Compressing Latent Space via Least Volume

Authors: Qiuyi Chen, Mark Fuge

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the intuition behind the regularization on some pedagogical toy problems, and its effectiveness on several benchmark problems, including MNIST, CIFAR-10 and Celeb A.
Researcher Affiliation Academia Qiuyi Chen & Mark Fuge Department of Mechanical Engineering University of Maryland, College Park {qchen88,fuge}@umd.edu
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes We make the code public on Git Hub1 to ensure reproducibility. 1https://github.com/IDEALLab/Least_Volume_ICLR2024
Open Datasets Yes the MNIST dataset (Deng, 2012) and the CIFAR-10 dataset (Krizhevsky et al., 2014)... Celeb A dataset (Liu et al., 2015)
Dataset Splits No The paper mentions "three cross validations" but does not specify the validation dataset splits (e.g., percentages or counts) or reference predefined splits that include validation. Table B.3 only lists "Training Set Size" and "Test Set Size".
Hardware Specification Yes All experiments are performed on NVIDIA A100 SXM GPU 80GB.
Software Dependencies No The paper mentions software components like "Torchvision", "Adam" optimizer, and activation functions, but does not specify any version numbers for these or other key software dependencies (e.g., Python, PyTorch).
Experiment Setup Yes The hyperparameters are listed in Table B.2 [for toy problems] and Table B.5 [for image datasets]. ... Batch Size, λ, η, Learning Rate, Epochs.