Sliced Mutual Information: A Scalable Measure of Statistical Dependence

Authors: Ziv Goldfeld, Kristjan Greenewald

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our theory is supported by numerical studies of independence testing and feature extraction, which demonstrate the potential gains SMI offers over classic MI for high-dimensional inference. ... We validate our theory on synthetic experiments, demonstrating that SMI is a scalable alternative to classic MI when dealing with high-dimensional data. ... 4 Empirical Results
Researcher Affiliation Collaboration Ziv Goldfeld Cornell University goldfeld@cornell.edu Kristjan Greenewald MIT-IBM Watson AI Lab kristjan.h.greenewald@ibm.com
Pseudocode Yes Pseudocode and computational complexity for b SIn,m can be found in Section B of the supplement.
Open Source Code No The paper does not provide an explicit statement about releasing code for the methodology or a link to a code repository.
Open Datasets Yes Figure 4: Solution for optimizing A-transformed SMI of the 0-1 MNIST setup using S-MINE. Rows 0 and 1 of A are shown in (a) and (b), respectively. We next combine S-MINE with independence testing, looking to maximize the SMI using transformations AX, AY , where X, Y are samples from a random MNIST class (either 0 or 1) and A 2 R10 d.
Dataset Splits No The paper describes generating '50 datasets comprising n positive samples' and '50 more dataset of negative samples' for independence testing, and uses Monte-Carlo sampling. However, it does not specify explicit train/validation/test splits with percentages or counts for model training or evaluation in a reproducible manner.
Hardware Specification No The paper does not explicitly describe the hardware (e.g., specific GPU/CPU models, cloud resources with specifications) used to run its experiments.
Software Dependencies No The paper mentions using the 'Kozachenko Leonenko estimator', 'MINE', and 'S-MINE', but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes The MC step for SMI estimation (see (6)) uses 1000 random slices... with g in S-MINE realized as a two-layer fully connected neural network with 100 hidden units for.