Models and Programmability Breakout III

1
Models and Programmability (Breakout III)

• Amr, Stan, Amol, Christos, Ugur, Cyrus

2
Vision What we want to see in 10 years? Sixth
Sense

• Preventive actions
• Healthcare disease outbreak
• Crisis traffic accidents, airplane crash
• Predictive actions
• National security predicts terrorists next
  action
• Efficiency
• Green houses, hybrid vehicles (lots of sensors)
• Real-time searching (sensor Google)
• Fusion of web, sensor data,

3
Motivating Application

• Moving sensors, soldiers with sensors
• Two types of queries/applications
• Monitoring environment and phenomena (e.g.,
  average temp of an area)
• Moving object queries (e.g., when the number of
  soldiers reaches n in a given area)

4
Assumed Architecture
Queries
Logical layer
Modeling
Feature layer
Feature extraction
Raw data layer
Data acquisition
Moving sensors
5
Challenge 1 A Generic way of Incorporating
Models

• Use of model
• Missing values
• Optimization
• Fault tolerance (exploiting redundancy in models)
• Prediction
• Compression (data reduction)
• Different types of models
• Statistical Models
• Geospatial Models
• Constraints
• Model source
• Supplied by domain expert
• Learned

6
Challenge 1

• Designing Models, using techniques from
• Data mining
• Time series analysis
• Computational geometry
• Spatiotemporal databases
• ..

7
Architecture
Queries
Logical layer
Modeling
Feature layer
Feature extraction
Raw data layer
Data acquisition
Moving sensors
8
Challenge 2 Incorporating more info into queries

• Resolution
• Confidence
• Pattern
• Models
• Uncertainty
• Triggers
• Linguistic constructs that retain the
  declarative-ness
• How each of the above can be defined
• Implementation issues

9
Architecture
Queries
Logical layer
Modeling
Feature layer
Feature extraction
Raw data layer
Data acquisition
Moving sensors
10
Challenge 3 Cross-layer optimization

• Examples
• Models can dictate data acquisition rate
  (adaptive sampling rate)
• Queries can dictate data acquisition rate
  (adaptive sampling rate)
• Given patterns in the query, what models
  algorithms are the best to use
• Given a trajectory at the query, move/actuate
  sensors at the right locations
• Which set(s) of sensors, historical data, models
  to use for a given query

11
Architecture
Queries
Logical layer
Modeling
Feature layer
Feature extraction
Raw data layer
Data acquisition
Moving sensors
12
Challenge 4 How to incorporate Uncertainty at
different layers

• Missing values
• Faulty sensors
• Conflicting sensed values

13
Visionary Outlook Privacy

• Exploiting space, time and latency to
  identify the level of privacy and data protection
• Trade-off between the required response time and
  level of security
• Secure enough given space and time
• At which layer(s) to incorporate the privacy