Presentation is loading. Please wait.

Presentation is loading. Please wait.

Utah School of Computing Intelligent User Interfaces: AI and Machine Learning CS5540 HCI ( Fall 2009 ) Rich Riesenfeld (based on a guest lecture by Cindy.

Similar presentations


Presentation on theme: "Utah School of Computing Intelligent User Interfaces: AI and Machine Learning CS5540 HCI ( Fall 2009 ) Rich Riesenfeld (based on a guest lecture by Cindy."— Presentation transcript:

1

2 Utah School of Computing Intelligent User Interfaces: AI and Machine Learning CS5540 HCI ( Fall 2009 ) Rich Riesenfeld (based on a guest lecture by Cindy Thompson, PhD Lecture Set 5

3 Student Name Server Utah School of Computing slide 2 Adaptive User Interfaces Definition: An adaptive user interface is a software artifact that improves its ability to interact with a user by constructing a user model based on partial experience with that user. Definition: An adaptive user interface is a software artifact that improves its ability to interact with a user by constructing a user model based on partial experience with that user. AUI are really at the intersection of Human Computer Interaction and Machine Learning (the latter is an area within ArtificialIntelligence) AUI are really at the intersection of Human Computer Interaction and Machine Learning (the latter is an area within ArtificialIntelligence)

4 Student Name Server Utah School of Computing slide 3 User Interface Dimensions More than one form of presentation modality (speech, vision, sound)More than one form of presentation modality (speech, vision, sound) More than one form of interaction (press buttons, type, speak)More than one form of interaction (press buttons, type, speak) -Example: Organization on web page different for blind users: key information at top of page where it is read to user sooner -Example: Customized on-screen keyboard for disabled users

5 Student Name Server Utah School of Computing slide 4 User Interface Dimensions Different content for different usersDifferent content for different users -Example: different levels of expertise at using interface and at the knowledge level 2

6 Student Name Server Utah School of Computing slide 5 What is Machine Learning? What is learning?What is learning? changes in [a] system that... enable [it] to do the same task or tasks drawn from the same population more efficiently and more effectively the next time. (Simon 1983)

7 Student Name Server Utah School of Computing slide 6 What is Machine Learning? What is learning?What is learning? any computer program that improves its performance at some task through experience. (Mitchell 1997) any computer program that improves its performance at some task through experience. (Mitchell 1997) 2

8 Student Name Server Utah School of Computing slide 7 What is Machine Learning? There are two ways that a system can improve: 1.By acquiring new knowledge -For example: acquiring new facts or skills 2.By adapting its behavior -For example: solving problems more accurately or efficiently 3

9 Student Name Server Utah School of Computing slide 8 Input to an ML System Most ML algorithms use a training set of examples as the basis for learningMost ML algorithms use a training set of examples as the basis for learning Each example is encoded using the instance representation chosen for the problemEach example is encoded using the instance representation chosen for the problem

10 Student Name Server Utah School of Computing slide 9 Input to an ML System Examples presented to a machine learning algorithm are typically represented as attribute-value pairs.Examples presented to a machine learning algorithm are typically represented as attribute-value pairs. -attribute is a general property associated with an object. For example, we might describe animals with the attributes: size, color, temp, has tail, has beak, covering 2

11 Student Name Server Utah School of Computing slide 10 Input to an ML System Examples presented to a machine learning algorithm are typically represented as attribute-value pairs.Examples presented to a machine learning algorithm are typically represented as attribute-value pairs. -value is one possible instantiation of the attribute -CANARY might be represented as: size=small, color=yellow, temp=warm- blooded, has tail=true, has beak=true, covering=feathers 3

12 Student Name Server Utah School of Computing slide 11 Training Experience Direct or indirect?Direct or indirect? -Direct experience consists of individual states/items and their individual classifications -Indirect experience consists of a sequence of states and a final classification only. Credit assignment is a major issue

13 Student Name Server Utah School of Computing slide 12 Training Experience Teacher or self-selected training?Teacher or self-selected training? -A teacher may select informative training examples -The learner may propose training examples that would be especially useful to learn from 2

14 Student Name Server Utah School of Computing slide 13 Training Experience Distribution of training data: Generally assume training data is representative of the examples to be judged when tested for final performanceDistribution of training data: Generally assume training data is representative of the examples to be judged when tested for final performance 3

15 Student Name Server Utah School of Computing slide 14 Adaptive Interfaces: Discussion Points Dimensions along which interfaces could adapt?Dimensions along which interfaces could adapt? What challenges do you anticipate for Artificial Intelligence when applied to interface adaptation?What challenges do you anticipate for Artificial Intelligence when applied to interface adaptation?

