SIFT has gathered extensive experience in multiple areas in the field of artificial intelligence. Automatic intent recognition, also referred to as plan recognition, is a critical challenge for intelligent user interfaces (to determine what a user is trying to do, and how the UI can offer help), computer security (to determine the objectives of an attacker), sketch understanding, natural language understanding, and other contemporary problems. SIFT researchers pioneered the use of Bayesian probabilistic methods in intent recognition. In cooperation with University researchers, SIFT has developed the Yappr system for intent recognition that achieves new levels of computational efficiency by using sophisticated techniques based on parsing.
Another prominent research area for SIFT is Automated Planning. SIFT researchers have used hierarchical task network (HTN) planning to develop a shared human/automation task model in Playbook® systems. As part of this effort, SIFT has led the maintenance of the open source SHOP2 HTN planner, originally developed by Dana Nau's research group at the University of Maryland. SIFT researchers have applied HTN planning in many areas including planning for unmanned air vehicles (UAVs), automatically composing semantic web services into workflows, etc.
SIFT researchers have been active in the area where planning, control theory and automatic verification intersect. Our work on the CIRCA system involves automatically planning (synthesizing) control programs that provide hard real-time performance guarantees, allowing them to be used in mission-critical applications. The CIRCA technology also allows non real-time computation techniques (planning, search, optimization) to be interleaved with real-time computations in an integrated system. SIFT researchers have applied the CIRCA technology to work on both deterministic and stochastic systems, in areas including UAV control and satellite and cyber-defense.
SIFT has utilized its core technologies of machine learning, planning and human interfaces in the context of cyber security. For example, SIFT has applied intent recognition techniques to the problem of identifying the objectives of cyber attackers in order to distinguish serious threats from "ankle biters." Projects in this area have included applying qualitative probabilistic reasoning to network intrusion detection, helping security analysts tame the floods of false positives generated by conventional intrusion detection systems, applying planning techniques to coordinate cyber defense across large enterprises, and applying machine learning to actively deceive cyber attackers so that they reveal facts about their true identity.
SIFT has developed a core body of research capabilities in etiquette models and their applications. SIFT has been developing the concept of computational representations of human etiquette under internal research and development funds since 2001.
SIFT researchers have been applying our core technologies and skills, such as etiquette modeling, human-centered interaction design, usability assessment and neurophysiological sensing and interpretation, to innovative applications in medical and health care fields. Our projects and demonstrations are among the most creative work we have done, necessarily bringing together hard science, creative design and deep-knowledge of human needs and capabilities – all in the service of improving patient care and quality of life.
SIFT researchers have developed novel tools to improve information flow between the system and the user, utilizing more than 20 years of experience in UI design and management. We have employed various techniques including designing adaptable interfaces, using ecological design techniques to improve the user’s situational awareness, and constructing computational reasoning approaches to UI design.
SIFT’s work in automated interface evaluation tools has collectively evolved, evaluated and demonstrated a representation and reasoning approach with more power and a far greater pedigree than most. Our approach in this field offers a rich description of the information needs of user tasks, of the information provided by displays and, through computational reasoning over these descriptions, the ability to quantify when, how and to what degree UI meets the needs of users. Much, but not all, of our work in computational reasoning approaches has been done for military aircraft, including the U.S. Army’s Rotorcraft Pilot’s Associate (RPA). In RPA, we demonstrated one of the few automated interface reasoning systems yet to operate and provide benefit in a live flight test vehicle. We have also applied these techniques to displays for military command and control, oil refinery operations, automobile and truck driving, building security and comfort management and coordinated aiding and monitoring devices for home care for the elderly.
SIFT researchers pioneered a human-automation integration architecture called Playbook®, in which humans and automation share a task model – the possible ways to achieve an outcome, and what everyone is supposed to do in the play. The user can select - either at a very high level, or if time permits and circumstances require, at much finer levels – how the tasks will be performed, and by whom. Thus, Playbook allows precise control over delegation of tasks to automation. We maintain that Playbook can be used as a testbed for examining different approaches to supervisory control.