adaptive goal recognition

15
5/10/2002 Adaptive Goal Adaptive Goal Recognition Recognition Neal Lesh Presented by Don Patterson

Upload: aggie

Post on 06-Jan-2016

19 views

Category:

Documents


0 download

DESCRIPTION

Adaptive Goal Recognition. Neal Lesh. Presented by Don Patterson. Goal Recognition. To infer a person’s intentions given a partial view of their actions Let A be a sequence of actions (  A*) Let G be a set of goals (  G*) A recognizer R : A*  ( G*  {nil}) - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Adaptive Goal Recognition

5/10/2002

Adaptive Goal Adaptive Goal RecognitionRecognition

Neal Lesh

Presented by Don Patterson

Page 2: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 25/10/2002

Goal RecognitionGoal Recognition

To infer a person’s intentions given a partial view of their actions

Let A be a sequence of actions ( A*)Let G be a set of goals ( G*)A recognizer R : A* ( G* {nil})

a program that takes an action sequence and predicts a goal or declines to make a prediction

Page 3: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 35/10/2002

Adaptive Goal RecognitionAdaptive Goal Recognition

Adds modifiers to RLet T* be a set of adaptationsA recognizer RT is a recognizer that has had a subset T of T* applied to it.

To personalize we will remember a different set, T, for each person

Page 4: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 45/10/2002

ExamplesExamples

A:(get pan,get egg,put butter in pan)

G*:(cook an egg, clean-up)

R(A): if (get egg) A )

then predict G = (cook an egg)else predict G = (clean-up)

Page 5: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 55/10/2002

ExamplesExamples

A: (get pan, get egg, put butter in pan)

G*:(cook an egg, clean-up)

T: (put butter in pan)RT(A):

if ((get egg) A ) or ((T A) {nil})then G = (cook an egg)else G = (clean-up)

Page 6: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 65/10/2002

MetricsMetrics

How are we going to compare our adapted recognizers?

Accuracy: How many times did RT(A) return the correct goal, G?Coverage: How many times did RT(A) return non-{nil}? How many times did it make a guess?Score: a function which unifies accuracy and coverage:

S(accuracy, coverage)

Page 7: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 75/10/2002

Adapt the goal recognizerAdapt the goal recognizer

Let D be the training dataConsists of a set E of “episodes”An episode is a start state and a sequence of actions (S,{a1,a2,a3,…,an})

Estimate(R,D)Returns Accuracy and Coverage of R on DRunning time

Verify the true goal |E| times, once per training example.Execute the Recognizer |E| * n. Once for every action in the training example.

R({a1}

R({a1,a2}) …

R({a1,a2,a3,…, an})

Page 8: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 85/10/2002

Adapt the goal recognizerAdapt the goal recognizer

Adapt(R,T,D)Greedily add adaptations to R until Estimate(RT,D) reaches a local maximum.

Running timeO(|T|2) calls to Estimate

Overall run time:O(|O(|T|T|22*[|*[|DD|*(O(Verify)+n*O(|*(O(Verify)+n*O(RRTT))])))])

O(|O(|T|T|22*|*|DD|*n*O(|*n*O(RRTT))))

Page 9: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 95/10/2002

ValidationValidation

Robustness to systematic non-goal oriented actions inserted into the training data:

Example: Every time I turn on the stove I open the door to check that nothing is in it.Example: Every time I type “cd” I also type “ls” regardless of what I’m doing.

Ability of the goal recognizer to helpHow often does the computer make a correct guess?

Impact of noisy training data

Page 10: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 105/10/2002

RobustnessRobustness

Page 11: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 115/10/2002

RobustnessRobustness

Page 12: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 125/10/2002

Ability to HelpAbility to Help

Page 13: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 135/10/2002

Impact of NoiseImpact of Noise

Frequency of Abandonment

0.0 0.05 0.10 0.15 0.20

Plan length 7.2 7.5 7.5 7.9 8.6% right 82 80 80 74 63% wrong 0 0 0 0 0% skipped 18 20 20 26 37

Page 14: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 145/10/2002

ObservationsObservations

Strengths:Recognizer independentProofs of trade-off between accuracy and coverageUnsupervised

Page 15: Adaptive Goal Recognition

590 HK Spring 2002: Adaptive Goal Recognition by Neal Lesh 155/10/2002

Challenges for Assisted Challenges for Assisted CognitionCognition

Adapt assumes discrete adaptationsProbabilities and continuous parameters are probably going to need to be handled.

Adaptations must be well chosenThis specific recognizer/adaptation pair wouldn’t appear to handle random noise wellIf actions are identified incorrectly then the system will likely fail.