Dimension-free Complexity Bounds for High-order Nonconvex Finite-sum Optimization

Dongruo Zhou, Quanquan Gu
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:27143-27158, 2022.

Abstract

Stochastic high-order methods for finding first-order stationary points in nonconvex finite-sum optimization have witnessed increasing interest in recent years, and various upper and lower bounds of the oracle complexity have been proved. However, under standard regularity assumptions, existing complexity bounds are all dimension-dependent (e.g., polylogarithmic dependence), which contrasts with the dimension-free complexity bounds for stochastic first-order methods and deterministic high-order methods. In this paper, we show that the polylogarithmic dimension dependence gap is not essential and can be closed. More specifically, we propose stochastic high-order algorithms with novel first-order and high-order derivative estimators, which can achieve dimension-free complexity bounds. With the access to $p$-th order derivatives of the objective function, we prove that our algorithm finds $\epsilon$-stationary points with $O(n^{(2p-1)/(2p)}/\epsilon^{(p+1)/p})$ high-order oracle complexities, where $n$ is the number of individual functions. Our result strictly improves the complexity bounds of existing high-order deterministic methods with respect to the dependence on $n$, and it is dimension-free compared with existing stochastic high-order methods.

Cite this Paper


BibTeX
@InProceedings{pmlr-v162-zhou22a, title = {Dimension-free Complexity Bounds for High-order Nonconvex Finite-sum Optimization}, author = {Zhou, Dongruo and Gu, Quanquan}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {27143--27158}, 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/zhou22a/zhou22a.pdf}, url = {https://proceedings.mlr.press/v162/zhou22a.html}, abstract = {Stochastic high-order methods for finding first-order stationary points in nonconvex finite-sum optimization have witnessed increasing interest in recent years, and various upper and lower bounds of the oracle complexity have been proved. However, under standard regularity assumptions, existing complexity bounds are all dimension-dependent (e.g., polylogarithmic dependence), which contrasts with the dimension-free complexity bounds for stochastic first-order methods and deterministic high-order methods. In this paper, we show that the polylogarithmic dimension dependence gap is not essential and can be closed. More specifically, we propose stochastic high-order algorithms with novel first-order and high-order derivative estimators, which can achieve dimension-free complexity bounds. With the access to $p$-th order derivatives of the objective function, we prove that our algorithm finds $\epsilon$-stationary points with $O(n^{(2p-1)/(2p)}/\epsilon^{(p+1)/p})$ high-order oracle complexities, where $n$ is the number of individual functions. Our result strictly improves the complexity bounds of existing high-order deterministic methods with respect to the dependence on $n$, and it is dimension-free compared with existing stochastic high-order methods.} }
Endnote
%0 Conference Paper %T Dimension-free Complexity Bounds for High-order Nonconvex Finite-sum Optimization %A Dongruo Zhou %A Quanquan Gu %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-zhou22a %I PMLR %P 27143--27158 %U https://proceedings.mlr.press/v162/zhou22a.html %V 162 %X Stochastic high-order methods for finding first-order stationary points in nonconvex finite-sum optimization have witnessed increasing interest in recent years, and various upper and lower bounds of the oracle complexity have been proved. However, under standard regularity assumptions, existing complexity bounds are all dimension-dependent (e.g., polylogarithmic dependence), which contrasts with the dimension-free complexity bounds for stochastic first-order methods and deterministic high-order methods. In this paper, we show that the polylogarithmic dimension dependence gap is not essential and can be closed. More specifically, we propose stochastic high-order algorithms with novel first-order and high-order derivative estimators, which can achieve dimension-free complexity bounds. With the access to $p$-th order derivatives of the objective function, we prove that our algorithm finds $\epsilon$-stationary points with $O(n^{(2p-1)/(2p)}/\epsilon^{(p+1)/p})$ high-order oracle complexities, where $n$ is the number of individual functions. Our result strictly improves the complexity bounds of existing high-order deterministic methods with respect to the dependence on $n$, and it is dimension-free compared with existing stochastic high-order methods.
APA
Zhou, D. & Gu, Q.. (2022). Dimension-free Complexity Bounds for High-order Nonconvex Finite-sum Optimization. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:27143-27158 Available from https://proceedings.mlr.press/v162/zhou22a.html.

Related Material