Download presentation
Presentation is loading. Please wait.
1
Attentive User Interfaces
by Falk Fleischer
2
Overview Introduction Principles & Methodologies Applications Summary
Attention - Sensitive Messaging Mixed - Initiative Interactions Summary 11/29/2018 Introduction
3
STRESS Handy Online Calendar Emails Instant Messages File Scout ...
11/29/2018 Introduction
4
Attentive User Interfaces
Human attention and working memory are limited resources! 11/29/2018 Introduction
5
Attentive User Interfaces
Goal of Attentive User Interfaces : To negotiate the volume and timing of human - computer communication in a natural way 11/29/2018 Introduction
6
Attentive User Interfaces
enhance robustness and fluidity of human - computer collaboration importance of attentive cues goals intentions topics of interest signal offering and accepting a contribution in dialogues 11/29/2018 Introduction
7
Principles & Methodologies
Overview Introduction Principles & Methodologies Applications Attention - Sensitive Messaging Mixed - Initiative Interactions Summary 11/29/2018 Principles & Methodologies
8
Principles & Methodologies
Key Features of AUIs sensing attention use perceptual sensors microphones cameras GPS accelerometers interaction with software devices online calendar history of user’s interests and activities 11/29/2018 Principles & Methodologies
9
Principles & Methodologies
Key Features of AUIs reasoning about attention model user’s attentive state reason about future attention and ideal attention-sensitive actions under uncertainty uses HMMs Bayesian Networks Influence Diagrams 11/29/2018 Principles & Methodologies
10
Principles & Methodologies
Bayesian Networks abstract probabilisitic knowledge base reflecting causal knowledge graph with following qualities 1. random variables as nodes 2. directed links between the nodes 3. each node with conditional probability table 4. no directed cycles 11/29/2018 Principles & Methodologies
11
Principles & Methodologies
Bayesian Networks Body Position Seminar- Room Head Pose Location Focus of Attention 11/29/2018 Principles & Methodologies
12
Principles & Methodologies
Bayesian Networks Task: compute probability distribution for a set of query variables given value of some evidence variables P(Query | Evidence) decisions about actions possible 11/29/2018 Principles & Methodologies
13
Principles & Methodologies
Bayesian Networks kinds of inferences: Diagnostic - from effect to causes Causal - from cause to effects Intercausal - between causes of an effects Mixed - combinations of the ones above inference also possible over time using Dynamic Bayesian Networks 11/29/2018 Principles & Methodologies
14
Principles & Methodologies
Influence Diagrams combines Bayesian Networks with nodes for action and utility 3 types of nodes : chance nodes : random variables action nodes : choice of action influencing following nodes utility nodes : computes the utility of actions given a state of the network 11/29/2018 Principles & Methodologies
15
Principles & Methodologies
Influence Diagrams Body Position Value of Action Location Focus of Attention Expected Utility Actions Seminar- Room Head Pose Cost of Action 11/29/2018 Principles & Methodologies
16
Principles & Methodologies
Influence Diagrams Algorithm 1. set evidence variable for current state 2. for all possible values of the action node do a) set action node to specific value b) calculate utility using standard probabilistic inference algorithm 3. return action with highest utility 11/29/2018 Principles & Methodologies
17
Principles & Methodologies
Key Features of AUIs graceful negotiation of turns and sense of acknowledgement of the request determine utility of interruption given the priority of the information signal information via nonintrusive channel sense user acknowledgement of the information 11/29/2018 Principles & Methodologies
18
Principles & Methodologies
Key Features of AUIs Communicating attention AUIs show their attention to user communicate focus of the user to other AUIs or remote people requesting user’s attention 11/29/2018 Principles & Methodologies
19
Principles & Methodologies
Key Features of AUIs Augmenting attentive resources presents information in estimated focus of user attenuating peripheral detail 11/29/2018 Principles & Methodologies
20
Principles & Methodologies
