Advanced Belief Reasoning in Intelligence (ABRI)
Abstract
The ABRI research project aims at investigating and developing software tools for intelligence analysis based on subjective logic. A typical aspect of intelligence analysis is that it takes place in environments of incomplete knowledge and uncertain evidence, so the ability to handle uncertainty formally and mathematically supports more realistic reasoning models that produce more realistic conclusions. The project will first investigate how the theory of subjective logic can best be applied and implement to provide practical solutions for handling uncertainty in intelligence analysis. Subsequently a basic intelligence analysis system will be implemented, tested and evaluated. While the original ACH (Analysis of Competing Hypotheses) methodology is specified as a manual and qualitative process, several enhancements have been proposed in the literature to make ACH more formal. Evidence used in intelligence analysis is typically incomplete and confidence in the evidence and their sources is often limited. This calls for a formal framework that explicitly takes uncertainty and missing evidence into account. The main advantage of subjective logic over many other reasoning frameworks is precisely that it takes degrees of uncertainty and ignorance explicitly into account, which mans that it formal reasoning based on subjective logic can include levels of confidence in evidence and their sources. A typical aspect of intelligence analysis is that it takes place in an environment of incomplete knowledge and uncertain evidence. Subjective logic is therefore particularly suitable for intelligence analysis, in particular when combined with the ACH approach because it is allows gradual assimilation of evidence and the ability to immediately see the effect it has on candidate hypotheses.. Building intelligence analysis models based on subjective logic represents a totally new approach which makes it possible to model and analyze real world situations when evidence becomes available incrementally. The uncertainty of initial input evidence will simply be reflected in the output at that stage. The crucial point is that the model and the derived conclusions will correctly reflect the evidence available to the analyst irrespectively of its degree of completeness and confidence. The main advantage of this is that evidence can be incrementally added to produce analysis results with increasing levels of certainty. It is also important for consumers of intelligence produced in this way to explicitly be informed about its level of confidence. ABRI focuses some or all the following tasks in the order presented: 1. Specify components from subjective logic that will be included in the tool. While the fundamental theory for belief reasoning with subjective logic is well defined, there is a need to expand this framework with methods specified by tasks 2) and 3) below. 2. Describe machine learning methods for subjective logic. 3. Describe classification methods for subjective logic. 4. Implement and evaluate a basic tool for Bayesian network modelling and intelligence analysis The method consists of the following: - Studying relevant literature - Developing theoretical models - Implementing prototypes - Validating prototypes The significance and high level goal of the ABRI project is to strengthen the intelligence capabilities of US military agencies.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1510470
Entities
People
- Audun Josang
Organizations
- Army Contracting Command
- United States Army
- University of Oslo