Intelligent Information Processing -- Projects

Analyze, Share, Know (ASK)

Audio Hot Spotting

Automated Discovery of Structural Patterns in Link Analysis

Automated Information Discovery and Retrieval from Asian Language Sources

Foundations for Next Generation Information Access

Graph Based Data Mining

Robot Platoon Command and Control

Social Information Retrieval

Intelligent Information Processing

Intelligent Information Processing investigates technologies, tools, and processes that support the discovery, processing, exploitation and dissemination of information, tools and knowledge. Intelligent agents are covered in this area.

Analyze, Share, Know (ASK)

David Anderson, Principal Investigator

Bedford and Washington

Problem
Analysts must form a correct view of the world from very large quantities of data that are not necessarily structured to help them answer their questions. Every analyst must potentially be able to leverage the insights gained by the work of others in the organization. The analyst should be able to know what the rest of the organization knows about the target subject and that information’s relevancy to the current analysis task.

Objectives
Using a mix of COTS, GOTS, and research components, the ASK program seeks to demonstrate novel analytic tools to support efficient inference against large data collections, analysis coordinated through structured argumentation, and an integrated, collaborative knowledge management environment for the analytical enterprise.

Activities
Leveraging previous work through judicious systems integration, the ASK team is creating prototypes of several advanced capabilities, including: automatic generation of timelines with event extraction; spatio-temporal fusion of imagery, signal processing, and linguistic sources; structured argumentation to organize all-source analysis; dialogue agent analytic tool interfaces mediated by an instant messaging system; reinforcement learning in the analytic environment; and analytic enterprise integration strategies.

Impacts
On-going interaction between members of the ASK team and specific government communities informs our choice of tasks. Capabilities demonstrated here are being fed back into the real-world work of intelligence systems engineering.

Audio Hot Spotting

Qian Hu, Principal Investigator

Bedford and Washington

Problem
Large volumes of recordings require rapid retrieval of segments potentially relevant to a given query (audio hot spotting). Spoken document retrieval systems that simply combine automatic speech recognition (ASR) with information retrieval (IR) do not meet this need in real applications. This is because of high ASR word error rates and the loss of important audio information in the speech transcription.

Objectives
We propose to research and develop audio-specific retrieval algorithms in critical domains by 1) exploiting multiple types of acoustic information from the audio signals; 2) exploring several adaptive techniques to improve existing ASR performance; and 3) fusing component technologies such as ASR, language/speaker identification, audio feature extraction, and information retrieval.

Activities
We will research algorithms and techniques to extend and improve ASR and audio feature extraction and to develop audio-based query algorithms making use of the multiple types of audio information. We will research and develop fusion algorithms to build an audio hot spotting system based on the extended ASR, audio feature extraction, language/speaker identification, and the new audio query language.

Impacts
Our research in audio hot spotting algorithms and prototype development will address the needs of MITRE's sponsors with warehouses of recordings waiting for efficient retrieval. It will extend MITRE's information retrieval capability from text to include audio. The expertise gained through the research will equip MITRE to better advise industry developers and our sponsors on audio information retrieval topics and evaluation standards.

Automated Discovery of Structural Patterns in Link Analysis

Daniel Venese, Principal Investigator

Washington only

Problem
The threat from terrorist actions is perceived to be immediate and growing. However, it is believed that clues to the capabilities, intentions and organization of terrorist and other criminal groups can be found in already available data. Identification of suspect behavior is made difficult through the intentional concealment of relationships, and thus the efficient discovery of patterns from large databases is a very difficult problem.

Objectives
We will explore promising, although unproven and technically risky, new approaches for automating the discovery of patterns of suspicious behavior and associations. We will attempt to train a classifier that can identify criminals or terrorists based on descriptions of the types of associations or relationships that they have in common.

Activities
We will develop a demonstration prototype for a large-scale repository supporting link discovery and analysis. Initial emphasis will be on techniques for transforming multiple, large databases into an integrated, searchable link representation. We will test approaches for storage and traversal of the links and mechanisms for inserting additional links. A series of three prototypes will demonstrate increasingly sophisticated techniques. The link repository will grow with each demonstration.

Impacts
The proposed research can contribute to the national effort by bringing new methodologies to bear in discovering terrorists, terrorist organizations, fraud, and other criminal behavior. There are challenges in attempting to scale current practices in link analysis to large-scale databases and to find suspects based upon data that is intentionally being manipulated. The methodology could be transferred into the counterterrorism and law enforcement domains if it proves effective.

Automated Information Discovery and Retrieval from Asian Language Sources

Ray LeBlanc, Principal Investigator

Bedford and Washington

Problem
While several commercial capabilities exist to address particular facets of machine translation (MT) needs, emphasis has been placed on European-based languages. Furthermore, none of the existing COTS products iis particularly well suited to the military environment. English translation of the Asian languages is a much more difficult problem than for European and has presented the MT community with significant challenges.

Objectives
This project will develop a capability to perform Chinese and Korean cross-language information retrieval, information discovery (ID), data mining (DM), and knowledge management (KM) in support of open source intelligence analysis. The project will develop a prototype capability that can support in-field experimentation with a broad spectrum of users.

