Rate-Distortion Theory by and for Energy-Based Models

被引:0
作者
Li, Qing [1 ]
Guyot, Cyril [1 ,2 ]
机构
[1] Western Digital Res, Milpitas, CA 95035 USA
[2] Alibaba Cloud, Sunnyvale, CA 94085 USA
关键词
Rate-distortion; Noise reduction; Distortion; Neural networks; Training; Electronics packaging; Source coding; Rate-distortion theory; energy-based models; batch denoising; Blahut-Arimoto algorithm;
D O I
10.1109/TCOMM.2024.3361531
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
In this work, we examine the relationship between rate-distortion theory and energy-based models (EBMs). We demonstrate that EBMs can be used to approximate the rate-distortion approaching posterior, as in the Blahut-Arimoto (BA) algorithm, and to solve batch denoising problems using the posterior distribution learned by EBMs. Our results highlight the potential of EBMs to enhance the efficiency of rate-distortion theory analysis and vice versa.
引用
收藏
页码:4072 / 4083
页数:12
相关论文
共 47 条
[1]  
[Anonymous], 2006, Predicting Structured Data
[3]  
Ba J, 2014, ACS SYM SER
[4]  
Bengio Y, 2001, ADV NEUR IN, V13, P932
[5]   COMPUTATION OF CHANNEL CAPACITY AND RATE-DISTORTION FUNCTIONS [J].
BLAHUT, RE .
IEEE TRANSACTIONS ON INFORMATION THEORY, 1972, 18 (04) :460-+
[6]  
Bora A, 2017, PR MACH LEARN RES, V70
[7]   StarGAN v2: Diverse Image Synthesis for Multiple Domains [J].
Choi, Yunjey ;
Uh, Youngjung ;
Yoo, Jaejun ;
Ha, Jung-Woo .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020), 2020, :8185-8194
[8]  
Cover TM., 2006, ELEMENTS INFORM THEO, DOI [DOI 10.1002/047174882X, DOI 10.1002/0471200611.CH2]
[9]   On the Tchebychef inequality of Bernstein [J].
Craig, CC .
ANNALS OF MATHEMATICAL STATISTICS, 1933, 4 :94-102
[10]  
Csiszar Imre, 1982, Information theory: coding theorems for discrete memoryless systems