16 Student Name Server Utah School of Computing slide 15 Dimensions of Adaptation Data processing levelData processing level Information Filtering levelInformation Filtering level Information Presentation levelInformation Presentation level

17 Student Name Server Utah School of Computing slide 16 AUI: Application of ML Fielding and AcceptanceFormulating the Problem Engineering the Representation New Knowledge InductionProcess Evaluating the Gathering Training Data

18 Student Name Server Utah School of Computing slide 17 AUI: App of ML

19 Student Name Server Utah School of Computing slide 18 Examples of Adaptive UIs Information filtering:Information filtering: -Syskill & Webert -NewsWeeder Generating new knowledge to satisfy users goals:Generating new knowledge to satisfy users goals: -Scheduling -Part layout

20 Student Name Server Utah School of Computing slide 19 Examples of Adaptive UIs Optimization:Optimization: -Route advisor -Scheduling Information entry:Information entry: -Predict keystrokes or Unix commands -Form filling 2

21 Student Name Server Utah School of Computing slide 20 Syskill & Webert in Depth (Pazzani et al, 1996) Technique: Recommends web pages on a user- specified topicRecommends web pages on a user- specified topic Accepts user feedback about pages user selectsAccepts user feedback about pages user selects Represents each web page as a bag of wordsRepresents each web page as a bag of words

22 Student Name Server Utah School of Computing slide 21 Syskill & Webert in Depth (Pazzani et al, 1996) Technique: Uses naive Bayes to adapt to users page preferencesUses naive Bayes to adapt to users page preferences Evaluation: 80% accurate predicting users movie opinions after training on only 35 pages 2

23 Student Name Server Utah School of Computing slide 22 Syskill & Webert in Depth 3

24 Student Name Server Utah School of Computing slide 23 Syskill & Webert in Depth Profile can be used to: Suggest which links might interest userSuggest which links might interest user Construct a Lycos query to find interesting (to user) pagesConstruct a Lycos query to find interesting (to user) pages Learns a profile for each topic and userLearns a profile for each topic and user 4

25 Student Name Server Utah School of Computing slide 24 Syskill & Webert in Depth Functionality is added to pages to collect user feedback: Hot (2 thumbs up),Hot (2 thumbs up), Llukewarm (1 up, 1 down), orLlukewarm (1 up, 1 down), or Cold (2 thumbs down);Cold (2 thumbs down); But learner only uses 2 classes by combining lukewarm & cold 5

26 Student Name Server Utah School of Computing slide 25 Syskill & Webert Discussion What are the advantages of this method?What are the advantages of this method? What are the disadvantages?What are the disadvantages? How might we improve the interface? The learning?How might we improve the interface? The learning?

27 Student Name Server Utah School of Computing slide 26 Form Filling in Depth (Schlimmer & Hermens, 1993) Collects traces of secretary filling out vacation formsCollects traces of secretary filling out vacation forms Learns rules that predict some fields based on othersLearns rules that predict some fields based on others Suggests default entries for fields that the user can overrideSuggests default entries for fields that the user can override Evaluation: Reduced keystrokes by 87%

28 Student Name Server Utah School of Computing slide 27 Issue: Where to get training data? Past user actionsPast user actions Domain/application-specific informationDomain/application-specific information ContextContext What other users have doneWhat other users have done

29 Student Name Server Utah School of Computing slide 28 Issue: Nature of User Feedback Least specific: Binary feedback: WebWatcher, Syskill & WebertLeast specific: Binary feedback: WebWatcher, Syskill & Webert Intermediate: ordering on choices: Adaptive Route Advisor, INCAIntermediate: ordering on choices: Adaptive Route Advisor, INCA Most specific: scoring: FireflyMost specific: scoring: Firefly Passive versus ActivePassive versus Active

30 Student Name Server Utah School of Computing slide 29 Alerting in Depth (Horvitz, et.al., 1999) Send an alert when important arrives Compares expected cost of interruption to expected cost of delaying notificationCompares expected cost of interruption to expected cost of delaying notification Bayes net to assess users focus of attentionBayes net to assess users focus of attention Learns to assess criticality of a message from with user-labeled criticality valuesLearns to assess criticality of a message from with user-labeled criticality values Related Application: filtering whether to send to users remote device

31 Student Name Server Utah School of Computing slide 30 General Design Issues What to predict? (how can we best help users)What to predict? (how can we best help users) Demand on user (obtrusive versus unobtrusive)Demand on user (obtrusive versus unobtrusive) Should user be able to examine and change their model?Should user be able to examine and change their model?

