Data-dependent PAC-Bayes priors via differential privacy

Authors: Gintare Karolina Dziugaite, Daniel M. Roy

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We study our datadependent bounds empirically, and show that they can be nonvacuous even when other distribution-dependent bounds are vacuous. We perform an experiment on a MNIST (and CIFAR10, with the same conclusion so we have not included it) using true and random labels and find that no bounds are violated. Our main focus is a sythentic experiment comparing the bounds of Lever, Laviolette, and Shawe Taylor (2013) to our new bounds based on privacy.
Researcher Affiliation Collaboration Gintare Karolina Dziugaite University of Cambridge; Element AI Daniel M. Roy University of Toronto; Vector Institute
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of its code.
Open Datasets Yes We perform an experiment on a MNIST (and CIFAR10, with the same conclusion so we have not included it) using true and random labels and find that no bounds are violated. We train two different architectures using SGLD on MNIST and a synthetic dataset, SYNTH.
Dataset Splits No The paper mentions using a "heldout test set" but does not provide specific percentages or counts for training, validation, or test splits. It refers to Appendix F for experimental setup, but the provided text does not contain these details.
Hardware Specification No The paper does not specify any hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., programming languages, libraries, frameworks).
Experiment Setup Yes The parameter settings for SYNTH experiments are: T1 = 100, T2 = 1000, g = 2; for MNIST: T1 = 500, T2 = 1000, g = 5. We run SGLD for T training epochs with a fixed value of the parameter t. We perform a two-stage training procedure: Stage One. We run SGLD for T1 epochs with inverse temperature t1... Stage Two. We continue SGLD for T2 T1 epochs with inverse temperature t2.