Towards Closed Loop Information: Predictive Information
Bernd Porr, Alice Egerton & Florentin Wörgötter
Log in to download the full text for free
> Citation
> Similar
> References
> Add Comment
Abstract
Motivation: Classical definitions of information, such as the Shannon information, are designed for open loop systems because they define information on a channel which has an input and an output. The main motivation of this paper is to present a closed loop information measure which is compatible with constructivist thinking. Design: Our information measure for a closed loop system reflects how additional sensor inputs are utilised to establish additional sensor-motor loops during learning. Our information measure is based on the assumption that it is not optimal to stay reactive and that it is beneficial to become proactive through increased learning about the environment. Consequently our information measure gauges the utilisation of new sensor inputs to generate anticipatory actions. We call this information measure “predictive information” (PI). Findings: Our PI is zero if the organism uses only its reflex reactions. It grows when the organism is able to use other sensor inputs to preempt reflex reactions and is able to replace reflexes by anticipatory reactions. This has been demonstrated with a real robot that had to learn to avoid obstacles. Conclusion: PI is a new measure which is able to quantify anticipatory learning and, in contrast to the Shannon information, is calculated only at the inputs of an agent. This information measure has been successfully applied to a simple robot task but its application is neither limited to a certain task nor to a certain learning rule.
Key words: closed loop system, information measure, differential Hebbian learning, reactive vs proactive systems
Citation
Porr B., Egerton A. & Wörgötter F. (2006) Towards closed loop information: Predictive information. Constructivist Foundations 1(2): 83–90. http://constructivist.info/1/2/083
Export article citation data:
Plain Text ·
BibTex ·
EndNote ·
Reference Manager (RIS)
Similar articles
References
Atmanspacher H. & Dalenoort G. (1994) Introduction. In: Atmanspacher H. & Dalenoort G. (eds.) Inside versus outside. Springer, Berlin: 1–12.
▸︎ Google︎ Scholar
D’Azzo J. J. (1988) Linear control system analysis and design. McGraw, New York.
▸︎ Google︎ Scholar
Foerster H. von (1960) On self-organizing systems and their environments. In: Yovits M. & Cameron S. (eds.) Self-organizing systems. Pergamon Press, London: 31–50.
▸︎ Google︎ Scholar
Hebb D. O. (1949) The organization of behavior: A neurophychological study. Wiley-Interscience, New York.
▸︎ Google︎ Scholar
Klyubin A. S., Polani D. & Nehaniv C. L. (2004) Organization of the information flow in the perception–action loop of evolved agents. In: Proceedings of 2004 NASA/DoD Conference on Evolvable Hardware. IEEE Computer Society: Seattle, Washington: 177–180.
▸︎ Google︎ Scholar
Kosco B. (1986) Differential hebbian learning. In: Denker J. S. (ed.) Neural Networks for computing. Volume 151 of AIP conference proceedings. American Institute of Physics, New York: 277–282.
▸︎ Google︎ Scholar
Linsker R. (1988) Self-organisation in a perceptual network. Computer 21(3): 105–117.
▸︎ Google︎ Scholar
Oja E. (1982) A simplified neuron model as a principal component analyzer. Journal of Mathematical Biology 15(3): 267–273.
▸︎ Google︎ Scholar
Palm W. J. (2000) Modeling, analysis and control of dynamic systems. Wiley, New York.
▸︎ Google︎ Scholar
Porr B. & Wörgötter F. (2005) Non-hebbian synaptic plasticity allows for the stable implementation of one-shot predicitive learning. In: Zimmermann H. & Krieglstein K. (eds.) Proceedings of the 6th Meeting of the German Neuroscience Society, 30th Göttingen Neurobiology Conference 2005, p. 1223.
▸︎ Google︎ Scholar
Porr B. & Wörgötter F. (2006) Strongly improved stability and faster convergence of temporal sequence learning by utilising input correlations only. Neural Computation 18(in press).
▸︎ Google︎ Scholar
Porr B. (2002) Systemtheorie und Naturwissenschaft Eine interdisziplinäre Analyse von Niklas Luhmanns Werk. Deutscher Universitäts-Verlag: Wiesbaden.
▸︎ Google︎ Scholar
Porr B., von Ferber C. & Wörgötter F. (2003) Iso-learning approximates a solution to the inverse-controller problem in an unsupervised behavioural paradigm. Computation 15: 865–884.
▸︎ Google︎ Scholar
Rieke F., Warland D., de Ruyter van Stevenick R. & Bialek W. (1997) Spikes: Exploring the neural code. MIT Press, Cambridge MA.
▸︎ Google︎ Scholar
Sutton R. (1988) Learning to predict by method of temporal differences. Machine Learning 3(1): 9–44.
▸︎ Google︎ Scholar
Touchette H. & Lloyd S. (2004) Information-theoretic approach to the study of control systems. Physica A 331: 140–172.
▸︎ Google︎ Scholar
Verschure P. & Coolen A. (1991) Adaptive fields: Distributed representations of classically conditioned associations. Network 2: 189–206.
▸︎ Google︎ Scholar
Comments: 0
To stay informed about comments to this publication and post comments yourself, please log in first.