Cognitive Systems

Navigation

Languages

Select Interface Language

Project-Homepage

Project: EANT / EANT2: Evolutionary Acquisition of Neural Topologies

Researchers: Siebel N. , Kassahun Y.

EANT2 ("Evolutionary Acquisition of Neural Topologies, Version 2") is a system for evolutionary reinforcement learning of artificial neural networks. It develops both the structure (topology) and the parameters (e.g. synaptic weights) of neural networks to create solutions for a given task. It has the following features:

The structural and parametrical development of the neural networks is clearly separated so as to enable the system to quickly develop useful network structures.
Neural Networks are encoded in a Linear Genome. Genes can represent neurons (with an arbitrary activation function), connections, biases etc. This encoding is compact and versatile; it facilitates manipulation and evaluation without decoding.
Parameter optimisation is done using CMA-ES, a derandomised variant of evolution strategies.
EANT2 requires no training data pairs for network training (as supervised/unsupervised/semi-supervised methods do) but instead uses Reinforcement Learning to adapt to the structure of the problem by means of a scalar fitness function.
There is no parameter tuning to a given problem; the method is designed to be as general as possible.

EANT2 is the name of the newer version of the algorithm which is based on the original version "EANT" by Yohannes Kassahun. More information on EANT and EANT2 as well as comparisons with different methods can be found on this page on evolutionary reinforcement learning and in the relevant references (see below).

Publications:

2009	Compiling neural networks for fast neuro-evolution Siebel N., Jordt A., Sommer G. In Proceedings of the 2nd International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems, St. Louis, USA, October 2009, pp. 23--29.	PDF , Bibtex , Abtract
2009	Efficient neural network pruning during neuro-evolution Siebel N., Boetel J., Sommer G. In Proceedings of the 2009 International Joint Conference on Neural Networks (IJCNN 2009), Atlanta, USA, pages 2920--2927, ISBN 987-1-4244-3553-1, June 2009.	PDF , Bibtex , Abtract
2008	Evolutionary learning of neural structures for visuo-motor control Siebel N., Sommer G., Kassahun Y. In Arpad Kelemen, Ajith Abraham and Yulan Liang (eds.), Computational Intelligence in Medical Informatics, pp. 93-115, ISBN 3-540-75766 -5, Springer-Verlag Berlin, 2008	PDF, PS , Bibtex , Abtract
2008	Learning defect classifiers for visual inspection images by neuro-evolution using weakly labelled training data Siebel N., Sommer G. In Proceedings of the IEEE Congress on Evolutionary Computation (CEC 2008), Hong Kong, China, pp. 3926--3932, ISBN 978-1-4244- 1823-7, June 2008.	PDF, PS , Bibtex , Abtract
2008	Creating edge detectors by evolutionary reinforcement learning Siebel N., Grünewald S., Sommer G. In Proceedings of the IEEE Congress on Evolutionary Computation (CEC 2008), Hong Kong, China, pp. 3552--3559, ISBN 978-1-4244- 1823-7, June 2008.	PDF, PS , Bibtex , Abtract
2007	Efficient learning of neural networks with evolutionary algorithms Siebel N., Krause J., Sommer G. In Pattern Recognition, F. Hamprecht, B. Jähne, C. Schnörr (Eds.), Heidelberg, LNCS, Vol. 4713, pp. 466-475, Springer-Verlag, Berlin. DOI: 10.1007/978-3-540-74936-3_47	PDF , Bibtex
2007	Self-organisation of neural topologies by evolutionary reinforcement learning Siebel N., Krause J., Sommer G. In Proceedings of the 6th International Workshop on Self-Organising Maps (WSOM 2007), Bielefeld, Germany, 7 pages (no page numbe rs), September 2007. DOI: 10.2390/biecoll-wsom2007-118.	PDF , Bibtex , Abtract
2007	Evolutionary reinforcement learning of artificial neural networks Siebel N., Sommer G. In International Journal of Hybrid Intelligent Systems (IJHIS), IOS Press, 4(3), pp. 171-183, October 2007.	PDF, PS , Bibtex , Abtract
2006	Learning neural networks for visual servoing using evolutionary methods Siebel N., Kassahun Y. In Proceedings of the 6th International Conference on Hybrid Intelligent Systems (HIS'06), Auckland, New Zealand, p. 6 (4 pages) , December 2006, ISBN 0-7695-2662-4, DOI: 10.1109/HIS.2006.41.	PDF , Bibtex , Abtract
2006	Towards a Unified Approach to Learning and Adaptation Kassahun Y. Dissertation, Institut für Informatik und Praktische Mathematik, Christian-Albrechts-Universität zu Kiel, 2006.	PDF , Bibtex
2006	Towards a Unified Approach to Learning and Adaptation Kassahun Y. Technical Report 0602, Institut für Informatik und Praktische Mathematik, Christian-Albrechts-Universität zu Kiel, 2006.	PDF , Bibtex
2006	Evolutionary reinforcement learning for simulated locomotion of a robot with a two-link arm Kassahun Y., Sommer G. In Proc. of the 9th Conference on Intelligent Autonomous Systems IAS-9, March 7-9, T. Arai, R. Pfeifer, T. Balch and H. Yokoi (Eds.), pp.263-271, IOS Press, 2006.	PDF , Bibtex
2005	Efficient reinforcement learning through evolutionary acquisition of neural topologies Kassahun Y., Sommer G. In Proc. 13th European Symposium on Artificial Neural Networks, Bruges, Belgium, pp. 259-266, d-side publications, April 2005.	PDF , Bibtex , Abtract
2005	Efficient reinforcement learning through evolutionary acquisition of neural topologies Kassahun Y., Sommer G. In Proc. 13th European Symposium on Artificial Neural Networks, Bruges, Belgium, pp. 259-266, d-side publications, April 2005.	PDF , Bibtex , Abtract
2005	Evolution of Neural Networks Through Incremental Acquisition of Neural Structures Kassahun Y., Sommer G. Technical Report 0508, Christian-Albrechts-Universität zu Kiel, Institut für Informatik und Praktische Mathematik, Juni 2005.	PDF , Bibtex
2005	Automatic neural robot controller design using evolutionary acquisition of neural topologies Kassahun Y., Sommer G. In Proc. 19. Fachgespräch Autonome Mobile Systeme (AMS2005), Stuttgart, pp. 315-321, Springer-Verlag, 2005	PDF , Bibtex