Github / Linkedin / CV

Research

I'm interested in network binarization and quantization. And my research goal is to enable state-of-the-art neural network models to be deployed on resource-limited hardware, including the compression for different neural architectures, and the flexible deployment on various hardware.

Recent News

[2023.02] One first-authored paper for FSMN binarization is accepted by IEEE TNNLS.

[2022.12] I am selected for the Rising Stars in AI Symposium 2023 organized by KAUST AI initiative.

[2022.10] I obtain the ByteDance Scholarship.

[2022.10] Our BiPointNet (ICLR'21) integrated into Amazon's Deep Graph Library (DGL).

[2022.09] One first-authored paper for model binarization is accepted by IJCV.

[2022.07] I obtain the China Scholarship Council (CSC) scholarship.

[2022.06] One co-authored paper for ViT quantization is accepted by ACM MM 2022.

[2022.06] I obtain the Beihang Top-10 PhD Students Award.

[2022.05] Our BiBERT (ICLR'22) integrated into Baidu's deep learning platform PaddlePaddle.

[2022.04] One first-authored paper for FSMN binarization is accepted by IJCAI 2022.

[2022.03] Our survey paper for binary neural networks is selected to ESI Highly Cited Papers.

[2022.03] One co-authored paper for physical world robustness is accepted by CVPR 2022.

[2022.01] One first-authored paper for BERT binarization is accepted by ICLR 2022.

[2021.10] I join ByteDance AI Lab as an research intern.

[2021.09] Our BiPointNet (ICLR'21) obtain the Most Popular Paper in Beijing Area.

[2021.09] I obtain China National Scholarship.

[2021.07] One co-authored paper for object detection is accepted by ICCV 2021.

[2021.05] I obtain Beihang-Huawei Scholarship.

[2021.03] One co-authored oral paper for data-free quantization is accepted by CVPR 2021.

[2021.01] One first-authored paper for PointNet binarization is accepted by ICLR 2021.

[2020.09] I obtain China National Scholarship.

[2020.09] I release our open source project "Awesome Model Quantization".

[2020.06] I join Tencent WXG as an research intern.

[2020.04] I am invited to present our IR-Net and survey paper at JD.com, Inc. Here are the Slides.

[2020.02] One first-authored paper for model binarization is accepted by CVPR 2020.

[2020.02] One co-authored paper for video hashing is accepted by TMM.

[2020.02] One first-authored survey paper for binary neural networks is accepted by PR.

[2018.11] I join MSRA as an research intern.

Recent Events

[Workshop@CVPR2023] I am co-organizing the challenge on The 3rd workshop of Adversarial Machine Learning on Computer Vision: Art of Robustness at CVPR 2023.

[Workshop@AAAI2023] I am co-organizing the 2nd international workshop on The Practical Deep Learning in the Wild (PracticalDL-23) and Practical AI Challenge at AAAI 2023.

[Workshop@VALSE2022] I am co-organizing the student workshop at VALSE 2022.

[Workshop@CVPR2022] I am co-organizing the workshop on The Art of Robustness: Devil and Angel in Adversarial Machine Learning at CVPR 2022.

[Workshop@AAAI2022] I am co-organizing the 1st international workshop on The Practical Deep Learning in the Wild (PracticalDL-22) at AAAI 2022.

[Thematic-forum@PRCV2022] I am co-organizing the thematic forum on The Hardware-friendly Lightweight Deep Learning at PRCV 2021.

Selected Publications

You can find the full list on Google Scholar and our group publication page.

PontTuset

BiFSMNv2: Pushing Binary Neural Networks for Keyword Spotting to Real-Network Performance [PDF]
Haotong Qin, Xudong Ma, Yifu Ding, Xiaoyang Li, Yang Zhang, Zejun Ma, Jiakai Wang, Jie Luo, Xianglong Liu
IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 2023
arXiv \ Code

We present a strong yet efficient binary neural network for KWS, namely BiFSMNv2, pushing it to the real-network accuracy performance.

PontTuset

BiBERT: Accurate Fully Binarized BERT [PDF]
Haotong Qin*, Yifu Ding*, Mingyuan Zhang*, Qinghua Yan, Aishan Liu, Qingqing Dang, Ziwei Liu, Xianglong Liu
International Conference on Learning Representations (ICLR), 2022
arXiv / News: (量子位, 百度) / Baidu-PaddlePaddle /
Code

We propose BiBERT, an accurate fully binarized BERT, introducing an efficient Bi-Attention structure and a DMD scheme. BiBERT yields impressive 59.2x and 31.2x saving on FLOPs and model size.

PontTuset

Distribution-sensitive Information Retention for Accurate Binary Neural Network [PDF]
Haotong Qin, Xiangguo Zhang, Ruihao Gong, Yifu Ding, Yi Xu, Xianglong Liu
International Journal of Computer Vision (IJCV), 2022
arXiv

We present a novel Distribution-sensitive Information Retention Network (DIR-Net), which improves BNNs by distribution-sensitive optimization without increasing the overhead in the inference process.

PontTuset

BiFSMN: Binary Neural Network for Keyword Spotting [PDF]
Haotong Qin*, Xudong Ma*, Yifu Ding*, Xiaoyang Li, Yang Zhang, Yao Tian, Zejun Ma, Jie Luo, Xianglong Liu
International Joint Conference on Artificial Intelligence (IJCAI), 2022
arXiv / News: (机器之心, PaperWeekly) / Code

We present BiFSMN, an accurate and extreme-efficient binary network for KWS, achieving impressive 22.3x speedup and 15.5x storage-saving on edge hardware.

