Disentangling Influence: Using disentangled representations to audit model predictions

Authors: Charles Marx, Richard Phillips, Sorelle Friedler, Carlos Scheidegger, Suresh Venkatasubramanian

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show through both theory and experiments that disentangled influence audits can both detect proxy features and show, for each individual or in aggregate, which of these proxy features affects the classifier being audited the most.
Researcher Affiliation Academia Charles T. Marx Haverford College cmarx@haverford.edu Richard Lanas Phillips Cornell University rlp246@cornell.edu Sorelle A. Friedler Haverford College sorelle@cs.haverford.edu Carlos Scheidegger University of Arizona cscheid@cs.arizona.edu Suresh Venkatasubramanian University of Utah suresh@cs.utah.edu
Pseudocode Yes DISENTANGLED-INFLUENCE-AUDIT(X, M) 1 for p in FEATURES(X) 2 (f , g, h) = DISENTANGLED-REPRESENTATION(X, p) // (h is not used) 3 M = g M 4 X = {f(x) for x in X} 5 SHAPp = DIRECT-INFLUENCE(X , p, M ) 6 return {SHAPp for p in FEATURES(X)}
Open Source Code Yes All data and code4 for the described method and below experiments is available in the Supplementary Materials. 4Code is available at: https://github.com/charliemarx/disentangling-influence
Open Datasets Yes It includes 5,000 instances of two variables x and y drawn independently from a uniform distribution over [0, 1]... The second synthetic dataset is the d Sprites dataset commonly used in the disentangled representations literature to disentangle independent factors that are sources of variation [23]... Finally, we will consider a real-world dataset containing Adult Income data that is commonly used as a test case in the fairness-aware machine learning community... [27].
Dataset Splits No The paper does not provide specific details on validation dataset splits, such as percentages or sample counts.
Hardware Specification Yes The Titan Xp used for this research was donated by the NVIDIA Corporation.
Software Dependencies No The paper mentions software like 'shap' and 'Black Box Auditing' (available via pip install) but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup No Preprocessing, model, and disentangled representation training information are included in the Supplementary Material.