Variational Bayesian Unlearning

Authors: Quoc Phong Nguyen, Bryan Kian Hsiang Low, Patrick Jaillet

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically demonstrate our unlearning methods on Bayesian models such as sparse Gaussian process and logistic regression using synthetic and real-world datasets.
Researcher Affiliation Academia Dept. of Computer Science, National University of Singapore, Republic of Singapore Dept. of Electrical Engineering and Computer Science, MIT, USA
Pseudocode No The paper describes algorithms and methods in prose and mathematical formulations but does not include explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes We empirically demonstrate our unlearning methods on Bayesian models such as sparse Gaussian process and logistic regression using synthetic and real-world datasets. Further experimental results on Bayesian linear regression and with a bimodal posterior belief are reported in Appendices C and D, respectively. ... The banknote authentication dataset [10] of size |D| = 1372 is partitioned into erased data of size |De| = 412 and remaining data of size |Dr| = 960. ... The fashion MNIST dataset of size |D| = 60000 (28 28 images of fashion items in 10 classes) is partitioned into erased data of size |De| = 10000 and remaining data of size |Dr| = 50000. ... This section illustrates the scalability of unlearning to the massive airline dataset of 2 million flights [15].
Dataset Splits Yes The banknote authentication dataset [10] of size |D| = 1372 is partitioned into erased data of size |De| = 412 and remaining data of size |Dr| = 960. ... The fashion MNIST dataset of size |D| = 60000 (28 28 images of fashion items in 10 classes) is partitioned into erased data of size |De| = 10000 and remaining data of size |Dr| = 50000.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software components like RMSProp, normalizing flows with MADE architecture, and Gaussians, but does not provide specific version numbers for these or other libraries.
Experiment Setup Yes We use RMSProp as the SGA algorithm with a learning rate of 10 4. ... We initialize qu(θ|Dr; λ) at q(θ|D) for achieving empirically faster convergence.