PontTuset

Towards Accurate Post-Training Quantizationfor Vision Transformer [PDF]
Yifu Ding, Haotong Qin, Qinghua Yan, Zhenhua Chai, Junjie Liu, Xiaolin Wei, Xianglong Liu
ACM Multimedia (ACM MM), 2022
arXiv

We propose a novel Accurate Post-training Quantization framework for Vision Transformer, namely APQ-ViT, which surpasses the existing post-training quantization methods by convincing margins, especially in lower bit settings.

PontTuset

BiPointNet: Binary Neural Network for Point Clouds [PDF]
Haotong Qin*, Zhongang Cai*, Mingyuan Zhang*, Yifu Ding, Haiyu Zhao, Shuai Yi, Xianglong Liu, Hao Su
International Conference on Learning Representations (ICLR), 2021
arXiv / Project / News: (量子位, 商汤学术) / Amazon-DGL /
Code
2021 The Most Popular Papers in Beijing Area (Field: Graphic and Image)

We present BiPointNet, the first model binarization approach for efficient deep learning on point clouds, giving an impressive 14.7x speedup and 18.9x storage saving on real-world devices.

PontTuset

Diversifying Sample Generation for Accurate Data-Free Quantization [PDF]
Xiangguo Zhang*, Haotong Qin*, Yifu Ding, Ruihao Gong, Qinghua Yan, Renshuai Tao, Yuhang Li, Fengwei Yu, Xianglong Liu
Computer Vision and Pattern Recognition (CVPR), 2021
Oral presentation, Acceptance Rate 4.7%
arXiv / News: (量子位, 商汤学术)

We proposed Diverse Sample Generation (DSG) scheme to mitigate the adverse effects caused by homogenization in data-free quantization, which obtained significant improvements over various networks and quantization methods.

PontTuset

Forward and Backward Information Retention for Accurate Binary Neural Networks [PDF]
Haotong Qin, Ruihao Gong, Xianglong Liu, Mingzhu Shen, Ziran Wei, Fengwei Yu, Jingkuan Song
Computer Vision and Pattern Recognition (CVPR), 2020
arXiv / News: (机器之心, 商汤学术, CVer, AI科技大本营) / Code

We propose a novel Information Retention Network (IR-Net) to retain the information that consists in the forward activations and backward gradients, and we are the first to implement and report advanced BNN speed on edge devices.

PontTuset

Binary Neural Networks: A Survey [PDF]
Haotong Qin, Ruihao Gong, Xianglong Liu, Xiao Bai, Jingkuan Song, Nicu Sebe
Pattern Recognition (PR), 2020
arXiv / News: (PaperWeekly, 机器之心) / Code
ESI Highly Cited Papers (2022)

We presented a comprehensive survey of BNNs. We investigated practical aspects of binary neural networks and gave the evaluation and discussions on different tasks. The challenges may be faced in future research were prospected.

Selected Honors and Awards

[2023]    Rising Star in AI (awarded by KAUST AI Initiative headed by Prof. Jürgen Schmidhuber)

[2022]    ByteDance Scholarship

[2022]    Beihang Top-10 PhD Students Award

[2021]    The Most Popular Paper in Beijing Area (Field: Graphic and Image)

[2021]    China National Scholarship

[2020]    China National Scholarship

[2020]    Tencent Rhino-Bird Elite

[2019]    ICPC China National Invitational Contest (Nanchang) - Gold Medal

[2018]    ACM-ICPC Chinese Collegiate Programming Contest (Ningxia) - Gold Medal

Academic Services

Organizer of Workshops: AAAI 2022/2023, CVPR 2022, PRCV 2021, VALSE 2022.

Reviewer of Journals: IEEE TPAMI/TIP/TNNLS/TMM, Pattern Recognition, etc.

Program Committee of Conferences: CVPR 2023, IJCAI 2021/22, ACM MM 2021/22, ICME 2023, etc.

Talks and Teaching

[2023.02] I am invited to talk about network binarization at KAUST. [Slides]

[2022.06] I am invited to host the VALSE Student Webinar & Panel namely "When CV meets NLP". [Media]

[2022.01] I am invited to host the VALSE Student Webinar about conference and journal rebuttal. [Media]

[2021.07] I am invited to talk about multi-scenario network quantization at J Ventures (将门创投). [Video]

[2021.06] I am invited to talk about data-free quantization at Zhidx (智东西公开课). [Video]

[2021.05] I am invited to present our DSG (CVPR 2021 oral) at MSRA Tech Talk. [Slides]

[2020.04] I am invited to present our IR-Net (CVPR 2020) and survey paper at JD AI Research. [Slides]

[Fall 2020] I am the teaching assistant in machine learning course (Beihang University).

About Me

In my free time, I like playing Chinese folk music, especially string music (Erhu, Zhonghu, etc.). In fact, I am almost a professional Erhu performer. I have studied Erhu supervised by Prof. Zaili Tian, Prof. Yang Gao, and Prof. Qingfu Zhu. I was the vice-president of the Beihang Folk Music Orchestra, here are some of the performance videos of our orchestra [Tencent Video][Bilibili].