|
Active Information Fusion for Decision Making
Project goal:
Develop a unified probabilistic
framework for effective representation of sensor data of different modalities,
for active fusion of sensory data, and for efficient and timely
decision-making.
Technical Challenges:
 Develop
and implement a DID for simultaneous sensor fusion, sensor selection, and
decision making
Learn model parameter
efficiently when data are incomplete
1). Develop computational methods to efficiently compute sensor selection
criteria
2).
Develop active sensor selection strategies to efficiently identify an optimal
sensor set
3).
Efficiently integrate the acquired sensory data, update the belief in
current hypothesis, and make optimal decisions
Accomplishments:
1. Develop and implement a framework for simultaneous fusion, sensor selection,
and decision making
i) A Dynamic Bayesian
Network for Active Information Fusion
ii) Active Infomation Fusion for Decision Making using a Dynamic Influence
Diagram
2. Develop a learning algorithm to estimate model
parameters when data are incomplete.
i) A Constrained Expectation-Maximization Learning Algorithm
3. Develop the sensor selection criterion and computational methods to
efficiently compute the criterion.
i) Mutual Information based
method
ii)Value-of-Information based
method
4. Develop the active sensor selection strategies to efficiently identify a
subset of optimal sensors
i)
Sensor Selection Method One
ii) Sensor Selection Method Two
5. Efficiently integrate the acquired sensory data and update the beleif to
current hypothesis
i) An Efficient BN Inference
Algorithm
|
