Editing a classifier by rewriting its prediction rules

Authors: Shibani Santurkar, Dimitris Tsipras, Mahalaxmi Elango, David Bau, Antonio Torralba, Aleksander Madry

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

Reproducibility Variable Result LLM Response
Research Type Experimental We present a methodology for modifying the behavior of a classifier by directly rewriting its prediction rules. Our approach requires virtually no additional data collection and can be applied to a variety of settings, including adapting a model to new environments, and modifying it to ignore spurious features.
Researcher Affiliation Academia Shibani Santurkar MIT shibani@mit.edu Dimitris Tsipras MIT tsipras@mit.edu Mahalaxmi Elango MIT melango@mit.edu David Bau MIT davidbau@mit.edu Antonio Torralba MIT torralba@mit.edu Aleksander M adry MIT madry@mit.edu
Pseudocode No The paper describes a process but does not include a formally labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes 1Our code is available at https://github.com/Madry Lab/Editing Classifiers.
Open Datasets Yes Image Net [Deng et al., 2009, Russakovsky et al., 2015] and Places-365 [Zhou et al., 2017] datasets (cf. Appendix A.2). ... MS-COCO [Lin et al., 2014] and LVIS [Gupta et al., 2019]
Dataset Splits Yes All other transformed images containing the concept, including those from classes other than the target one, are used for validation and testing (30-70 split).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions models like VGG, ResNet, and CLIP, but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup No In each case, we select the best hyperparameters including the choice of the layer to modify based on the validation set performance (cf. Appendix A.6.3). ...Here, hyperparameters (cf. Appendix Table 2) are chosen based on the large-scale synthetic study in Section 4.1.