A Model-free Closeness-of-influence Test for Features in Supervised Learning

Authors: Mohammad Mehrabi, Ryan A. Rossi

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our findings through extensive numerical simulations, specifically we adopt the datamodel framework (Ilyas, et al., 2022) for CIFAR-10 dataset to identify pairs of training samples with different influence on the trained model via optional black box training mechanisms.
Researcher Affiliation Collaboration 1Department of Data Sciences and Operations, University of Southern California, Los Angeles, USA 2Adobe, San Jose, USA.
Pseudocode Yes Algorithm 1 Test statistic for hypothesis testing 1
Open Source Code No The paper provides a link to
Open Datasets Yes We consider the CIFAR-10 dataset (Krizhevsky et al., 2009)
Dataset Splits No The paper mentions training and testing samples but does not explicitly describe a validation split for model training or evaluation. It mentions
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments. It vaguely mentions
Software Dependencies No The paper mentions using
Experiment Setup Yes We run our method with the score function T(x, y) = |y bθTx| with bθ N(0, Id). The estimate bθ is fixed across all 45 tests. We suppose that we have access to 1000 data points, and we consider three different significance levels α = 0.1, 0.15, and 0.2.