Adaptive Machine Unlearning

Authors: Varun Gupta, Christopher Jung, Seth Neel, Aaron Roth, Saeed Sharifi-Malvajerdi, Chris Waites

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

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 6, we complement our main result with a set of experimental results on CIFAR-10, MNIST, and Fashion-MNIST that demonstrate differential privacy may be useful in giving adaptive guarantees beyond the statement of our theorems.
Researcher Affiliation Academia Varun Gupta1, Christopher Jung1, Seth Neel2, Aaron Roth1, Saeed Sharifi-Malvajerdi1, and Chris Waites3 1University of Pennsylvania 2Harvard University 3Stanford University
Pseudocode Yes The paper contains "Algorithm 1: Interaction between (A, RA) and Upd Req", "Algorithm 2: Adistr: Distributed Learning Algorithm", and "Algorithm 3: RAdistr: Distributed Unlearning Algorithm: t th round of unlearning".
Open Source Code Yes The code for our experiments can be found at https://github.com/Chris Waites/adaptive-machine-unlearning.
Open Datasets Yes Experimental results on CIFAR-10 [Krizhevsky and Hinton, 2009], MNIST [Lecun et al., 1998], and Fashion-MNIST [Xiao et al., 2017]
Dataset Splits No The paper mentions using CIFAR-10, MNIST, and Fashion-MNIST but does not explicitly provide details about the train/validation/test splits, such as percentages, sample counts, or specific splitting methodology. It only refers to test sets in general.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions "JAX" and "DP-SGD" but does not provide specific version numbers for these or any other key software dependencies used in their experiments.
Experiment Setup Yes Full experimental details can be found in the appendix. (Appendix A. Experimental Details provides network architecture, batch size 128, ADAM optimizer, learning rate 1e-3, 50 epochs, and k shards for SISA framework.)