UZH ETH UZH|ETH
  Zurich

Skip to content. | Skip to navigation

Sections
Personal tools
You are here: Home Projects EvoSpike Methods and Results

Methods and Results

by Nikola Kasabov last modified Nov 17, 2012 11:38 AM
Research questions addressed and results produced by the project

Methods

The project creates evolving probabilistic SNN (epSNN) models and systems for SSTPR. The epSNN are built on the principles of evolving connectionist systems [1], of eSNN [1], [2], [3] and of probabilistic neuronal models [4], previously developed by prof. Kasabov’s group at KEDRI. The latter extend the widely used leaky integrate-and-fire spiking model by introducing some biologically plausible probabilistic parameters [4]. The epSNN are evolving structures that learn from and adapt to new incoming data. The research explores a range of approaches to creating epSNN for STPR, from a single neuron to ‘reservoir’ computing and neuro-genetic systems.

Novel epSNN models and systems for STPR

A single neuron is a complex information processing machine. A single neuronal model, namely Spike Pattern-Association Neuron SPAN [5], [6], [7], has been developed as part of this project. SPAN can be trained to capture SSTD patterns of hundreds and thousands of input spike trains and to generate in response a precise time spike sequence. Mutiple SPAN networks can be used for classification of complex spatial patterns through generating spike trains at different times on differnt output neurons. This model can be potentially used for neurposthetics.

Apart from the eSNN [1], [2] and SPAN [5], [6], [7], the project explores other neuronal models and dynamic synapses, namely Fusi’s model [8] implemented on the INI Zurich (www.ini.unizh.ch) SNN chip [9]. Two novel models have been developed that combine eSNN and SPAN with the dynamic synapse model from [8]. Improved accuracy for STPR have been achieved when address-event representation (AER) was used [10], [11]. AER is implemented in the INI silicon retina chip and the silicon cochlea chip.

The research explores further ensembles of probabilistic neuronal models and recurrent deep learning connectionist structures (reservoirs). These structures capture correlated spatial and temporal components from incoming data. The epSNN can learn data in an on-line manner using a frame-based input information representation, or alternatively AER. Some preliminary experiments on gesture- and sign language recognition [12], moving object recognition [13], sound recognition, EEG data recognition [14] (fig.3) have been conducted.

Computational neurogenetic models

A major issue in the EvoSpike model and system development is how to optimize the numerous epSNN parameters. Here we combine local learning of synaptic plasticity with global optimisation of probability and other parameters. Three approaches are being investigated: evolutionary computation methods [12]; gene regulatory network (GRN) model [15], [16] – fig.4, or using both together [15], [16]. Linking gene/protein expression to epSNN parameters may also lead to new types of neuron-synapse-astrocyte models inspired by new findings in neuroscience. Neurogenetic models are promising for cognitive robotic systems and for the prognosis of neurodegenerative diseases such as Alzheimer’s disease and for personalized medicine [16]. The research is expected to contribute to the fast developing area of neuromorphic engineering [17], [18]. Future research is expected to continue through tighter integration of knowledge and methods from information science, bioinformatics and neuroinformatics [19], [20].

Results

EvoSpike Simulator

One outcome of this project is EvoSpikeSim. EvoSpikeSim is a collection of modules and functions written in Python language and using functions from the Brian library.

EvoSpikeSim includes modules and functions as follows:

  • Modules for converting continuous-value input data into spike trains;
  • Evolving spiking neural network (eSNN) models for spatio-temporal pattern recognition, including: rank-order coding eSNN; spike-time coding eSNN; eSNN with dynamic SDSP synapses; spike pattern association neuron (SPAN) and neural network models; models for classification; probabilistic eSNN models (epSNN); gene-regulatory network models for epSNN parameter optimization; reservoir computing models; other.
  • Functions for knowledge extraction from trained eSNN.
  • Functions for presenting results and for visualization of learning processes in the peSNN.
  • Functions for connecting software modules to neuromorphic SNN hardware realisations.

