Argumentation-based Distributed WAT-09, 9-11-2009, Seville Induction Santi Ontañón & Enric Plaza IIIA-CSIC dijous 12 de novembre de 2009 1
Outline • Motivation • Approach • Evaluation • Future dijous 12 de novembre de 2009 2
Motivation CONCEPT A1 DATA dijous 12 de novembre de 2009 3
Motivation CONCEPT CONCEPT A1 A2 DATA DATA dijous 12 de novembre de 2009 3
Motivation CONCEPT CONCEPT COMMUNICATION A1 A2 DATA DATA dijous 12 de novembre de 2009 3
Motivation CONCEPT CONCEPT COMMUNICATION A1 A2 DATA DATA dijous 12 de novembre de 2009 4
Motivation ALIGNEMENT CONCEPT CONCEPT COMMUNICATION A1 A2 DATA DATA dijous 12 de novembre de 2009 4
Motivation ALIGNEMENT CONCEPT CONCEPT ARGUMENTATION A1 A2 DATA DATA dijous 12 de novembre de 2009 4
Goals 1.Distributed induction 2.argumentation-based communication process 3.on top of existing ML methods • ID3 (decision trees) • CN2 (rule induction) • INDIE (relational inductive learning) dijous 12 de novembre de 2009 5
Induction p 1 ∧ p 2 ∧ p 3 − → C Hypothesis (an example is a concept C when rule is satisfed) dijous 12 de novembre de 2009 6
Induction p 1 ∧ p 2 ∧ p 3 − → C dijous 12 de novembre de 2009 7
Induction p 1 ∧ p 2 ∧ p 3 − → C p 3 ∧ p 4 − → C Hypothesis for C p 5 ∧ p 6 ∧ p 7 − → C = disjunction of rules dijous 12 de novembre de 2009 8
Induction with 2 agents p ′ 1 ∧ p ′ 2 ∧ p ′ → C p 1 ∧ p 2 ∧ p 3 − → C 3 − p ′ 3 ∧ p ′ p 3 ∧ p 4 − → C → C 4 − p ′ 5 ∧ p ′ 6 ∧ p ′ → C 7 − p 5 ∧ p 6 ∧ p 7 − → C dijous 12 de novembre de 2009 9
Agreement? p ′ 1 ∧ p ′ 2 ∧ p ′ → C p 1 ∧ p 2 ∧ p 3 − → C 3 − p ′ 3 ∧ p ′ → C p 3 ∧ p 4 − → C 4 − p ′ 5 ∧ p ′ 6 ∧ p ′ p 5 ∧ p 6 ∧ p 7 − → C → C 7 − dijous 12 de novembre de 2009 10
Agreement? p ′ 1 ∧ p ′ 2 ∧ p ′ → C p 1 ∧ p 2 ∧ p 3 − → C 3 − p ′ 3 ∧ p ′ → C p 3 ∧ p 4 − → C 4 − p ′ 5 ∧ p ′ 6 ∧ p ′ p 5 ∧ p 6 ∧ p 7 − → C → C 7 − dijous 12 de novembre de 2009 10
Agreement? p ′ 1 ∧ p ′ 2 ∧ p ′ → C p 1 ∧ p 2 ∧ p 3 − → C 3 − p ′ 3 ∧ p ′ → C p 3 ∧ p 4 − → C 4 − p ′ 5 ∧ p ′ 6 ∧ p ′ p 5 ∧ p 6 ∧ p 7 − → C → C 7 − dijous 12 de novembre de 2009 10
Agreement? p ′ 1 ∧ p ′ 2 ∧ p ′ → C p 1 ∧ p 2 ∧ p 3 − → C 3 − p ′ 3 ∧ p ′ → C p 3 ∧ p 4 − → C 4 − p ′ 5 ∧ p ′ 6 ∧ p ′ p 5 ∧ p 6 ∧ p 7 − → C → C 7 − dijous 12 de novembre de 2009 10
Approach dijous 12 de novembre de 2009 11
Argumentation • Argumentation as a process : • to reach an agreed concept between 2 agents • regulated interchange for contrasting, attacking, and revising beliefs • Working upon existing ML induction methods • ID3 • CN2 • INDIE dijous 12 de novembre de 2009 12
Argumentation e = � P, S � where ( S ∈ S ) Examples Hypotheses H = { r 1 , ..., r m } Rules r = � H, S � dijous 12 de novembre de 2009 13
Argumentation e = � P, S � where ( S ∈ S ) Examples Hypotheses H = { r 1 , ..., r m } Rules r = � H, S � dijous 12 de novembre de 2009 14
Argumentation e = � P, S � where ( S ∈ S ) Examples Hypotheses H = { r 1 , ..., r m } Rules r = � H, S � Argument α = � A, r � Counter-example β = � A, e, α � dijous 12 de novembre de 2009 14
ID3 rule conversion TL red green CC Wait no yes Cross Wait dijous 12 de novembre de 2009 15
CN2 post-process Post-processing removes order dependencies among rules used in CN2 C CN2 output: A B A default: C dijous 12 de novembre de 2009 16
Argument Solution Prediction ) Sponge SpiculateSkeleton Megascleres • Megascleres • SmoothForm Tylostyle • SpiculateSkeleton X • ExternalFeatures ExternalFeatures • Osc AbsentOsc α 1 dijous 12 de novembre de 2009 17
Argument evaluation The other’s arguments are contrasted with one’s examples ) Sponge SpiculateSkeleton Megascleres • Megascleres • SmoothForm Tylostyle • SpiculateSkeleton • ExternalFeatures ExternalFeatures • Osc AbsentOsc Case Base Finding Couter-examples of an argument dijous 12 de novembre de 2009 18
Argumentation-based ADI Distributed Induction Induction by individual agent T 1 T 2 using a specific ML method Induction Induction (ID3, CN2, INDIE) H 1 H 2 Argumentation about each rule held by an agent (first ARGUMENTATION one agent then the other) Hypotheses union eliminating redundancies dijous 12 de novembre de 2009 19
