submodularity in machine learning
play

Submodularity in Machine Learning - New Direc9ons - - PowerPoint PPT Presentation

Submodularity in Machine Learning - New Direc9ons - Andreas Krause Stefanie Jegelka Network Inference How learn who influences whom? 2


  1. Submodularity ¡ ¡ in ¡Machine ¡Learning ¡ -­‑ ¡New ¡Direc9ons ¡-­‑ ¡ Andreas ¡Krause ¡ Stefanie ¡Jegelka ¡

  2. Network ¡Inference ¡ How ¡learn ¡who ¡influences ¡whom? ¡ 2 ¡

  3. Summarizing ¡Documents ¡ How ¡select ¡representa9ve ¡sentences? ¡ 3 ¡

  4. MAP ¡inference ¡ sky ¡ tree ¡ house ¡ grass ¡ max p ( x | z ) x How ¡find ¡the ¡MAP ¡labeling ¡in ¡discrete ¡graphical ¡models ¡ ¡ efficiently ? ¡ 4 ¡

  5. What’s ¡common? ¡ � Formaliza9on: ¡ ¡Op9mize ¡a ¡set ¡func9on ¡F(S) ¡ ¡under ¡constraints ¡ ¡ ¡ � generally ¡very ¡hard ¡ ¡ � but: ¡structure ¡helps! ¡ ¡ ¡ ¡ ¡ ¡ … ¡if ¡F ¡is ¡submodular, ¡we ¡can ¡… ¡ � solve ¡op9miza9on ¡problems ¡with ¡strong ¡guarantees ¡ � solve ¡some ¡learning ¡problems ¡ 5 ¡

  6. Outline ¡ many ¡new ¡ � What ¡is ¡submodularity? ¡ results! ¡ J ¡ Part ¡I ¡ � Op9miza9on ¡ � Minimiza9on ¡ ¡ Break ¡ � Maximiza9on ¡ Part ¡II ¡ � Learning ¡ � Learning ¡for ¡Op9miza9on: ¡new ¡seXngs ¡ 6 ¡

  7. Outline ¡ many ¡new ¡ � What ¡is ¡submodularity? ¡ results! ¡ J ¡ Part ¡I ¡ � Op9miza9on ¡ � Minimiza9on: ¡new ¡algorithms, ¡constraints ¡ ¡ � Maximiza9on: ¡new ¡algorithms ¡(unconstrained) ¡ Part ¡II ¡ � Learning ¡ � Learning ¡for ¡Op9miza9on: ¡new ¡seXngs ¡ … ¡and ¡many ¡new ¡applica9ons! ¡ 7 ¡

  8. submodularity.org ¡ slides, ¡links, ¡references, ¡workshops, ¡… ¡ 8 ¡

  9. Example: ¡placing ¡sensors ¡ OF F ICE OF F ICE QUIE T PHONE CONF E RENCE STO R AGE LA B ELEC COPY SE R VER KITCHE N Place ¡sensors ¡to ¡monitor ¡temperature ¡ 9 ¡

  10. Set ¡func9ons ¡ � finite ¡ground ¡set ¡ V = { 1 , 2 , . . . , n } � set ¡func9on ¡ ¡ ¡ OF F ICE OF F ICE QUIE T PHONE F : 2 V → R CONF E RENCE STO R AGE LA B ELEC COPY ¡ SE R V ER KITCHE N ¡ � will ¡assume ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ (w.l.o.g.) ¡ F ( ∅ ) = 0 � assume ¡black ¡box ¡that ¡can ¡evaluate ¡ F ( A ) for ¡any ¡ ¡ A ⊆ V 10 ¡

  11. Example: ¡placing ¡sensors ¡ U9lity ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡of ¡having ¡sensors ¡at ¡subset ¡ ¡ ¡ ¡ ¡of ¡all ¡loca9ons ¡ F ( A ) A OF F ICE OF F ICE QUIE T PHONE OF F ICE OF F ICE QUIE T PHONE X 4 CONF E RENCE CONF E RENCE X 1 STO R AGE STO R AGE X 1 X 3 LA B LA B X 5 ELEC COPY ELEC COPY SE R VER SE R VER X 2 KITCHE N KITCHE N A={1,4,5}: ¡Redundant ¡info ¡ A={1,2,3}: ¡Very ¡informa9ve ¡ Low ¡value ¡F(A) ¡ High ¡value ¡F(A) ¡ 11 ¡

