Learning by Observation

• J. Vargas, R. F. DeMara, A. J. Gonzalez, M. Georgiopoulos, and H. Marshall, "PDU Bundling and Replication for Reduction of Distributed Simulation Communication Traffic," Journal of Defense Modeling and Simulation, Vol. 1, No. 3, August, 2004, pp. 167 * 185.
  
  [ FULL DOCUMENT ] (Full document will be loaded onto a new page.)
  
  
• A. E. Henninger, A. J. Gonzalez, M. Georgiopoulos, and R. F. DeMara, "A Connectionist-Symbolic Approach to Modeling Agents: Neural Networks Grouped by Contexts," Proceedings of the Third International and Interdisciplinary Conference on Modeling and Using Context (CONTEXT'01), pp. 198 * 209, Dundee Scotland, July 26 * 29, 2001.
  
  [ FULL DOCUMENT ] (Full document will be loaded onto a new page.)
  
  
• A. E. Henninger, A. J. Gonzalez, M. Georgiopoulos, and R. F. DeMara, "The Limitations of Static Performance Metrics for Dynamic Tasks Learned Through Observation," Proceedings of the Tenth Conference on Computer Generated Forces and Behavioral Representation (CGF-BR'01), pp. 147 * 154, Norfolk, Virginia, U.S.A., May 14 * 17, 2001.
  
  [ FULL DOCUMENT ] (Full document will be loaded onto a new page.)
  
  
• A. E. Henninger, A. J. Gonzalez, W. Gerber, M. Georgiopoulos, and R. F. DeMara, "On the Fidelity of SAFs: Can Performance Data Help?" Proceedings of the 2000 Interservice/Industry Training, Simulation and Education Conference (I/ITSEC-2000), pp. 147 * 154, Orlando, Florida, U.S.A., November 27 * 30, 2000.
  
  [ FULL DOCUMENT ] (Full document will be loaded onto a new page.)
  
  
• Henninger, A., Gonzalez, A.J., Georgiopoulos, M., and DeMara, R.F., "Modelling Semi-automated Forces with Neural Networks: Performance Improvement through a Modular Approach," Computer Generated Forces and Behavior Representation, 2000.
  
  [

  ABSTRACT

  | FULL DOCUMENT ] (Full document will be loaded onto a new page.)

  
  A recent report by the National Research Council (NRC) declares neural networks “hold the most promise for providing powerful learning models”. While some researchers have experimented with using neural networks to model battlefield behavior for Computer Generated Forces (CGF) systems used in distributed simulations, the NRC report indicates that further research is needed to develop a hybrid system that will integrate the newer neural network technology into the current rule-based paradigms. This paper supports this solicitation by examining the use of a context structure to modularly organize the application of neural networks to a low-level Semi-Automated Forces (SAF) reactive task. Specifically, it reports on the development of a neural network movement model and illustrates how its performance is improved through the use of the modular context paradigm. Further, this paper introduces the theory behind the neural networks’ architecture and training algorithms as well as the specifics of how the networks were developed for this investigation. Lastly, it illustrates how the networks were integrated with SAF software, defines the networks’ performance measures, presents the results of the scenarios considered in this investigation, and offers directions for future work.
  
  
• Gonzalez, A.J., and et al., "Automating the CGF Model Development and Refinement Process by Observing Expert Behavior in a Simulation," Computer Generated Forces and Behavior Representation, 1998.
  
  [

  ABSTRACT

  | FULL DOCUMENT ] (Full document will be loaded onto a new page.)

  
  Ultimate widespread use of CGF entities in tactical simulations will depend on how easy it will be to develop, refine and maintain models of the behaviors to be represented. However, proper vehicle behavior model development for CGF applications can be difficult as well as expensive. Means to quickly and effectively create models for new vehicles and/or behaviors must be developed to permit CGF models to be widely used in the future. One realistic approach to overcoming this model generation bottleneck is to create and refine vehicle models through automated observation of the behavior of an entity being controlled by a human expert in a simulation. This is a learning paradigm quite commonly used by humans. This paper describes an on-going research effort that introduces some new ideas on how to accomplish autonomous model development through observation.
  
  
• Gonzalez, A.J., DeMara, R.F., and Georgiopoulos, M., "Vehicle Model Generation and Optimization for Embedded Simulation," Simulator Interoperability Workshop, 1998.
  
  [

  ABSTRACT

  | FULL DOCUMENT ] (Full document will be loaded onto a new page.)

  
  The development and use of Inter-Vehicle Embedded Simulation Technology (INVEST) offers several distinct advantages for 21 st Century training environments. Benefits of embedded simulations include the ability to perform “in-situ” exercises on actual equipment, more direct provision of support for the wide array of equipment in the field, and a greater opportunity to develop new training exercises using much shorter lead times than previously possible with stand-alone training systems. Nonetheless, the INVEST program also presents several challenging, yet surmountable obstacles to interconnecting the real and virtual layers within their actual environments. These challenges include 1) how to more efficiently and effectively create, refine and maintain vehicle models, and 2) how to optimize the operation of these models within the INVEST environment, so as to best utilize the computing resources available on board the vehicle and the available communication bandwidth. Addressing these technology challenges remains a prerequisite to the feasibility of realistic INVEST simulations involving ground combat vehicles. In this paper, we describe a multifaceted investigation aimed at addressing these tasks.
  
  
• Sidani, T.A., and Gonzalez, A.J., "Learning Situational Awareness by Observing Expert Actions," Florida A.I. Research Society Conference, 1995.
  
  [

  ABSTRACT

  | FULL DOCUMENT ] (Full document will be loaded onto a new page.)

  
  The aim of many research projects in the field of artificial intelligence entail the incorporation of human-like intelligent behavior in the computer. The represented level of intelligence is highly dependent upon the amount of knowledge possessed by experts in the field and the efficiency and effectiveness of transferring this expertise from man to machine. This knowledge acquisition process is difficult and time consuming. Furthermore, most of the current knowledge acquisition techniques focus on gathering information about a domain which is static in nature. They capture explicit static information which does not vary over time and is easy to mold into a symbolic representation. Most real-life situations, however, involve dynamically changing information and require a non-symbolic form of representation. Current knowledge acquisition techniques are not well suited for the extraction of implicit expert knowledge while handling a situation in a dynamic environment.
  
  This research describes a general methodology for learning implicit situational knowledge by observing the expert while reacting to a real-time simulation. The paper outlines the IASKNOT system methodology which gathers, represents, and learns expert knowledge by examining the expert's simulated surroundings while simultaneously monitoring the expert's actions for a given situation. It utilizes recent advances in the areas of neural networks and artificial intelligence. The method demonstrates the ability to train on basic skills and to generalize learned actions to handle more complex situations not previously encountered. It was implemented and tested for handling specific situations in the driving domain.