Small world topology of dynamic reservoir for effective solution of memory guided tasks
-
1
Bernstein Centre for Computational Neuroscience, III Physikalisches Institut , Department of computational neuroscience, Germany
A large number of biological, technological and social networks, such as the neural network of C-Elegans, power-grid networks, citation networks and numerous cellular networks, exhibit a small-world topology.Inspired by this we propose a highly clustered neural topology with a short characteristic path length, for the dynamic reservoir of a standard Echo state network. The purpose is to generate a stable short term memory for delayed response tasks such as maze navigation. Characteristic to delayed response tasks, like the road sign problem and navigation through generic T-shaped maze, is the requirement of a temporal memory such that the system is capable of learning slowly and access previously learned pattern over varying time delays. For this purpose we use continuous time leaky integrator neurons with an experimentally determined leak decay rate coupled with the small world architecture. This mechanism controls the time scale of the internal dynamics of the ESN network.
Restricting our neural pool to 100 neurons we tested our setup initially with the task of learning a Mackey-Glass system with mild to wild chaotic behavior and finally with the task of off-line learning of navigation through a maze. For the maze task at each step the state-action pair (sensor-motor value) serves as the input to the network, while the output gives the required action which the agent needs to take for a required state. After initial training on a simple maze, we test the learning of the network by generalizing to a larger arena with varying starting and ending points. Results of the simulations on the time series data of the Mackey-glass dynamic system show higher prediction ability than standard ESN networks. The small-world model along with the ability to control the internal time scale of the dynamic reservoir in our setup, results in a stable temporal memory capable of handling varying delays to solve the navigation task with efficient generalization capability to more complex maze environments.
Acknowledgements
This research was supported by the Emmy Noether Program (DFG, MA4464/3-1) and the Bernstein Center for Computational Neuroscience II Goettingen (01GQ1005A, project D1).
References
1. Jaeger, H. Short term memory in echo state networks. GMD-Report 152, GMD - German National Research Institute for Computer Science (2002).
2. Watts, D. J. & Strogatz, S. H. Collective dynamics of 'small-world' networks. Nature 393, 440-442 (1998).
3. Fette, G. & Eggert, J. Short term memory and pattern matching with simple echo state networks. In Proc. ICANN, 2005.
Keywords:
Echo state networks,
maze navigation,
memory guided tasks,
recurrent neural networks,
reservoir computing,
sequence memory,
short term memory,
Small world topology
Conference:
BC11 : Computational Neuroscience & Neurotechnology Bernstein Conference & Neurex Annual Meeting 2011, Freiburg, Germany, 4 Oct - 6 Oct, 2011.
Presentation Type:
Poster
Topic:
neurons, networks and dynamical systems (please use "neurons, networks and dynamical systems" as keywords)
Citation:
Dasgupta
S,
Manoonpong
P and
Woergoetter
F
(2011). Small world topology of dynamic reservoir for effective solution of memory guided tasks.
Front. Comput. Neurosci.
Conference Abstract:
BC11 : Computational Neuroscience & Neurotechnology Bernstein Conference & Neurex Annual Meeting 2011.
doi: 10.3389/conf.fncom.2011.53.00177
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
22 Aug 2011;
Published Online:
04 Oct 2011.
*
Correspondence:
Mr. Sakyasingha Dasgupta, Bernstein Centre for Computational Neuroscience, III Physikalisches Institut, Department of computational neuroscience, Goettingen, Germany, sakya.dasgupta@gmail.com