《构建强健的人工智能:原因及方式(54页).pdf》由会员分享,可在线阅读,更多相关《构建强健的人工智能:原因及方式(54页).pdf(54页珍藏版)》请在三个皮匠报告上搜索。
1、ROBUST ARTIFICIAL INTELLIGENCE:WHY AND HOW Tom Dietterich Distinguished Professor(Emeritus)Oregon State University Past-President AAAI 1 Outline The Need for Robust AI High Stakes Applications Need to Act in the face of Unknown Unknowns Approaches toward Robust AI Robustness to Known Unknowns Robust
2、ness to Unknown Unknowns Concluding Remarks 2 CCAI-2017 Technical Progress is Encouraging the Development of High-Stakes Applications 3 CCAI-2017 Self-Driving Cars Credit:The Verge Tesla AutoSteer Credit:Tesla Motors Credit: 4 CCAI-2017 Automated Surgical Assistants 5 Credit:Wikipedia CC BY-SA 3.0 D
3、aVinci CCAI-2017 AI Hedge Funds 6 CCAI-2017 AI Control of the Power Grid 7 Credit:DARPA Credit:EBM Netz AG CCAI-2017 Autonomous Weapons 8 Samsung SGR-1 Credit:AFP/Getty Images CCAI-2017 Northroop Grumman X-47B Credit:Wikipedia UK Brimstone Anti-Armor Weapon Credit:Duch.seb-Own work,CC BY-SA 3.0 High
4、-Stakes Applications Require Robust AI Robustness to Human user error Cyberattack Misspecified goals Incorrect models Unmodeled phenomena 9 CCAI-2017 Why Unmodeled Phenoma?It is impossible to model everything It is not desirable to model everything 10 CCAI-2017 It is impossible to model everything Q
5、ualification Problem:It is impossible to enumerate all of the preconditions for an action Ramification Problem:It is impossible to enumerate all of the implicit consequences of an action 11 CCAI-2017 It is important to not model everything Fundamental theorem of machine learning error rate model com
6、plexitysample size Corollary:If sample size is small,the model should be simple We must deliberately oversimplify our models!12 CCAI-2017 Conclusion:An AI system must act without having a complete model of the world 13 CCAI-2017 Outline The Need for Robust AI High Stakes Applications Need to Act in
7、the face of Unknown Unknowns Approaches toward Robust AI Lessons from Biology Robustness to Known Unknowns Robustness to Unknown Unknowns Concluding Remarks 14 CCAI-2017 Robustness Lessons from Biology Evolution is not optimization You cant overfit if you dont optimize Competition against adversarie
8、s“Survival of the Fittest”Populations of diverse individuals A“portfolio”strategy Redundancy within individuals diploidy/polyploidy=recessive alleles can be passed to future generations alternative metabolic pathways Dispersal Search for healthier environments 15 CCAI-2017 Approaches to Robust AI Ro
9、bustness to Model Errors Probabilistic Methods Robust optimization Regularize the model Optimize a risk-sensitive objective Employ robust inference algorithms Robustness to Unmodeled Phenomena Detect model weaknesses (including anomaly detection)Use a big model Learn a causal model Employ a portfoli
10、o of models 16 CCAI-2017 Idea 1:Decision Making under Uncertainty Observe Choose to maximize ,Uncertainty modeled as(|,)“Maximize Expected Utility”CCAI-2017 17 A Robustness to Downside Risk ,ignores the distribution of ,In this case ,1=,2 But action 2 has larger down-side risk and larger variance Ri
11、sk-sensitive measures will prefer 1 CCAI-2017 18 Utility P(U|Y,A)Idea 2:Robust Optimization Many AI reasoning problems can be formulated as optimization problems max1,2 (1,2)subject to 1+2 1+2 19 CCAI-2017 1,2 1 2 Uncertainty in the constraints max1,2 (1,2)subject to 1+2 1+2 Define uncertainty regio
12、ns 20 CCAI-2017 1,2 1 2 Minimax against the uncertainty max1,2min,(1,2;,)subject to 1+2 1+2 Problem:Solutions can be too conservative 21 CCAI-2017 Impose a Budget on the Adversary max1,2min,(1,2;,)subject to(+)1+(+)2 +(+)1+2 +22 Bertsimas,et al.CCAI-2017 Existing AI Algorithms Implicitly Implement R
13、obust Optimization Given:training examples(,)for an unknown function =()a loss function ,:how serious it is to output when the right answer is?Find:the model that minimizes ,+loss +complexity penalty CCAI-2017 23 Regularization can be Equivalent to Robust Optimization Xu,Caramanis&Mannor(2009)Suppos