ADI argumentation C ( α 1 ) = { e 1 , e 2 } β = � A 2 , e 1 , α 1 � α 1 = � A 1 , r 1 � Belief revision: Agent A 1 incorporates counter-exmple e 1 and updates induction hypotheses dijous 12 de novembre de 2009 20
Reduced Argumentation- RADI based Distributed Induction Induction by individual agent T 1 T 2 using a specific ML method Induction Induction (ID3, CN2, INDIE) H 1 H 2 Argumentation about hypothesis of one agent ARGUMENTATION (then the other agent) Hypotheses union eliminating redundancies dijous 12 de novembre de 2009 21
Argumentation in RADI dijous 12 de novembre de 2009 22
Agreement H 1 H 2 E + E + 1 2 H 1 ⊑ E + H 2 ⊑ E + 1 2 dijous 12 de novembre de 2009 23
Agreement Semantically H 1 H 2 Equivalent E + E + 1 2 H 1 ⊑ E + H 2 ⊑ E + 1 2 H 1 ⊑ E + H 2 ⊑ E + 2 1 dijous 12 de novembre de 2009 23
Evaluation of ADI & RADI dijous 12 de novembre de 2009 24
Distribution Centralized Individual Union T 1 T 2 T 1 T 2 Induction Induction Induction Induction Induction H 1 H 2 H 1 H 2 dijous 12 de novembre de 2009 25
T 1 T 2 T 1 T 2 Induction Induction Induction Induction H 1 H 2 H 1 H 2 DAGGER DAGGER ARGUMENTATION Induction DAGGER ADI/RADI dijous 12 de novembre de 2009 26
Evaluation Accuracy Training Test Soybean Zoology Cars Sponges Soybean Zoology Cars Sponges ID3-centralized ID3-centralized 100,00 100,00 100,00 99,44 85,00 99,00 88,95 58,57 ID3-Individual ID3-Individual 85,67 93,85 93,84 80,20 76,50 90,00 86,83 55,54 ID3-union ID3-union 90,25 94,73 97,73 94,05 81,00 94,00 90,99 60,36 ID3-DAGGER ID3-DAGGER 99,57 100,00 76,36 99,76 80,67 92,50 68,95 62,50 ID3-ADI 100,00 100,00 100,00 99,70 88,50 99,00 88,95 58,21 ID3-RADI 100,00 100,00 100,00 99,74 87,67 99,00 89,24 58,21 Data set: 90% training; 10% test 2 Agents: 50% training set Best results in bold (when not statistically significant more than one results are in bold) dijous 12 de novembre de 2009 27
Evaluation Accuracy Training Test Soybean Zoology Cars Sponges Soybean Zoology Cars Sponges ID3-centralized ID3-centralized 100,00 100,00 100,00 99,44 85,00 99,00 88,95 58,57 ID3-Individual ID3-Individual 85,67 93,85 93,84 80,20 76,50 90,00 86,83 55,54 ID3-union ID3-union 90,25 94,73 97,73 94,05 81,00 94,00 90,99 60,36 ID3-DAGGER ID3-DAGGER 99,57 100,00 76,36 99,76 80,67 92,50 68,95 62,50 ID3-ADI 100,00 100,00 100,00 99,70 88,50 99,00 88,95 58,21 ID3-RADI 100,00 100,00 100,00 99,74 87,67 99,00 89,24 58,21 Training: ADI & RADI indistinguishable results from Centralized DAGGER good accuracy but not as Centralized dijous 12 de novembre de 2009 28
Evaluation Accuracy Training Test Soybean Zoology Cars Sponges Soybean Zoology Cars Sponges ID3-centralized ID3-centralized 100,00 100,00 100,00 99,44 85,00 99,00 88,95 58,57 ID3-Individual ID3-Individual 85,67 93,85 93,84 80,20 76,50 90,00 86,83 55,54 ID3-union ID3-union 90,25 94,73 97,73 94,05 81,00 94,00 90,99 60,36 ID3-DAGGER ID3-DAGGER 99,57 100,00 76,36 99,76 80,67 92,50 68,95 62,50 ID3-ADI 100,00 100,00 100,00 99,70 88,50 99,00 88,95 58,21 ID3-RADI 100,00 100,00 100,00 99,74 87,67 99,00 89,24 58,21 Test: ADI & RADI accuracy equal or better than Centralized DAGGER sometimes is better Union works very well only for Cars data set ADI & RADI are less prone to overfitting dijous 12 de novembre de 2009 29
Evaluation Accuracy Training Test Soybean Zoology Cars Sponges Soybean Zoology Cars Sponges CN2-centralized CN2-centralized 100,00 100,00 100,00 100,00 84,66 94,00 80,64 78,57 CN2-Individual CN2-Individual 87,82 94,62 89,90 88,29 77,83 87,50 80,84 74,46 CN2-union CN2-union 54,91 91,65 80,41 70,71 53,66 86,00 80,00 68,20 CN2-DAGGER CN2-DAGGER 99,49 99,65 95,86 99,88 79,33 92,50 75,34 78,93 CN2-ADI 100,00 100,00 100,00 100,00 84,90 93,50 80,61 79,11 CN2-RADI CN2-RADI 100,00 100,00 100,00 100,00 84,66 93,50 80,17 78,93 dijous 12 de novembre de 2009 30
Recommend
More recommend