Activities
The project will provide an automated capability to translate electronic textual information between Chinese and English, and between Korean and English. We will characterize and subsequently retrieve information, based on user-specified profiles, from Chinese and Korean language sources by means of a prototype analytic tool. A dictionary management capability will allow users to build, import/export, and aggregate custom dictionaries.

Impacts
This project has the potential for improving the efficiency and effectiveness of intelligence organizations currently impacted by foreign language translation issues. It is expected to provide the beneficiaries with needed interim capabilities and validation of the most fertile areas for the future application of government funds.

Foundations for Next Generation Information Access

Warren Greiff, Principal Investigator

Bedford and Washington

Problem
Computerized support for information gathering is fragmented across multiple research communities and integration is difficult due to a lack of an underlying formalism that cuts across the different technologies. Statistical techniques developed for individual components have been developed in isolation and without a common theoretical foundation. As a result we are left with a number of reasonably effective, semi-principled, incompatible techniques.

Objectives
The principal objective is the development of statistical foundations for information access. A successful foundation will comprise rigorous characterizations of the issues of modeling and estimation, together with principled methodologies for adapting to new languages, genres, information domains, auxiliary knowledge sources and tasks.

Activities
We will develop simulations that model the stochastic generation of latent document features, observable document features, the determination of document relevance, and the distribution of query characteristics. We will perform exploratory data analysis on available research corpora to verify our models. A central focus will be research into the importance of variance reduction and the potential benefit of various bias-variance strategies.

Impacts
This research is of direct relevance to existing MITRE projects. The results will allow MITRE to develop information access systems incorporating new sources of evidence and to tailor information systems to meet specific military and intelligence needs. MITRE will then be strategically positioned to set the direction of research into, and development of, next generation information access technology.

Graph Based Data Mining

Daniel Venese, Principal Investigator

Washington only

Problem
The threat from terrorist actions is perceived to be immediate and growing. However, it is believed that clues to the capabilities, intentions and organization of terrorist and other criminal groups can be found in already available data. Identification of suspect behavior is made difficult through the intentional concealment of relationships, and thus the efficient discovery of patterns from large databases is a very difficult problem.

Objectives
We will explore promising, although unproven and technically risky, new approaches for automating the discovery of patterns of suspicious behavior and associations. We will attempt to train a classifier that can identify criminals or terrorists based on descriptions of the types of associations or relationships that they have in common.

Activities
We will develop a demonstration prototype for a large-scale repository supporting link discovery and analysis, emphasizing techniques for transforming multiple, large databases into an integrated, searchable link representation. Three prototypes will demonstrate increasingly sophisticated techniques for storage and traversal of the links and mechanisms for inserting additional links. Sensitive but unclassified databases from the USCS will form the basis for the second and third demonstrations.

Impacts
The research can contribute to the national effort by applying new methodologies to discover terrorists, terrorist organizations, fraud, and other criminal behavior. Feasibility of these techniques in the USCS environment will be demonstrated in the areas of counterdrugs and counterterrorism. The methodology could be transferred into the counterterrorism and law enforcement domains if it proves effective.

Robot Platoon Command and Control

Alan Christiansen, Principal Investigator

Washington only

Problem
Reliable autonomous soldier robot teams will not be possible for many years. However, an intermediate level of autonomy, where a commander gives high-level commands (e.g., go to the top of Hill 203), is achievable in the near future. This supervisory control requires only occasional intervention by a commander during a mission.

Objectives
This proposal asserts that one human is adequate for directing a small team of robots. We will use reconnaissance tasks in urban terrain as our test bed. Validating the assertion will require us to demonstrate a working team system where robots exhibit some automated reasoning (route planning, navigation) and cooperative behavior, while attending to human guidance.

Activities
We will extend behavior-based robotics approaches to include the memory and communication required for human participation in the team. Our principal demonstration task will be to produce a team entry for the RoboCup-Rescue annual competition. We will also investigate the utility of platform mobility for reconnaissance-directed sensor networks.

Impacts
MITRE's capability in robotics will be of considerable importance to our customers in the near future. This proposal builds on MITRE's current expertise in command and control and artificial intelligence. Robot platoon command and control defines a niche that is a natural extension of this expertise.

Social Information Retrieval

Raymond D'Amore, Principal Investigator

Washington only

Problem
Our research is focused on developing new technology for tracking internet-based networked organizations, and using those results to identify potential vulnerabilities and threats. Current information retrieval technology does not directly address the problem of detecting activist networks, assessing behavior, and tracking their evolution; new technology is needed to detect networks based on their structure and context.

Objectives
The main objective is to develop technology for a worldwide monitoring system used to detect the emergence of new groups (e.g., activists) and track the evolution of existing organizations based on their online activity. The focus will be on assessing an organization's behavior and its vulnerabilities.

Activities
We are exploring the confluence of information retrieval for collecting distributed information, social network analysis for determining network structure and characteristics, and dynamical systems modeling for determining network function or behavior. Work includes the development of advanced smart crawler collection tools that will use adaptive and cooperative searching techniques to provide efficient and high-coverage collection from the Web or other network search environments.

Impacts
This research will provide new tools for detecting emergent networked organizations in the open Web and enterprise environments, and will provide a basis for modeling their behavior, identifying critical nodes for assessing vulnerabilities and network robustness. Our initial work has already had impact on several sponsor mission areas.