Bioinspired framework for general-purpose learning
Learning applications in a robotic system are normally developed for specific areas and is difficult to generalize them for use in many other applications. Constraining the learning process to limited domains is good for fast development, but, unfortunately, what is gained in time is lost in generalization. Some basic principles of learning, valid for any area, and a framework where, what is learnt in a lower level application can be used in a higher level one, are shown here as a solution to this problem. First, some principles of the learning process, based on the statistical occurrence of inputs and outputs are explained. Additionally, a number of “accelerators” of the learning process are presented, which reduce and facilitate this process, and again are valid for all applications. Finally, a system simulation of a learning application is presented.
Unable to display preview. Download preview PDF.
- Laird, J.E., Rosenbloom, P.S. and Newell, A. “Chunking in SOAR: the anatomy of a general learning mechanism”. Machine Learning. Volumen/no: 1. pp. 11–47, 1986.Google Scholar
- Langley, P. and Carbonell, J.G. “Machine Learning, Techniques and Foundations”. Knowledge Engineering Master Documentation. Facultad Informática. Madrid. 1987.Google Scholar
- Lehman et al.: “Invitation to Cognitive Science” (A gentle introduction to SOAR, an architecture for human cognition). MIT Press, 1996.Google Scholar
- McFarland, D. and Bosser, T.: “Intelligent Behaviour in Animals and Robots”. Complex Adaptive Systems, MIT Press, 1993.Google Scholar
- Michalski, R.S. “Understanding the nature of learning: Issues and research directions”. Machine Learning: An Artificial Intelligence Approach. Volumen/no: 2Pag. 3–25. 1986.Google Scholar
- Michalski, R.S. “Toward a unified theory of learning: Multistrategy taskadaptative learning”. Readings in Knowledge Acquisition and Learning. Ed. Buchanan and Wilkins (Eds.). Morgan Kauffman. pp. 7–38. 1993.Google Scholar