Learning Online Algorithms with Distributional Advice

We study the problem of designing online algorithms given advice about the input. While prior work had focused on deterministic advice, we only assume distributional access to the instances of interest, and the goal is to learn a competitive algorithm given access to i.i.d. samples. We aim to be competitive against an adversary with prior knowledge of the distribution, while also performing well against worst-case inputs. We focus on the classical online problems of skirental and prophet-inequalities, and provide sample complexity bounds for the underlying learning tasks. First, we point out that for general distributions it is information-theoretically impossible to beat the worst-case competitive-ratio with any finite sample size. As our main contribution, we establish strong positive results for wellbehaved distributions. Specifically, for the broad class of log-concave distributions, we show that poly(1/ ) samples suffice to obtain (1 + )competitive ratio. Finally, we show that this sample upper bound is close to best possible, even for very simple classes of distributions.

[1]  Sergei Vassilvitskii,et al.  Algorithms with predictions , 2020, Beyond the Worst-Case Analysis of Algorithms.

[2]  S. Matthew Weinberg,et al.  Matroid prophet inequalities , 2012, STOC '12.

[3]  Ronitt Rubinfeld,et al.  Online Page Migration with ML Advice , 2020, AISTATS.

[4]  Christoph Dürr,et al.  Online Bin Packing with Advice of Small Size , 2015, Theory of Computing Systems.

[5]  M. Bagnoli,et al.  Log-concave probability and its applications , 2004 .

[6]  S. Matthew Weinberg,et al.  Optimal Single-Choice Prophet Inequalities from Samples , 2019, ITCS.

[7]  Vahab S. Mirrokni,et al.  Online Allocation with Traffic Spikes: Mixing Adversarial and Stochastic Models , 2015, EC.

[8]  R. Pagh,et al.  Composable Sketches for Functions of Frequencies: Beyond the Worst Case , 2020, ICML.

[9]  Michael Mitzenmacher,et al.  A Model for Learned Bloom Filters and Optimizing by Sandwiching , 2018, NeurIPS.

[10]  L. L. Cam,et al.  Asymptotic Methods In Statistical Decision Theory , 1986 .

[11]  Antonios Antoniadis,et al.  Secretary and Online Matching Problems with Machine Learned Advice , 2020, NeurIPS.

[12]  Christoph Dürr,et al.  Online Computation with Untrusted Advice , 2019, ITCS.

[13]  J. Kiefer,et al.  Asymptotic Minimax Character of the Sample Distribution Function and of the Classical Multinomial Estimator , 1956 .

[14]  David P. Woodruff,et al.  Learning-Augmented Data Stream Algorithms , 2020, ICLR.

[15]  Benjamin Moseley,et al.  Learnable and Instance-Robust Predictions for Online Matching, Flows and Load Balancing , 2020, ESA.

[16]  Maria-Florina Balcan,et al.  Learning to Branch , 2018, ICML.

[17]  Tim Roughgarden,et al.  A PAC Approach to Application-Specific Algorithm Selection , 2015, SIAM J. Comput..

[18]  P RenaultMarc,et al.  On Online Algorithms with Advice for the k-Server Problem , 2015 .

[19]  Piotr Indyk,et al.  Learning-Based Frequency Estimation Algorithms , 2018, ICLR.

[20]  Santosh S. Vempala,et al.  The geometry of logconcave functions and sampling algorithms , 2007, Random Struct. Algorithms.

[21]  Google,et al.  Improving Online Algorithms via ML Predictions , 2024, NeurIPS.

[22]  Dhruv Rohatgi,et al.  Near-Optimal Bounds for Online Caching with Machine Learned Advice , 2019, SODA.

[23]  Debmalya Panigrahi,et al.  Customizing ML Predictions for Online Algorithms , 2020, ICML.

[24]  Prabhakar Raghavan,et al.  A Statistical Adversary for On-line Algorithms , 1991, On-Line Algorithms.

[25]  Le Song,et al.  2 Common Formulation for Greedy Algorithms on Graphs , 2018 .

[26]  Silvio Lattanzi,et al.  Online Scheduling via Learned Weights , 2020, SODA.

[27]  Rohan Kodialam,et al.  Optimal Algorithms for Ski Rental with Soft Machine-Learned Predictions , 2019, ArXiv.

[28]  Aditya Bhaskara,et al.  Online Learning with Imperfect Hints , 2020, ICML.

[29]  Edward F. Grove.  Online bin packing with lookahead , 1995, SODA '95.

[30]  Ola Svensson,et al.  The Primal-Dual method for Learning Augmented Algorithms , 2020, NeurIPS.

[31]  Tim Roughgarden.  Beyond the Worst-Case Analysis of Algorithms (Introduction) , 2020, ArXiv.

[32]  Sergei Vassilvitskii,et al.  Competitive caching with machine learned advice , 2018, ICML.

[33]  Amin Saberi,et al.  Online Optimization with Uncertain Information , 2012, ACM Trans. Algorithms.

[34]  Zhenwei Dai,et al.  Adaptive Learned Bloom Filter (Ada-BF): Efficient Utilization of the Classifier , 2019, NeurIPS.

[35]  Tim Kraska,et al.  Partitioned Learned Bloom Filter , 2020, ArXiv.

[36]  Eldar Fischer,et al.  On the power of conditional samples in distribution testing , 2013, ITCS '13.

[37]  Maria-Florina Balcan,et al.  Learning-Theoretic Foundations of Algorithm Configuration for Combinatorial Partitioning Problems , 2016, COLT.

[38]  Adi Rosén,et al.  On Online Algorithms with Advice for the k-Server Problem , 2011, Theory of Computing Systems.

[39]  Rocco A. Servedio,et al.  Testing probability distributions using conditional samples , 2012, Electron. Colloquium Comput. Complex..

[40]  Adam Tauman Kalai,et al.  Learning to Prune: Speeding up Repeated Computations , 2019, COLT.

[41]  Anna R. Karlin,et al.  Competitive randomized algorithms for non-uniform problems , 1990, SODA '90.

[42]  Tim Kraska,et al.  The Case for Learned Index Structures , 2018 .

[43]  E. Samuel-Cahn.  Comparison of Threshold Stop Rules and Maximum for Independent Nonnegative Random Variables , 1984 .

[44]  Christos H. Papadimitriou,et al.  Beyond competitive analysis [on-line algorithms] , 1994, Proceedings 35th Annual Symposium on Foundations of Computer Science.

[45]  Sreenivas Gollapudi,et al.  Online Algorithms for Rent-Or-Buy with Expert Advice , 2019, ICML.