14、e an adversary can move each training data point by an amount Optimizing the linear support vector objective(,)+is equivalent to minimaxing against this adversary who has a total budget =24 CCAI-2017 Idea 3:Optimize a Risk-Sensitive Objective Setting:Markov Decision Process States:,+1,+2 Actions:,+1
15、 Control policy=()Rewards:,+1 Total reward Transitions:+1,25 +1+1 +1+2 CCAI-2017 0.00.10.20.302468VP(V)Optimize Conditional Value at Risk For any fixed policy,the cumulative return=1 will have some distribution The Conditional Value at Risk at quantile is the expected return of the bottom quantile B
16、y changing we can change the distribution ,so we can try to push the probability to the right“Minimize downside risks”26 CCAI-2017 0.00.10.20.302468VP(V)Optimize Conditional Value at Risk For any fixed policy,the cumulative return=1 will have some distribution The Conditional Value at Risk at quanti
17、le is the expected return of the bottom quantile By changing we can change the distribution ,so we can try to push the probability to the right“Minimize downside risks”27 =0.1 =3.06 CCAI-2017 0.00.10.20.302468VP(V)=3.94 Optimize Conditional Value at Risk For any fixed policy,the cumulative return=1
18、will have some distribution The Conditional Value at Risk at quantile is the expected return of the bottom quantile By changing we can change the distribution ,so we can try to push the probability to the right“Minimize downside risks”28 =0.1 =3.06 CCAI-2017 Optimizing CVaR gives robustness Suppose
19、that for each time,an adversary can choose a vector and define a new probability distribution +1,()Optimizing CVaR at quantile is equivalent to minimaxing against this adversary with a budget along each trajectory of Chow,Tamar,Mannor&Pavone(NIPS 2014)Conclusion:Acting Conservatively Gives Robustnes
20、s to Model Errors 29 CCAI-2017 Many Other Examples Credal Bayesian Networks Convex uncertainty sets over the probability distributions at nodes Upper and lower probability models(Cosman,2000)Robust Classification(Antonucci&Zaffalon,2007)Robust Probabilistic Diagnosis(etc.)(Chen,Choi,Darwiche,2014,20
21、15)30 CCAI-2017 Approaches to Robust AI Robustness to Model Errors Robust optimization Regularize the model Optimize a risk-sensitive objective Employ robust inference algorithms Robustness to Unmodeled Phenomena Detect model weaknesses Repair or expand the model Learn a causal model Employ a portfo
22、lio of models 31 CCAI-2017 Idea 4:Detect Surprises An AI system should monitor itself and its environment to detect surprises that may signal an“unknown unknown”When a surprise is detected Ask the user to help Execute a fallback safety policy CCAI-2017 32 Monitor the Distribution of Predicted Classe
23、s Supervised classification On validation data,measure expected class frequencies Detect departures from these on test data Mismatch can indicate a change in the class distribution or a failure in the classifier CCAI-2017 33 Letter frequencies in English Credit:Nandhp,Wikipedia Look for Violated Exp
24、ectations In search and reinforcement learning,we expect the estimated value to increase as we near the goal When false,this signals potential change in world,new obstacle,etc.CCAI-2017 34 51015200.40.50.60.70.80.91.01.1StepValueMonitor Auxiliary Regularities Hermansky(2013):Each phoneme has charact
25、eristic inter-arrival time Monitor the inter-arrival times of recognized phonemes Apply to detect and suppress noisy frequency bands CCAI-2017 35 Monitor Auxiliary Tasks ALVINN auto-steer system Main task:Determine steering command Auxiliary task:Predict input image Perform both tasks with the same
