Universal Rate-Distortion-Perception Representations for Lossy Compression

Authors: George Zhang, Jingjing Qian, Jun Chen, Ashish Khisti

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide experimental results on MNIST and SVHN suggesting that on image compression tasks, the operational tradeoffs achieved by machine learning models with a fixed encoder suffer only a small penalty when compared to their variable encoder counterparts.
Researcher Affiliation Academia George Zhang Electrical and Computer Engineering University of Toronto gq.zhang@mail.utoronto.ca Jingjing Qian Electrical and Computer Engineering Mc Master University qianj40@mcmaster.ca Jun Chen Electrical and Computer Engineering Mc Master University chenjun@mcmaster.ca Ashish Khisti Electrical and Computer Engineering University of Toronto akhisti@ece.utoronto.ca
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper within the main text.
Open Datasets Yes We provide experimental results on MNIST and SVHN suggesting that on image compression tasks...
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning into train/validation/test sets.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes The architecture we use is a stochastic autoencoder with GAN regualarization, wherein a single model consists of an encoder f, a decoder g, and a critic h. The loss function is given by L = E[ X X 2] + λW1(p X, p X), where... The particular tradeoff point achieved by the model is controlled by the weight λ. ... Optimization alternates between minimizing over f, g with h fixed and maximizing over h with f, g fixed. Figure 4: λ1 = 0.0000 λ2 = 0.0050 λ3 = 0.0100 (MNIST) and λ1 = 0.0000 λ2 = 0.0005 λ3 = 0.0010 (SVHN).