Stability Based Generalization Bounds for Exponential Family Langevin Dynamics

Arindam Banerjee, Tiancong Chen, Xinyan Li, Yingxue Zhou
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:1412-1449, 2022.

Abstract

Recent years have seen advances in generalization bounds for noisy stochastic algorithms, especially stochastic gradient Langevin dynamics (SGLD) based on stability (Mou et al., 2018; Li et al., 2020) and information theoretic approaches (Xu & Raginsky, 2017; Negrea et al., 2019; Steinke & Zakynthinou, 2020). In this paper, we unify and substantially generalize stability based generalization bounds and make three technical contributions. First, we bound the generalization error in terms of expected (not uniform) stability which arguably leads to quantitatively sharper bounds. Second, as our main contribution, we introduce Exponential Family Langevin Dynamics (EFLD), a substantial generalization of SGLD, which includes noisy versions of Sign-SGD and quantized SGD as special cases. We establish data dependent expected stability based generalization bounds for any EFLD algorithm with a O(1/n) sample dependence and dependence on gradient discrepancy rather than the norm of gradients, yielding significantly sharper bounds. Third, we establish optimization guarantees for special cases of EFLD. Further, empirical results on benchmarks illustrate that our bounds are non-vacuous, quantitatively sharper than existing bounds, and behave correctly under noisy labels.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-banerjee22a, title = {Stability Based Generalization Bounds for Exponential Family {L}angevin Dynamics}, author = {Banerjee, Arindam and Chen, Tiancong and Li, Xinyan and Zhou, Yingxue}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {1412--1449}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/banerjee22a/banerjee22a.pdf}, url = {https://proceedings.mlr.press/v162/banerjee22a.html}, abstract = {Recent years have seen advances in generalization bounds for noisy stochastic algorithms, especially stochastic gradient Langevin dynamics (SGLD) based on stability (Mou et al., 2018; Li et al., 2020) and information theoretic approaches (Xu & Raginsky, 2017; Negrea et al., 2019; Steinke & Zakynthinou, 2020). In this paper, we unify and substantially generalize stability based generalization bounds and make three technical contributions. First, we bound the generalization error in terms of expected (not uniform) stability which arguably leads to quantitatively sharper bounds. Second, as our main contribution, we introduce Exponential Family Langevin Dynamics (EFLD), a substantial generalization of SGLD, which includes noisy versions of Sign-SGD and quantized SGD as special cases. We establish data dependent expected stability based generalization bounds for any EFLD algorithm with a O(1/n) sample dependence and dependence on gradient discrepancy rather than the norm of gradients, yielding significantly sharper bounds. Third, we establish optimization guarantees for special cases of EFLD. Further, empirical results on benchmarks illustrate that our bounds are non-vacuous, quantitatively sharper than existing bounds, and behave correctly under noisy labels.} }
Endnote
%0 Conference Paper %T Stability Based Generalization Bounds for Exponential Family Langevin Dynamics %A Arindam Banerjee %A Tiancong Chen %A Xinyan Li %A Yingxue Zhou %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-banerjee22a %I PMLR %P 1412--1449 %U https://proceedings.mlr.press/v162/banerjee22a.html %V 162 %X Recent years have seen advances in generalization bounds for noisy stochastic algorithms, especially stochastic gradient Langevin dynamics (SGLD) based on stability (Mou et al., 2018; Li et al., 2020) and information theoretic approaches (Xu & Raginsky, 2017; Negrea et al., 2019; Steinke & Zakynthinou, 2020). In this paper, we unify and substantially generalize stability based generalization bounds and make three technical contributions. First, we bound the generalization error in terms of expected (not uniform) stability which arguably leads to quantitatively sharper bounds. Second, as our main contribution, we introduce Exponential Family Langevin Dynamics (EFLD), a substantial generalization of SGLD, which includes noisy versions of Sign-SGD and quantized SGD as special cases. We establish data dependent expected stability based generalization bounds for any EFLD algorithm with a O(1/n) sample dependence and dependence on gradient discrepancy rather than the norm of gradients, yielding significantly sharper bounds. Third, we establish optimization guarantees for special cases of EFLD. Further, empirical results on benchmarks illustrate that our bounds are non-vacuous, quantitatively sharper than existing bounds, and behave correctly under noisy labels.
APA
Banerjee, A., Chen, T., Li, X. & Zhou, Y.. (2022). Stability Based Generalization Bounds for Exponential Family Langevin Dynamics. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:1412-1449 Available from https://proceedings.mlr.press/v162/banerjee22a.html.

Related Material