This clip demonstrates a brief session with a preliminary version of a visualization tool developed in KEDRI by Dr. Stefan Schliebs and Johannes Bopp. The tool visualizes the activity in time of 120,000 LIF neurons, connected with recurrent connections in a reservoir structure, when spatio-temporal data is entered. The example shows how and when the neurons fire when boxing movement data is entered. The reservoir has 30x40x100 neurons to capture moving images of 30x40 pixels over 100 time points.

EvoSpike Organized Events

The following events are organized as part of the EvoSpike project:

  1. NCEI 2012 - Neuro-Computing and Evolving Intelligence Workshop, Auckland, 8 June 2012 (for more details, see the `NCEI 2012 flyer at http://www.kedri.info)

    System Message: WARNING/2 (<string>, line 47); backlink

    Inline interpreted text or phrase reference start-string without end-string.

  2. Special session at the WCCI - World Congress of Computational Intelligence, 10-15 June 2012. (see WCCI PDF flyer for more details)

The two events disseminate results of the EvoSpike project and include international participation. They are organised jointly by KEDRI (www.kedri.info) and INI.

Invited/plenary talks with dissemination of EvoSpike results:

System Message: WARNING/2 (<string>, line 55)

Title underline too short.

Invited/plenary talks with dissemination of  EvoSpike results:
-------------------------------------------------------------
  • CIBB2011(Gargnano, Italy);
  • Workshop IJCNN 2011(San Jose);
  • Irish CICS 2011(Derry, UK);
  • EANN2011 (Corfu);
  • ICONIP2011 (Shanghai);
  • WCCI2012 (Brisbane),
  • IEEE IS 2012 (Sofia),
  • ANNPR 2012 (Trento, Italy)
  • EANN 2012 (London)
  • ICONIP 2012 (Qatar)

System Message: WARNING/2 (<string>, line 67)

Bullet list ends without a blank line; unexpected unindent.

Publications resulted from the project

Some prior EvoSpike related publications (published before 2011) and current publications resulted from this project are listed here

References

[1](1, 2, 3) N.Kasabov, Evolving connectionist systems: The knowledge engineering approach, Springer, 2007 (first edition 2002).
[2](1, 2) S.Wysoski, L.Benuskova, N.Kasabov, Evolving spiking neural networks for audiovisual information processing, Neural Networks, vol 23, 7, pp 819-835, 2010.
[3]L.Benuskova and N.Kasabov, Computational neuro-genetic modelling, Springer, New York, 2007, 290 pages
[4](1, 2) N.Kasabov, To spike or not to spike: A probabilistic spiking neuron model, Neural Networks, 23(1), 16–19, 2010.
[5](1, 2) Mohemmed, A. and Schliebs, S. and Matsuda, S. and Kasabov, N. Method for training a spiking neuron to associate input-output spike trains, Proceedings of the EANN/AIAI 2011, Part I, IFIP AICT 363, 219-228, 2011 (http://www.springer.com/computer/theoretical+computer+science/book/978-3-642-23956-4?changeHeader)
[6](1, 2) A.Mohemmed,S.Schliebs,S.Matsuda,Kasabov(2011) SPAN: Spike Pattern Association Neuron for Learning Spatio-Temporal Sequences, Int. J. Neural Systems, in print 2012.
[7](1, 2) A.Mohemmed,S.Schliebs,S.Matsuda,N.Kasabov, Spike Pattern Association Neuronal Networks for Learning Spatio-Temporal Sequences, Neurocomputing, accepted for publication in 2012.
[8](1, 2) S.Fusi et al, Spike driven synaptic plasticity: Theory, simulation, VLSI implementation (2000), Neural Computation, 12, 2227-2258
[9]G.Indiveri et al, Neuromorphic silicon neuron circuits, Frontiers in neuroscience, vol.5, 1-23, May 2011.
[10]Nuntalid, N. and Dhoble, K. and Kasabov, N. EEG Classification with BSA Spike Encoding Algorithm and Evolving Probabilistic Spiking Neural Network, Neural Information Processing, Proceedings of the 18th International Conference on Neural Information Processing, ICONIP, 2011, Shanghai, China, Springer, Heidelberg, LNCS vol. 7062, 451-460.

