Faster Relative Entropy Coding with Greedy Rejection Coding

Authors: Gergely Flamich, Stratis Markou, José Miguel Hernández-Lobato

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we evaluate GRC in a variational autoencoderbased compression pipeline on MNIST, and show that a modified ELBO and an index-compression method can further improve compression efficiency.
Researcher Affiliation Academia Gergely Flamich Department of Engineering University of Cambridge gf332@cam.ac.uk Stratis Markou Department of Engineering University of Cambridge em626@cam.ac.uk José Miguel Hernández Lobato Department of Engineering University of Cambridge jmh233@cam.ac.uk
Pseudocode Yes Algorithm 1 Harsha et al. s rejection algorithm; equivalent to GRC with a global partition Algorithm 2 GRC with partition process Z; differences to Harsha et al. s algorithm shown in green
Open Source Code Yes Our code is available at https://github. com/cambridge-mlg/fast-rec-with-grc.
Open Datasets Yes Finally, we evaluate GRC in a variational autoencoderbased compression pipeline on MNIST
Dataset Splits No The paper mentions training on MNIST and evaluating on the MNIST test set, but it does not specify a separate validation split or its size.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies or libraries used.
Experiment Setup Yes We trained our VAE with L 2 {20, 50, 100} latent dimensions optimized using the negative ELBO and its modified version in Equation (12), and experimented with encoding the heap indices of GRCD with both δ and coding.