  12. Marginal ¡gain ¡ � Given ¡set ¡func9on ¡ F : 2 V → R ¡ ¡ ¡ � Marginal ¡gain: ¡ ∆ F ( s | A ) = F ( { s } ∪ A ) − F ( A ) ¡ OF F ICE OF F ICE QUIE T PHONE X 2 X 1 CONF E RENCE STO R AGE LA B ELEC COPY SE R VER KITCHE N X s ¡ ¡ ¡ new ¡sensor ¡s ¡ 12 ¡

  13. Decreasing ¡gains: ¡submodularity ¡ placement ¡A ¡= ¡{1,2} ¡ placement ¡B ¡= ¡{1,…,5} ¡ OF F ICE OF F ICE QUIE T PHONE OF F ICE OF F ICE QUIE T PHONE X 2 X 2 ¡ X 1 CONF E RENCE X 1 ¡ CONF E RENCE STO R AGE X 3 ¡ STO R AGE LA B LA B ELEC COPY ELEC COPY SE R VER SE R VER X 5 ¡ KITCHE N X 4 ¡ KITCHE N Big ¡gain ¡ Adding ¡ s ¡helps ¡a ¡lot! ¡ X s ¡ ¡ ¡ Adding ¡ s ¡doesn’t ¡help ¡much ¡ small ¡gain ¡ new ¡sensor ¡s ¡ A ¡ B ¡ + ¡ ¡ ¡ ¡ ¡ ¡ s ¡ + ¡ ¡ ¡ ¡ ¡ ¡ s ¡ A ⊆ B F ( A ∪ s ) − F ( A ) F ( B ∪ s ) − F ( B ) ≥ ∆ ( s | A ) ∆ ( s | B ) ≥ 13 ¡

  14. Equivalent ¡characteriza9ons ¡ � Diminishing ¡gains: ¡ ¡for ¡all ¡ A ⊆ B A ¡ B ¡ + ¡ ¡ ¡ ¡s ¡ + ¡ ¡ ¡ ¡s ¡ F ( A ∪ s ) − F ( A ) F ( B ∪ s ) − F ( B ) ≥ � Union-­‑Intersec9on: ¡for ¡all ¡ ¡ A, B ⊆ V ¡ A A A ∪ B B B F ( A ∪ B ) + F ( A ∩ B ) F ( A ) + F ( B ) ≥ A ∩ B 14 ¡

  15. Ques9ons ¡ How ¡do ¡I ¡prove ¡my ¡problem ¡is ¡ submodular? ¡ Why ¡is ¡submodularity ¡useful? ¡ 15 ¡

  16. Example: ¡Set ¡cover ¡ place sensors goal: cover floorplan with discs in building Possible OFFICE OFFICE QUIET PHONE locations CONFERENCE STO R AGE V LAB ELEC COPY SERVER KITCHEN ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡: ¡ ¡ F ( A ) = A ⊆ V Node ¡predicts ¡ “area ¡covered ¡by ¡sensors ¡placed ¡at ¡A” ¡ values ¡of ¡posi9ons ¡ with ¡some ¡radius ¡ Formally: ¡ ¡ Finite ¡set ¡ ¡ ¡ ¡ ¡, ¡collec9on ¡of ¡n ¡subsets ¡ W S i ⊆ W For ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡define ¡ F ( A ) = A ⊆ V � [ � � i ∈ A S i � 16 ¡

  17. Set ¡cover ¡is ¡submodular ¡ A={s 1 ,s 2 } OFFICE OFFICE QUIET PHONE CONFERENCE S 1 S 2 STO R AGE LAB ELEC COPY SERVER S ’ KITCHEN F(A U {s’}) – F(A) ≥ OFFICE OFFICE QUIET PHONE CONFERENCE F(B U {s ’ }) – F(B) S 1 S 2 STO R AGE LAB ELEC COPY S 3 SERVER S 4 KITCHEN S ’ B = {s 1 ,s 2 ,s 3 ,s 4 } 17 ¡