System Message: WARNING/2 (<string>, line 96)

Explicit markup ends without a blank line; unexpected unindent.

(http://www.springer.com/generic/search/results?SGWID=5-40109-24-653415-0&sortOrder=relevance&searchType=EASY_CDA&searchScope=editions&queryText=iconip+2011)

[11]Dhoble, K., N. Nuntalid, G. Indivery and N.Kasabov, On-line Spatiotemporal Pattern Recognition with Evolving Spiking Neural Networks utilising Address Event Representation, Rank Oder- and Temporal Spike Learning, WCCI 2012 IEEE World Congress on Computational Intelligence, June, 10-15, 2012 - Brisbane, Australia, 554-560
[12](1, 2) Schliebs, S. and Hamed, H. N. A. and Kasabov, N. A reservoir-based evolving spiking neural network for on-line spatio-temporal pattern learning and recognition, Neural Information Processing, Proceedings of the 18th International Conference on Neural Information Processing, ICONIP, 2011, Shanghai, China, Springer, Heidelberg, LNCS vol. 7063, pp.160-168

(http://www.springer.com/generic/search/results?SGWID=5-40109-24-653415-0&sortOrder=relevance&searchType=EASY_CDA&searchScope=editions&queryText=iconip+2011)

[13]Kasabov, N. and Dhoble, K. and Nuntalid, N. and Mohemmed, A. Evolving probabilistic spiking neural networks for spatio-temporal pattern recognition: A preliminary study on moving object recognition, Neural Information Processing, Proceedings of the 18th International Conference on Neural Information Processing, ICONIP, 2011, Shanghai, China, Springer, Heidelberg, LNCS vol. 7064, 230-239.

System Message: WARNING/2 (<string>, line 107)

Explicit markup ends without a blank line; unexpected unindent.

(http://www.springer.com/generic/search/results?SGWID=5-40109-24-653415-0&sortOrder=relevance&searchType=EASY_CDA&searchScope=editions&queryText=iconip+2011)

[14]N.Nuntalid, N.Kasabov, EEG spatio/spectro temporal pattern recognition with evolving probabilistic spiking neural networks, IEEE Trans. Neural Networks, submitted, 2012.
[15](1, 2) N.Kasabov, A.Mohhemed, S.Schliebs (2011) Modelling the Effect of Genes on the Dynamics of Probabilistic Spiking Neural Networks for Computational Neurogenetic Modelling, Proc. CIBB 2011, Springer, LNBI
[16](1, 2, 3) Kasabov, N. R.Schliebs, H.Kojima Probabilistic Computational Neurogenetic Framework: From Modelling Cognitive Systems to Alzheimer’s Disease, IEEE Transactions of Autonomous Mental Development, 3:(4) 300-3011, 2011 (http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=6097099&punumber=4563672)
[17]G.Indiveri and T.Horiuchi (2011) Frontiers in Neuromorphic Engineering, Frontiers in Neuroscience, 5:118.
[18]Kasabov, N., Dhoble, K., Nuntalid, N. and G. Indiveri, On-line Spatio- and Spectro-Temporal Pattern Recognition with Evolving Spiking Neural Networks utilising Integrated Rank Oder- and Spike-Time Learning, Neural Networks, submitted, 2012
[19]N.Kasabov (ed) The Springer Handbook of Bio- and Neuroinformatics, Springer, 2012, in print
[20]
  1. Kasabov, Evolving Spiking Neural Networks and Neurogenetic Systems for Spatio- and Spectro-Temporal Data Modelling and Pattern Recognition, Springer-Verlag Berlin Heidelberg 2012, J. Liu et al. (Eds.): IEEE WCCI 2012, LNCS 7311, pp. 234–260, 2012.