The Meta-AQUA system implements a theory of introspective multistrategy learning by providing a computational realization of the concepts as described in Cox (1996; Cox & Ram, 1999).It is an introspective learning system that chooses and combines multiple learning methods from a toolbox of algorithms in order to repair faulty components responsible for failures encountered during the system's performance task. The program incorporates an reflective version of the AQUA (Ram, 1991, 1993, 1994) story-understanding system as the performance task from which learning can take place. As a front end module to the performance system, a specially modified version of the Tale-Spin (Meehan, 1981) story-generation program automatically produces input data. At the back end, the UM Nonlin planning system (Ghosh et al., 1992) creates a learning plan designed to improve the performance. An extensive frame system (Minsky, 1975; Wilensky, 1986) was built to provide the formalism with which to represent the system's knowledge, both of the domain and of itself. This knowledge is stored in and retrieved from a simple indexed memory. The memory is partitioned into a working memory (FK) and a long-term store (BK).

The system architecture and flow of information within Meta-AQUA is shown below.

The problem generation module outputs a story to the performance system with the initial goal to understand the input (i.e., build a coherent conceptual interpretation). The story understanding system uses schemas from the BK to build a representation of the story in the FK. If this task fails, then a trace of the reasoning that preceded the failure is passed to the learning subsystem. A CBR subsystem within the learner uses past cases of introspective reasoning from the BK to explain the failure and to generate a set of learning goals. These goals, along with the trace, are then passed to a nonlinear planner. The planner subsequently builds a learning strategy from its toolbox of learning methods. The learning plan is then passed to an execution system that examines and changes items in the BK. These changes enable improved performance in subsequent processing.

Links:

• The Introspection Home Page
• SMIRKS - A Meta-AQUA Enhancement
• INTRO - A Meta-AQUA + Prodigy/Agent Integration
  
• Meta-AQUA, Version 6, code release public download
  
• Meta-AQUA Publications

References

Ghosh, S., Hendler, J., Kambhampati, S., & Kettler, B. (1992). UM Nonlin [a Common Lisp implementation of A. Tate's Nonlin planner]. Available FTP: Hostname: cs.umd.edu Directory: /pub/nonlin Files: nonlin-files.tar.Z

Meehan, J. (1981). Talespin. In R. C. Schank & C. Riesbeck (Eds.), Inside computer understanding: Five programs plus miniatures (pp. 197-258). Hillsdale, NJ: Lawrence Erlbaum Associates.

Minsky, M. L. (1975). A framework for representing knowledge. In P. H. Winston (Ed.), The psychology of computer vision (pp. 211-277). New York: McGraw Hill.

Ram, A. (1991). A theory of questions and question asking. Journal of the Learning Sciences, 1, (3&4), 273-318.

Ram, A. (1993). Indexing, elaboration and refinement: Incremental learning of explanatory cases. Machine Learning, 10, 201-248.

Ram, A. (1994). AQUA: Questions that drive the understanding process. In R. C. Schank, A. Kass, & C. K. Riesbeck (Eds.), Inside case-based explanation (pp. 207-261). Hillsdale, NJ: Lawrence Erlbaum Associates.

Wilensky, R. (1986). Some problems and proposals for knowledge representation (Tech. Rep. No. UCB/CSD 86/294). Berkeley, CA: University of California.