Key Features of AUIs Sensing attention Reasoning about attention Negotiation of turns and sense of acknowledgement Communicating attention Augmenting attentive resources 11/29/2018 Principles & Methodologies
21
Overview Introduction Principles & Methodologies Applications Summary
Attention - Sensitive Messaging Mixed - Initiative Interactions Summary 11/29/2018 Applications
22
The Notification Platform Project
Microsoft Research, Redmond centered on economic based principles: value of information vs. attentive-sensitive cost of disruption 11/29/2018 Applications
23
The Notification Platform Project
subscription of different sources and output devices possible considers user’s attention and location under uncertainty aware of fidelity and relevance of different communication channels 11/29/2018 Applications
24
The Notification Platform Project
s Handy Online Calendar Notification Platform Instant Messages File Scout ... 11/29/2018 Applications
25
The Notification Platform Project
Universal Inbox Bayesian attentive model Context Server & Context Whiteboard Decision model Profile of user’s preferences 11/29/2018 Applications
26
The Notification Platform Project
11/29/2018 Applications
27
The Notification Platform Project
Example: A new message coming into the Universal Inbox 11/29/2018 Applications
28
The Notification Platform Project
1. How important is this message? - question of informing user now or later - computes cost of delayed review Information Value 11/29/2018 Applications
29
The Notification Platform Project
2. Which device and modality should I use? a) What is the user doing now? Focus of Attention b) Will the user receive my notification on this device? c) Will I lose fidelity of information when using this device? Transmitted Value Value of Action Device & Modality 11/29/2018 Applications
30
The Notification Platform Project
3. How expensive is it to alert the user now? - uses preferences stored in profile Attentional cost Focus of Attention Cost of Action Device & Modality 11/29/2018 Applications
31
The Notification Platform Project
for every device and modality compute the Expected Utility = Transmitted Value - Attentional cost choose action with highest utility! Device & Modality Value of Action Cost of Action Expected Utility 11/29/2018 Introduction
32
The Notification Platform Project
11/29/2018 Applications
33
Overview Introduction Principles & Methodologies Applications Summary
Attention - Sensitive Messaging Mixed - Initiative Interactions Summary 11/29/2018 Applications
34
DeepListener Project spoken language system
address the “speech-target problem” use probabilistic models to compute the likelihood of being the target of speech, of what was said and of different intentions clarification dialogues use cues for conversational turn taking attentional lens or animated agent 11/29/2018 Applications
35
DeepListener Project 11/29/2018 Applications
36
Quartet Project continuous speech recognition system
advanced model of attention: keyboard events visual pose disruption of attention between device and human natural language processing realises misrecognition and intention for someone else 11/29/2018 Applications
37
Quartet Project 11/29/2018 Applications
38
Overview Introduction Principles & Methodologies Applications Summary
Attention - Sensitive Messaging Mixed - Initiative Interactions Summary 11/29/2018 Summary
39
Summary if we extend devices with attention sensors
advanced representation & inferential methods communication capabilities 11/29/2018 Summary
40
Summary it is possible to build
attentive sensible unintrusive sociable devices! but there is still much research to do : recognition, modelling, reasoning, psychology 11/29/2018 Summary
41
Literature E. Horvitz, C. M. Kadie, T. Paek, D. Hovel. Models of Attention in Computing and Communications: From Principles to Applications, Communications of the ACM 46(3):52-59, March 2003 Jeffrey S. Shell, Ted Selker, Roel Vertegaal: Interacting with groups of computers. Communications of the ACM 46(3): 40-46, March 2003 Paul P. Maglio, Christopher S. Campbell: Attentive agents. Communications of the ACM 46(3): , March 2003 Roel Vertegaal: Introduction. Communications of the ACM 46(3): 30-33, March 2003 Stuart Russell, Peter Norvig. Artificial Intelligence: A Modern Approach. Prentice Hall, 1995 11/29/2018 Summary
42
Thank you for your kind attention!
11/29/2018
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.