Minimum Entropy Coupling with Bottleneck

Authors: Reza Ebrahimi, Jun Chen, Ashish Khisti

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Furthermore, we illustrate the practical application of MEC-B through experiments in Markov Coding Games (MCGs) under rate limits. These games simulate a communication scenario within a Markov Decision Process, where an agent must transmit a compressed message from a sender to a receiver through its actions. Our experiments highlight the trade-offs between MDP rewards and receiver accuracy across various compression rates, showcasing the efficacy of our method compared to conventional compression baseline. This section presents the experimental results of the method described in Section 4.1, applied to Markov Coding Games.
Researcher Affiliation Academia M.Reza Ebrahimi University of Toronto mr.ebrahimi@mail.utoronto.ca Jun Chen Mc Master University chenjun@mcmaster.ca Ashish Khisti University of Toronto akhisti@ece.utoronto.ca
Pseudocode Yes Algorithm 1 Deterministic EBIM Solver Input: p X, R Output: p XT; Algorithm 2 Source; Algorithm 3 Agent; Algorithm 4 Receiver; Algorithm 5 Uniform Quantizer Encoder; Algorithm 6 Max-Seeking Minimum Entropy Coupling; Algorithm 7 Zero-Seeking Minimum Entropy Coupling; Algorithm 8 Soft Q-Value Iteration
Open Source Code Yes The submission includes the codes used to generate the results presented in the paper. Additionally, a README.md file accompanies these codes, offering detailed instructions on how to execute them, along with an example to guide users.
Open Datasets Yes For our experiments, we utilize a noisy Grid World environment for the Markov Decision Process. Figures 12 and 13 present sample output results after training the networks to achieve 4 -upscaling of the input images, using the MNIST [44] and SVHN [45] datasets.
Dataset Splits No The paper does not explicitly provide specific training, validation, and test dataset splits with percentages or sample counts for any of its experiments.
Hardware Specification No The paper states, 'The experiments performed in the paper are not computationally demanding. The runtime for the experiments on Markov Coding Games is mentioned in the Appendix Section C.2.' However, Appendix C.2 (Environment Setup) describes the Grid World environment but does not provide any specific hardware details such as GPU or CPU models, or memory specifications.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions, or other libraries).
Experiment Setup Yes The marginal policy is learned through Soft Q-Value iteration, as described in Algorithm 8. By increasing the value of β in Equation (14), we induce more randomness into the marginal policy. The rewards received are discounted by a factor of 0.95. For our experiments, we utilize a noisy Grid World environment for the Markov Decision Process.