  18. More ¡complex ¡model ¡for ¡sensing ¡ OF F ICE OF F ICE QUIE T PHONE Y s : ¡temperature ¡ Y 1 ¡ Y 2 ¡ Y 3 ¡ CONF E RENCE at ¡loca9on ¡s ¡ STO R AGE X 1 ¡ X 3 ¡ LA B X 2 ¡ ELEC COPY X s : ¡sensor ¡value ¡ Y 6 ¡ at ¡loca9on ¡s ¡ SE R VER Y 4 ¡ KITCHE N Y 5 ¡ X s ¡= ¡Y s ¡+ ¡noise ¡ X 6 ¡ X 4 ¡ X 5 ¡ Joint ¡probability ¡distribu9on ¡ ¡ P(X 1 ,…,X n ,Y 1 ,…,Y n ) ¡ ¡= ¡P(Y 1 ,…,Y n ) ¡P(X 1 ,…,X n ¡| ¡Y 1 ,…,Y n ) ¡ Prior ¡ Likelihood ¡ 18 ¡

  19. Example: ¡Sensor ¡placement ¡ U9lity ¡of ¡having ¡sensors ¡at ¡subset ¡A ¡of ¡all ¡loca9ons ¡ ¡ F ( A ) = H ( Y ) − H ( Y | X A ) Uncertainty ¡ Uncertainty ¡ about ¡temperature ¡Y ¡ about ¡temperature ¡Y ¡ a7er ¡ sensing ¡ before ¡ sensing ¡ OF F ICE OF F ICE QUIE T PHONE OF F ICE OF F ICE QUIE T PHONE X 4 CONF E RENCE CONF E RENCE X 1 STO R AGE STO R AGE X 1 X 3 LA B LA B X 5 ELEC COPY ELEC COPY SE R VER SE R VER X 2 KITCHE N KITCHE N A={1,2,3}: ¡High ¡value ¡F(A) ¡ A={1,4,5}: ¡Low ¡value ¡F(A) ¡ 19 ¡

  20. Submodularity ¡of ¡Informa9on ¡Gain ¡ ¡Y 1 ,…,Y m , ¡X 1 , ¡…, ¡X n ¡discrete ¡RVs ¡ ¡F(A) ¡= ¡I(Y; ¡X A ) ¡= ¡H(Y)-­‑H(Y ¡| ¡X A ) ¡ � F(A) ¡is ¡ ¡NOT ¡always ¡submodular ¡ ¡ If ¡X i ¡are ¡all ¡condi9onally ¡independent ¡given ¡Y, ¡ then ¡F(A) ¡is ¡submodular! ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡[Krause ¡& ¡Guestrin ¡`05] ¡ Y 1 Y 2 Y 3 Proof: ¡ “informa9on ¡never ¡hurts” ¡ X 1 X 2 X 3 X 4 20 ¡

  21. Example: ¡costs ¡ cost: ¡ Market ¡3 ¡ 9me ¡to ¡reach ¡shop ¡ + ¡price ¡of ¡items ¡ breakfast?? ¡ t 3 ¡ ground ¡set ¡ ¡ ¡ ¡ ¡ V t 1 ¡ t 2 ¡ Market ¡1 ¡ Market ¡2 ¡ each ¡item ¡ 1 ¡$ ¡ 21 ¡

  22. Example: ¡costs ¡ cost: ¡ Market ¡3 ¡ 9me ¡to ¡shop ¡ + ¡price ¡of ¡items ¡ breakfast?? ¡ F( ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡) ¡= ¡ ¡cost( ¡ ¡ ¡ ¡ ¡ ¡) ¡+ ¡cost( ¡ ¡ ¡ ¡, ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡) ¡ ¡ = ¡ ¡t 1 ¡+ ¡1 ¡ ¡ ¡+ ¡ ¡ ¡ ¡t 2 ¡ ¡ ¡ ¡ + ¡2 ¡ ¡ = ¡ ¡#shops ¡ ¡+ ¡ ¡#items ¡ Market ¡1 ¡ Market ¡2 ¡ submodular? ¡ 22 ¡

  23. Shared ¡fixed ¡costs ¡ A ¡ ∆ ( b | A ) = 1 + t 3 B ¡ ∆ ( b | B ) = 1 marginal ¡cost: ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡#new ¡shops ¡ ¡+ ¡#new ¡items ¡ decreasing ¡ ¡ è ¡ ¡cost ¡is ¡submodular! ¡ shops: ¡ ¡ ¡ ¡ ¡shared ¡fixed ¡cost ¡ • economies ¡of ¡scale ¡ • 23 ¡

  24. Another ¡example: ¡Cut ¡func9ons ¡ 1 2 3 V={a,b,c,d,e,f,g,h} a c e g 2 3 2 2 2 3 3 3 X F ( A ) = w s,t b d f h 2 1 3 s ∈ A,t/ ∈ A Cut ¡func9on ¡is ¡submodular! ¡ 24 ¡

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend