learning augmented online selection algorithms
play

Learning-Augmented Online Selection Algorithms Themis Gouleakis - PowerPoint PPT Presentation

Learning-Augmented Online Selection Algorithms Themis Gouleakis Joint work with: Antonios Antoniadis, Pieter Kleer and Pavel Kolev September 2, 2020 Introduction Secretary problem with prediction Online bipartite matching The end Online


  1. Learning-Augmented Online Selection Algorithms Themis Gouleakis Joint work with: Antonios Antoniadis, Pieter Kleer and Pavel Kolev September 2, 2020

  2. Introduction Secretary problem with prediction Online bipartite matching The end Online selection problem Elements E = { e 1 , . . . , e n } arrive online. – Uniformly random arrival order σ of elements in E Element e i has value v i ≥ 0 (revealed upon arrival). Upon arrival of element e i : Select or reject it (irrevocably). Goal: Select feasible set S of elements that maximizes � f ( S ) = v j . j ∈ S Focus is on (constant-factor) approximation algorithms. Examples: Online (bipartite) matching, Matroid secretary problem. September 2, 2020 2/20

  3. Introduction Secretary problem with prediction Online bipartite matching The end Learning augmentation Machine learning oracle predicts aspect of – Input that has not yet arrived. – (Offline) optimal solution. We do not know quality of prediction. – Measured in terms of prediction error η . Goal: Include predictions in existing α -approximation such that: Improved approximation guarantee if η is small. Minor loss in approximate guarantee if η is large. “Best of both worlds”-scenario: Improved guarantees if ML oracle is accurate. Still guarantee in worst-case when oracle is inaccurate. September 2, 2020 3/20

  4. Introduction Secretary problem with prediction Online bipartite matching The end Approx. guarantee 1 α 0 Prediction error η September 2, 2020 4/20

  5. Introduction Secretary problem with prediction Online bipartite matching The end (Some) related work Machine learned advice: Ski rental – [Purohit-Svitkina-Kumar, NIPS 2018], [Wang-Wang, 2020]. Scheduling – [Purohit-Svitkina-Kumar, NIPS 2018], [Mitzenmacher, 2019], [Lattanzi-Lavastida-Moseley-Vassilvitskii, SODA 2020]. Caching – [Lykouris-Vassilvitskii, ICML 2018], [Rothagi, SODA 2020]. Metric Algorithms – [Antoniadis-Coester-Eliás-Polak-Simon, ICML 2020]. Online selection problems with distributional information: v i ∼ F i . Prophet inequalities (adversarial arrival order) – Single item: [Krengel-Sucheston, 1978]. – Matroid prophet inequality: [Kleinberg-Weinberg, 2012]. – Unknown distribution: e.g., [Correa-Dütting-Fischer-Schewior, ’19]. September 2, 2020 5/20

  6. Introduction Secretary problem with prediction Online bipartite matching The end Secretary problem Elements (secretaries) { e 1 , . . . , e n } arrive over time. – Uniform random arrival order σ = ( e 1 , . . . , e n ) . Value v i revealed upon arrival of e i . Goal: Select secretary with maximum value v ∗ = max i v i . Secretary algorithm [Lindley, 1961]/[Dynkin, 1963] Phase I: For i = 1 , . . . , n e : Select nothing. Phase II: Set threshold t = max j = 1 ,..., n e v j . For i = n e + 1 , . . . , n : If v i > t , select e i and STOP . Gives 1 v ] ≥ 1 e -approximation for maximum value v ∗ , i.e., E σ [¯ e · v ∗ . September 2, 2020 6/20

  7. Introduction Secretary problem with prediction Online bipartite matching The end Example Value 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  8. Introduction Secretary problem with prediction Online bipartite matching The end Example Value 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  9. Introduction Secretary problem with prediction Online bipartite matching The end Example Value 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  10. Introduction Secretary problem with prediction Online bipartite matching The end Example Value 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  11. Introduction Secretary problem with prediction Online bipartite matching The end Example Value 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  12. Introduction Secretary problem with prediction Online bipartite matching The end Example Value t 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  13. Introduction Secretary problem with prediction Online bipartite matching The end Example Value t 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  14. Introduction Secretary problem with prediction Online bipartite matching The end Example Value t 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  15. Introduction Secretary problem with prediction Online bipartite matching The end Example Value t 0 i Phase I Phase II i = 1 , . . . , n i = n e + 1 , . . . , n e September 2, 2020 7/20

  16. Introduction Secretary problem with prediction Online bipartite matching The end Prediction We include prediction p ∗ for optimal value v ∗ . Prediction error η = | p ∗ − v ∗ | . Goal (informal): Design (deterministic) algorithm such that: Approximation guarantee > 1 e when η is small. Approximation guarantee ≈ 1 ce when η is large. – For some constant c > 1. 1 1 e 1 ce 0 Prediction error η September 2, 2020 8/20

  17. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? September 2, 2020 9/20

  18. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: September 2, 2020 9/20

  19. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . September 2, 2020 9/20

  20. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . September 2, 2020 9/20

  21. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  22. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  23. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  24. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  25. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  26. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  27. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  28. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

  29. Introduction Secretary problem with prediction Online bipartite matching The end What to do when prediction is good? Choose element with value ‘close’ to prediction: Fix λ > 0, and select first element with v i > p ∗ − λ . Parameter λ can be seen as estimator for η . Value p ∗ p ∗ − λ 0 i September 2, 2020 9/20

Recommend


More recommend