Xiangguo Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Xiangguo Zhang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Haotong Qin
ETH Zurich
1 shared publication - Jiakai Wang
Zhongguancun Laboratory, Beijing, China
1 shared publication - Jiwen Lu
Tsinghua University
1 shared publication - Jun Xie
School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu 610054, China
1 shared publication Weiliang Zeng
Division of Wireless R&D, Qualcomm Research, San Diego, CA 92121, USA
1 shared publication- Xianglong Liu
Beihang University, Beijing, China
1 shared publication
Latest Publications
- 4 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 45 (10) , 11689-11706
Diverse Sample Generation: Pushing the Limit of Generative Data-Free Quantization
Generative data-free quantization emerges as a practical compression approach that quantizes deep neural networks to low bit-width without accessing the real data. This approach generates data utilizing batch normalization (BN) statistics of the full-precision networks to quantize the networks. However, it always faces the serious challenges of accuracy degradation in practice. We first give a theoretical analysis that the diversity of synthetic samples is crucial for the data-free quantization, while in existing approaches, the synthetic data completely constrained by BN statistics experimentally exhibit severe homogenization at distribution and sample levels. This paper presents a generic D iverse S ample G eneration ( DSG ) scheme for the generative data-free quantization, to mitigate detrimental homogenization. We first slack the statistics alignment for features in the BN layer to relax the distribution constraint. Then, we strengthen the loss impact of the specific BN layers for different samples and inhibit the correlation among samples in the generation process, to diversify samples from the statistical and spatial perspectives, respectively. Comprehensive experiments show that for large-scale image classification tasks, our DSG can consistently quantization performance on different neural architectures, especially under ultra-low bit-width. And data diversification caused by our DSG brings a general gain to various quantization-aware training and post-training quantization approaches, demonstrating its generality and effectiveness.
- 4 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 45 (10) , 11689-11706
Diverse Sample Generation: Pushing the Limit of Generative Data-Free Quantization
Generative data-free quantization emerges as a practical compression approach that quantizes deep neural networks to low bit-width without accessing the real data. This approach generates data utilizing batch normalization (BN) statistics of the full-precision networks to quantize the networks. However, it always faces the serious challenges of accuracy degradation in practice. We first give a theoretical analysis that the diversity of synthetic samples is crucial for the data-free quantization, while in existing approaches, the synthetic data completely constrained by BN statistics experimentally exhibit severe homogenization at distribution and sample levels. This paper presents a generic D iverse S ample G eneration ( DSG ) scheme for the generative data-free quantization, to mitigate detrimental homogenization. We first slack the statistics alignment for features in the BN layer to relax the distribution constraint. Then, we strengthen the loss impact of the specific BN layers for different samples and inhibit the correlation among samples in the generation process, to diversify samples from the statistical and spatial perspectives, respectively. Comprehensive experiments show that for large-scale image classification tasks, our DSG can consistently quantization performance on different neural architectures, especially under ultra-low bit-width. And data diversification caused by our DSG brings a general gain to various quantization-aware training and post-training quantization approaches, demonstrating its generality and effectiveness.
- 27 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 24 (11) , 2518-2521
Deep Learning-Based Precoder Design in MIMO Systems With Finite-Alphabet Inputs
It is a challenge to apply the precoder design maximizing the mutual information with finite-alphabet inputs to practical multiple-input multiple-output (MIMO) systems, because it needs to iteratively solve an optimization problem, which is difficult to satisfy the requirement of real time. This paper develops a deep learning based precoding scheme, which employs the property of deep neural network (DNN) as approximator of functions. Simulation results show that a DNN can accurately learn the input-output relationship of a nearly optimal precoder achieved by the traditional interior-point method (IPM); moreover, in different MIMO scenarios, a trained DNN of small size offers almost the same performance as the nearly optimal precoder, but with huge improvement in efficiency, especially in cases of higher modulation and more antennas. The improved efficiency makes it possible to be applied to practical communication systems.
- 27 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Communications Letters
- Vol. 24 (11) , 2518-2521
Deep Learning-Based Precoder Design in MIMO Systems With Finite-Alphabet Inputs
It is a challenge to apply the precoder design maximizing the mutual information with finite-alphabet inputs to practical multiple-input multiple-output (MIMO) systems, because it needs to iteratively solve an optimization problem, which is difficult to satisfy the requirement of real time. This paper develops a deep learning based precoding scheme, which employs the property of deep neural network (DNN) as approximator of functions. Simulation results show that a DNN can accurately learn the input-output relationship of a nearly optimal precoder achieved by the traditional interior-point method (IPM); moreover, in different MIMO scenarios, a trained DNN of small size offers almost the same performance as the nearly optimal precoder, but with huge improvement in efficiency, especially in cases of higher modulation and more antennas. The improved efficiency makes it possible to be applied to practical communication systems.