Data-Copying in Generative Models: A Formal Framework

Robi Bhattacharjee, Sanjoy Dasgupta, Kamalika Chaudhuri
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:2364-2396, 2023.

Abstract

There has been some recent interest in detecting and addressing memorization of training data by deep neural networks. A formal framework for memorization in generative models, called “data-copying” was proposed by Meehan et. al (2020). We build upon their work to show that their framework may fail to detect certain kinds of blatant memorization. Motivated by this and the theory of non-parametric methods, we provide an alternative definition of data-copying that applies more locally. We provide a method to detect data-copying, and provably show that it works with high probability when enough data is available. We also provide lower bounds that characterize the sample requirement for reliable detection.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-bhattacharjee23a, title = {Data-Copying in Generative Models: A Formal Framework}, author = {Bhattacharjee, Robi and Dasgupta, Sanjoy and Chaudhuri, Kamalika}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {2364--2396}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/bhattacharjee23a/bhattacharjee23a.pdf}, url = {https://proceedings.mlr.press/v202/bhattacharjee23a.html}, abstract = {There has been some recent interest in detecting and addressing memorization of training data by deep neural networks. A formal framework for memorization in generative models, called “data-copying” was proposed by Meehan et. al (2020). We build upon their work to show that their framework may fail to detect certain kinds of blatant memorization. Motivated by this and the theory of non-parametric methods, we provide an alternative definition of data-copying that applies more locally. We provide a method to detect data-copying, and provably show that it works with high probability when enough data is available. We also provide lower bounds that characterize the sample requirement for reliable detection.} }
Endnote
%0 Conference Paper %T Data-Copying in Generative Models: A Formal Framework %A Robi Bhattacharjee %A Sanjoy Dasgupta %A Kamalika Chaudhuri %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-bhattacharjee23a %I PMLR %P 2364--2396 %U https://proceedings.mlr.press/v202/bhattacharjee23a.html %V 202 %X There has been some recent interest in detecting and addressing memorization of training data by deep neural networks. A formal framework for memorization in generative models, called “data-copying” was proposed by Meehan et. al (2020). We build upon their work to show that their framework may fail to detect certain kinds of blatant memorization. Motivated by this and the theory of non-parametric methods, we provide an alternative definition of data-copying that applies more locally. We provide a method to detect data-copying, and provably show that it works with high probability when enough data is available. We also provide lower bounds that characterize the sample requirement for reliable detection.
APA
Bhattacharjee, R., Dasgupta, S. & Chaudhuri, K.. (2023). Data-Copying in Generative Models: A Formal Framework. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:2364-2396 Available from https://proceedings.mlr.press/v202/bhattacharjee23a.html.

Related Material