Update meeting November 24 th
2 Outline Paper presented in Int’l Conference on Ubiquitous Computing (UCS’04), Japan Current work on new paper – Data Fusion Algorithm
3 Recap on previous talk: The Context Spaces Model SjSj SiSi RiRi Operators:
4 UCS’04 - Situations in the Smart Room: problem of uncertainty a b c x y z - A Meeting - A Presentation - Friday Gathering visualization of situation definitions in smart room and context states
5 UCS’04 - Verification I State-space difference = where, User in a meeting User working in office
6 UCS’04 - Verification II The Objective: Recognize conflicts and intelligently resort to other elements in the environment for verification. ‘On’‘Off’
7 UCS’04 - Verification II Event TypesSituation Subspaces Contradicting Event Values Value corresponds to def. of Identifying supporting events
8 UCS’04 - Verification III Healthy Walking Standing Running Sick Respiratory Rate Heart Rate Running Sick Standing The problem: ‘Running’? ’Sick’? (‘Running’ ‘Sick’)? The procedure: We reveal that: (If ‘Walking’ AND ’Sick’ then infer ‘Running’) ’Walking’ ’Sick’ It is more likely that we are in: ‘Running’
9 UCS’04 - The Reasoning Process Knowledge Synthesis Conflict Analysis Verification Situation Composition Low-Level Discrepancies Discovery Raw Data input Sensor data Knowledge Synthesis Conflict analysis Verification Situations Composition input output Complex Situations Basic Situations Refined Situations Reasoning Process
10 UCS’04 - Verification Layer - the ReaGine Prototype Load Monitor Reasoning Engine Cleaner KB Data Assimilator Data Inserter Monito r Stability Analyzer MQMQ MQMQ Senso Sensor Event Router Algorithms Library Monito r Same run with added uncertainty Experimental run Experiment deployment architecture
11 New Paper – Synthesising Sensor Data for Context-Aware Applications Introduces a new sensor fusion technique: –Built over a general context model –Consider intuitions relevant to context-aware computing –Provides a different perspective to model knowledge & uncertainty e2 e3 e4 e5 e6 e7 e8 e9 situation1 situation2 situation3 situation4 e1 eventssituationssensor fusion Relating events to situations Dempster-Shafer starting point Context Spaces starting point
12 New Paper – Synthesising Sensor Data for Context-Aware Applications for If < C then = 0; for where I. II. III. where = = IV. where and Built over the context spaces model Based on multi-attribute utility theory and probability theory Considers intuitions from human perception for context reasoning
13 New Paper – Algorithm Characteristics B B A A attribute nameserialimportance (1-5)optionalweight User RFID Y Location14No User RFID X Location14No User PDA Y Location13Yes User PDA X Location13Yes MR Light Level14No MR Light Level24No MR Noise Level12.5No MR Motion Detected12.5No MR Projector Active14Yes MR Microphone Active14Yes Examples of factors covered by algorithm: Individual significance of events The importance of a complete match Degree of trust in evidence accuracy Changing significance of match rejection
14 Use Synthesis process to reason about and distinguish between: User presenting User attending another’s presentation User in a meeting Context Attributes: User notebook’s keyboard activity User notebook active presentation processes Light level in room User position (by tracing his mobile devices) New Paper – Smart Room Experiment
15 New Paper – Experiment Deployment & Design Motes Interface Service Positioning Engine Service Hook user activity Identify presentation act. Synthesis Process Data Interpretation Positioning Engine Notebook Location Light sensors Notebook Inferred Location
16 New Paper – Sensor Technology Ekahau Positioning Engine: Berkeley Motes:
17 New Paper – Results Synthesis results in differentiating between situations Time (in min.) 15min 30min45min U. Presenting U. in Presentation U. in Meeting U. in Presentation U. Presenting U. in Meeting U. in Presentation U. Presenting U. in Meeting Unpredictable user behavior SynthesisDempster-Shafer