Surprised-based learning may improve predictions Published June 29, 2006 By William J. Sharp Air Force Office of Scientific Research Public Affairs ARLINGTON, Va. -- Featuring a team of researchers from the University of Southern California, the Air Force Office of Scientific Research here recently began funding a new research area that includes a study in improving the process prediction for defense planners. The new area, formally launched at a meeting with AFOSR principal investigators at Syracuse University, Syracuse, N.Y., is entitled Information Forensics and Process Integration. The overall portfolio of projects consists of three areas of research emphasis – incomplete information and metrics; search, interactive design, and active querying; and cognitive processing. Dr. Wei-min Shen, director, Polymorphic Robotics Lab, University of Southern California, Marina del Rey, Calif., is the principal investigator for the USC team. The team’s area of focus is in the area of incomplete information and forensics. The team’s estimated $600,000, 3-year project is entitled “Surprised Based Learning.” “The goal of the project is to model information networks,” Dr. Shen said. “We want users to be able to predict, analyze, manage, and control network-centric warfare effectively and efficiently. We want to construct models from whatever information is available at the outset and then make continuous improvements to the models based upon changes in the environment.” For example, Dr. Shen’s team envisions information analysts able to simultaneously access multiple information systems and have those systems communicate with one another. These systems could include radar, satellites, weapons, etc. Collaborative communication among several systems can lead to better decisions on the battlefield. “We want to better predict surprises in complex operating systems,” Dr. Shen said. “The better we can understand surprise, the better we can improve on our ability to analyze complex information and have our model improve continuously. Dr. Shen’s team wants to quantify and control surprise for information analysts so it does not occur at a magnitude that is disruptive to analyst operations. Information analysts feel more comfortable when information is predictable. Dr. Shen wants to explore levels of comfort sustainable for analysts especially when information analyses do not go as planned or are disrupted. “Every surprise contains critical information for modeling the environment that is new to the observer. This idea is intuitively simple and attractive but requires rigid investigation and mathematical formalism,” Dr. Shen said. “Unexpected information can come from a variety of sources such as computer malfunctions,” he added. “Other unexpected events can happen when adversaries get involved with friendly force operations. “What we want the analyst to better understand is information not directly observable to the analyst may be just as valuable as the information that is exposed,” Dr. Shen said. He said surprise modeling has many potential applications such as equipment operation, social organization analysis, and human learning techniques.