HARD AND SOFT SENSOR INFORMATION FUSION USING COGNITIVE INJECTION PROCESS FOR DECISION MAKING
Open Access
- Author:
- Sudit, David M
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- April 24, 2012
- Committee Members:
- Soundar Kumara, Dissertation Advisor/Co-Advisor
Soundar Kumara, Committee Chair/Co-Chair
David J Hall, Committee Member
David Arthur Nembhard, Committee Member
Guodong Pang, Committee Member - Keywords:
- Information Fusion
Human Fuser
Cognitive
Uncertainty
Sensors - Abstract:
- Advances in technology have led to the abundant availability of data from heterogeneous sensors. This combined with knowledge about the sensed data has improved our ability to turn data into reliable and concise information. The process of combining data from different sources is termed Data or Information Fusion. In particular, a new area of focus has been on Hard and Soft fusion. The idea of Hard and Soft fusion stems from a classification system that divides sensors into two groups: Hard and Soft sensors. Hard sensors are the traditional electronic sensing sources like satellites and RFIDs that use a programmed system to collect the data. Soft sensing occurs when a human is collecting data. This type of sensing is clearly less predictable, but gives a different dimension to the complete set of data. With the availability of new communication methods (eg. Facebook, Twitter, YouTube) any person in any setting could be a soft sensor. With the advent of crowd sourcing it will be common to encounter the context of soft sensing in the future. With the inclusion of humans as sources of data, Information Fusion algorithms should be able to combine these disparate sources. In this thesis I focus on including humans in a decision making system to improve the knowledge of a Decision Maker (DM). I accomplish this by using a subject matter expert, whom I call a Human Fuser, to manipulate the reliability values extracted by the Information Fusion process. The Information Fusion process is a hard process as it uses programs to align and associate data to be able to estimate pieces of information. These pieces of information have a reliability or certainty value attached with them that has been extracted by looking at the sources (the hard and soft sensors) error functions. I introduce a soft process, called Cognitive Injection Process into the Information Fusion process. Including soft sensors will change the DM’s knowledge of the problem scenario because the reliability values of the different pieces of information will be changing. In a decision making context I need to consider several features, such as the color of a car, speed at which it is travelling etc. I should consider these features and select the best feature-estimate set, which can help the decision maker in inferring for example in this case, color and speed. In this thesis I address three sub-problems: 1. Feature-Estimate selection through linear mathematical models, 2. Feature-Estimate selection through complex mathematical models, and 3. Introducing multiple human fusers into the information fusion process. In the first problem I model the feature-estimate selection for soft sensing as a knapsack problem and solve it through heuristics. I consider entropy modeling and minimize entropy, thus leading to best feature-estimate set, in the second problem. Finally I use learning and forgetting models to address the problem of introducing multiple humans in the information fusion process.