Learning piecewise Lipschitz functions in changing environments

Dravyansh Sharma, Maria-Florina Balcan, Travis Dick
Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, PMLR 108:3567-3577, 2020.

Abstract

Optimization in the presence of sharp (non-Lipschitz), unpredictable (w.r.t. time and amount) changes is a challenging and largely unexplored problem of great significance. We consider the class of piecewise Lipschitz functions, which is the most general online setting considered in the literature for the problem, and arises naturally in various combinatorial algorithm selection problems where utility functions can have sharp discontinuities. The usual performance metric of ‘static’ regret minimizes the gap between the payoff accumulated and that of the best fixed point for the entire duration, and thus fails to capture changing environments. Shifting regret is a useful alternative, which allows for up to $s$ environment {\it shifts}. In this work we provide an $O(\sqrt{sdT\log T}+sT^{1-\beta})$ regret bound for $\beta$-dispersed functions, where $\beta$ roughly quantifies the rate at which discontinuities appear in the utility functions in expectation (typically $\beta\ge1/2$ in problems of practical interest \cite{2019arXiv190409014B,balcan2018dispersion}). We also present a lower bound tight up to sub-logarithmic factors. We further obtain improved bounds when selecting from a small pool of experts. We empirically demonstrate a key application of our algorithms to online clustering problems on popular benchmarks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v108-sharma20a, title = {Learning piecewise Lipschitz functions in changing environments}, author = {Sharma, Dravyansh and Balcan, Maria-Florina and Dick, Travis}, booktitle = {Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics}, pages = {3567--3577}, year = {2020}, editor = {Chiappa, Silvia and Calandra, Roberto}, volume = {108}, series = {Proceedings of Machine Learning Research}, month = {26--28 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v108/sharma20a/sharma20a.pdf}, url = {https://proceedings.mlr.press/v108/sharma20a.html}, abstract = {Optimization in the presence of sharp (non-Lipschitz), unpredictable (w.r.t. time and amount) changes is a challenging and largely unexplored problem of great significance. We consider the class of piecewise Lipschitz functions, which is the most general online setting considered in the literature for the problem, and arises naturally in various combinatorial algorithm selection problems where utility functions can have sharp discontinuities. The usual performance metric of ‘static’ regret minimizes the gap between the payoff accumulated and that of the best fixed point for the entire duration, and thus fails to capture changing environments. Shifting regret is a useful alternative, which allows for up to $s$ environment {\it shifts}. In this work we provide an $O(\sqrt{sdT\log T}+sT^{1-\beta})$ regret bound for $\beta$-dispersed functions, where $\beta$ roughly quantifies the rate at which discontinuities appear in the utility functions in expectation (typically $\beta\ge1/2$ in problems of practical interest \cite{2019arXiv190409014B,balcan2018dispersion}). We also present a lower bound tight up to sub-logarithmic factors. We further obtain improved bounds when selecting from a small pool of experts. We empirically demonstrate a key application of our algorithms to online clustering problems on popular benchmarks.} }
Endnote
%0 Conference Paper %T Learning piecewise Lipschitz functions in changing environments %A Dravyansh Sharma %A Maria-Florina Balcan %A Travis Dick %B Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2020 %E Silvia Chiappa %E Roberto Calandra %F pmlr-v108-sharma20a %I PMLR %P 3567--3577 %U https://proceedings.mlr.press/v108/sharma20a.html %V 108 %X Optimization in the presence of sharp (non-Lipschitz), unpredictable (w.r.t. time and amount) changes is a challenging and largely unexplored problem of great significance. We consider the class of piecewise Lipschitz functions, which is the most general online setting considered in the literature for the problem, and arises naturally in various combinatorial algorithm selection problems where utility functions can have sharp discontinuities. The usual performance metric of ‘static’ regret minimizes the gap between the payoff accumulated and that of the best fixed point for the entire duration, and thus fails to capture changing environments. Shifting regret is a useful alternative, which allows for up to $s$ environment {\it shifts}. In this work we provide an $O(\sqrt{sdT\log T}+sT^{1-\beta})$ regret bound for $\beta$-dispersed functions, where $\beta$ roughly quantifies the rate at which discontinuities appear in the utility functions in expectation (typically $\beta\ge1/2$ in problems of practical interest \cite{2019arXiv190409014B,balcan2018dispersion}). We also present a lower bound tight up to sub-logarithmic factors. We further obtain improved bounds when selecting from a small pool of experts. We empirically demonstrate a key application of our algorithms to online clustering problems on popular benchmarks.
APA
Sharma, D., Balcan, M. & Dick, T.. (2020). Learning piecewise Lipschitz functions in changing environments. Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 108:3567-3577 Available from https://proceedings.mlr.press/v108/sharma20a.html.

Related Material