26、hidden layer information CCAI-2017 36 Pomerleau,NIPS 1992 Watch for Anomalies Machine Learning Training examples drawn from()Classifier =()is learned Test examples from()If=then with high probability()will be correct for test queries What if?CCAI-2017 37 Automated Counting of Freshwater Macroinverte
27、brates Goal:Assess the health of freshwater streams Method:Collect specimens via kicknet Photograph in the lab Classify to genus and species 38 www.epa.gov CCAI-2017 Open Category Object Recognition Train on 29 classes of insects Test set may contain additional species 39 CCAI-2017 Prediction with A
28、nomaly Detection 40 Source:Dietterich&Fern,unpublished CCAI-2017 Anomaly Detector?Classifier Training Examples(,)no =()yes reject Novel Class Detection via Anomaly Detection Train a classifier on data from 2 classes Test on data from 26 classes Black dot:Best previous method 41 CCAI-2017 Related Eff
29、orts Open Category Classification (Salakhutdinov,Tenenbaum,&Torralba,2012)(Da,Yu&Zhou,AAAI 2014)(Bendale&Boult,CVPR 2015)Change-Point Detection(Page,1955)(Barry&Hartigan,1993)(Adams&MacKay,2007)Covariate Shift Correction(Sugiyama,Krauledat&Mller,2007)(Quinonero-Candela,Sugiyama,Schwaighofer&Lawrence
30、,2009)Domain Adaptation(Blitzer,Dredze,Pereira,2007)(Daume&Marcu,2006)43 CCAI-2017 Idea 5:Use a Bigger Model The risk of Unknown Unknowns may be reduced if we model more aspects of the world Knowledge Base Construction Cyc(Lenat&Guha,1990)Information Extraction&Knowledge Base Population Dankel(1980)
31、NELL(Mitchell,et al.,AAAI 2015)TAC-KBP(NIST)Robust Logic(Valiant;AIJ 2001)Risk:Every new component added to a model may introduce an error 44 CCAI-2017 Idea 6:Use Causal Models Causal relations are more likely to be robust Require less data to learn (Heckerman&Breese,IEEE SMC 1997)Can be transported
32、 to novel situations (Pearl&Bareinboim,AAAI 2011)(Schoelkopf,et al.,ICML 2012)(Lee&Honavar,AAAI 2013)45 CCAI-2017 Idea 7:Employ a Portfolio of Models Ensemble machine learning methods regularly win Kaggle competitions Portfolios for SAT solving Portfolios for Question Answering and Search CCAI-2017
33、46 Portfolio Methods in SAT&CSP SATzilla:Xu,Hoos,Hutter,Leyton-Brown(JAIR 2008)47 Presolver 1 Presolver 2 Feature Computation Algorithm Selector Final Algorithm Problem Instance CCAI-2017 SATzilla Results HANDMADE problem set Presolvers:March_d104(5 seconds)SAPS(2 seconds)48 Cumulative Distribution
34、Xu,Hutter,Hoos,Leyton-Brown(JAI R2008)CCAI-2017 IBM Watson/DeepQA Combines 100 different techniques for analyzing natural language identifying sources finding and generating hypotheses finding and scoring evidence merging and ranking hypotheses 49 Ferrucci,IBM JRD 2012 CCAI-2017 Summary Robustness t
35、o Model Errors Probability models with risk-sensitive objectives Optimize against an adversary Regularize the model Optimize a risk-sensitive objective Employ robust inference algorithms Robustness to Unmodeled Phenomena Detect model weaknesses Use a big model Learn a causal model Employ a portfolio
36、 of models 50 CCAI-2017 Outline The Need for Robust AI High Stakes Applications Need to Act in the face of Unknown Unknowns Approaches toward Robust AI Lessons from Biology Robustness to Known Unknowns Robustness to Unknown Unknowns Concluding Remarks 51 CCAI-2017 Concluding Remarks High Risk Emergi
37、ng AI applications Require Robust AI Systems AI systems cant model everything AI needs to be robust to“unknown unknowns”52 CCAI-2017 We have many good ideas We need many more!53 CCAI-2017 Acknowledgments Juan Augusto Randall Davis Trevor Darrell Pedro Domingos Alan Fern Boi Faltings Stephanie Forres
38、t Helen Gigley Barbara Grosz Vasant Honavar Holgar Hoos Eric Horvitz Michael Huhns Rebecca Hutchinson Pat Langley Sridhar Mahadevan Shie Mannor Melanie Mitchell Dana Nau Jeff Rosenschein Dan Roth Stuart Russell Tuomas Sandholm Rob Schapire Scott Sanner Prasad Tadepalli Milind Tambe Zhi-hua Zhou 54 CCAI-2017 Questions?55 CCAI-2017