32 Student Name Server Utah School of Computing slide 31 General Design Issues Should user be able to override the systems changing itself?Should user be able to override the systems changing itself? EvaluationEvaluation Design of interface more tied to design of adaptation process?Design of interface more tied to design of adaptation process? 2

33 Student Name Server Utah School of Computing slide 32 Evaluation Discussion What aspects of the system would you want to improve as a result of its learning about you?What aspects of the system would you want to improve as a result of its learning about you? How easy are these to measure?How easy are these to measure? Is there a difference between real (quantitative) and perceived (qualitative) improvements?Is there a difference between real (quantitative) and perceived (qualitative) improvements? -Which would be more important to you?

34 Student Name Server Utah School of Computing slide 33 Dimensions of Evaluation Dependent measuresDependent measures -Solution speed -Solution quality -Amount of effort reduced -User satisfaction -Predictive accuracy

35 Student Name Server Utah School of Computing slide 34 Dimensions of Evaluation Independent variablesIndependent variables -Number of user interactions -Characteristics of system, user, and task 2

36 Student Name Server Utah School of Computing slide 35 Ethical Issues PrivacyPrivacy EtiquetteEtiquette Is it legal?Is it legal? Could we do anonymous user models?Could we do anonymous user models? Where is the model stored and who owns it?Where is the model stored and who owns it?

37 Student Name Server Utah School of Computing slide 36 Ethical Issues Is the interface trying to influence the user?Is the interface trying to influence the user? -Adaptive tutor vs. -Selling you a product you would buy anyway vs. -(Trying to) Change your buying habits or preferences 2

38 Student Name Server Utah School of Computing slide 37 User Modeling & Intelligent Interfaces StereotypesStereotypes Dialog systemsDialog systems Just do it? Or ask first?Just do it? Or ask first? PersonificationPersonification

39 Student Name Server Utah School of Computing slide 38 Dialog Systems Bring in techniques from natural language understanding, speech recognition, human factors, and problem solving and inferenceBring in techniques from natural language understanding, speech recognition, human factors, and problem solving and inference Most existing systems are task specificMost existing systems are task specific Developing the knowledge base needed to control a system is time and resource intensiveDeveloping the knowledge base needed to control a system is time and resource intensive

40 Student Name Server Utah School of Computing slide 39 Dialog Systems We focus on decreasing that time by improving a simple system,online, after it is builtWe focus on decreasing that time by improving a simple system,online, after it is built Our first effort improves along the dimension of asking better questions when helping the user find informationOur first effort improves along the dimension of asking better questions when helping the user find information 2

41 Student Name Server Utah School of Computing slide 40 Demo Adaptive Place Advisor discussion and demonstration

42 Student Name Server Utah School of Computing slide 41 What is the Future? Aides, not AgentsAides, not Agents Intelligent EnvironmentsIntelligent Environments Wearable computersWearable computers Conversational interfacesConversational interfaces CommercializationCommercialization

43 Student Name Server Utah School of Computing slide 42 Personalization on the Web FireflyFirefly WiseWireWiseWire News DudeNews Dude Many othersMany others

44 Student Name Server Utah School of Computing slide 43 Issues for Web Personalization Of 1400 random web sites, 92% collected great amounts of personal data (US Federal Trade Commission, 1998)Of 1400 random web sites, 92% collected great amounts of personal data (US Federal Trade Commission, 1998) Many users do not wish to register with web sites, or if they do, they provide fake infoMany users do not wish to register with web sites, or if they do, they provide fake info Some countries do not allow usage logs to be kept from session to sessionSome countries do not allow usage logs to be kept from session to session Security of your personal information!Security of your personal information!

45 Student Name Server Utah School of Computing slide 44 Music Recommendation in Depth (Shardanand & Maes, 1994) Users provide ratings of music itemsUsers provide ratings of music items System provides recommendations of other items user may also enjoySystem provides recommendations of other items user may also enjoy Compare ratings to other users ratings and suggest music accordinglyCompare ratings to other users ratings and suggest music accordingly

46 Utah School of Computing Intelligent User Interfaces End End Lecture Set 5


Download ppt "Utah School of Computing Intelligent User Interfaces: AI and Machine Learning CS5540 HCI ( Fall 2009 ) Rich Riesenfeld (based on a guest lecture by Cindy."

Similar presentations


Ads by Google