Relative time scales in the self-organization of pattern classification: From “one-shot” to statistical learning
We propose a biologically plausible learning scheme which enables a system to classify patterns based on the presentation of one single example. During a learning mode, the system recognizes whether a category for a presented pattern has been instantiated before, or whether it must be classified as unknown. In this case a new category is created autonomously. The proposed “one-shot” learning rules are characterized by certain time scale relations between system parameter dynamics and input dynamics. We show that reversing these relations (leading to a statistical learning regime, the learning dynamics can be reduced to a Kohonen learning scheme. Our results show that both “one-shot” and statistical learning in biological systems might be governed by identical laws.
Unable to display preview. Download preview PDF.
- 1.S. Amari. Dynamics of pattern formation in lateral-inhibition type neural fields. Biol. Cybern., 27:77–87, 1977.Google Scholar
- 2.G.A. Carpenter and S. Grossberg. Art2: Stable self-organization of pattern recognition codes for analogue input, patterns. Appl. Opt., 26:4919–4930, 1987.Google Scholar
- 3.J Hertz, A Krogh, and RG Palmer. Introduction to the Theory of Neural Computation. Addison-Wesley, 1991.Google Scholar
- 4.R. Mükkulainen. Self-organizing process based on lateral inhibition and synaptic resource distribution. In T. Kohonen, K. Mäkisara, O. Simula, and J. Kangas, editors, Artificial Neural Networks, pages 415–420. Elsevier, North-Holland, 1991.Google Scholar
- 5.H Ritter and K Schulten. On the stationary state of kohonen's self-organizing sensory mapping. Biol. Cybern., 54:99–106, 1986.Google Scholar