J. David Schaffer
ABSTRACT
One of the "black arts" of evolutionary computation is the design of effective fitness functions. For some tasks the appropriate function is easy to identify, but many times the "obvious" approach induces unforeseen failures of the evolutionary process to discover genomes with the desired properties. We present a series of fitness functions we have tried on the task of evolving spiking neural networks. The paradigm is to compare the output spike trains produced by evolving networks to provided target spike trains. The initial attempts failed dramatically, and subsequent versions revealed new failure modes until the third version which seems to be yielding better performance. We close with some speculations on possible limitations to this approach.
- Eshelman, L. The CHC Adaptive Search Algorithm: How to have safe search while engaging in nontraditional genetic recombination, Foundations of Genetic Algorithms 1, pp. 265--283, Morgan Kaufmann, San Francisco, 1991.Google Scholar
Cross Ref
- Gerstner, W. and Kistler, W.M., Spiking neuron models -- single neurons, populations, plasticity, Cambridge University Press, August 2002. Google ScholarDigital Library
- Markram, H., Wang, Y., Tsodyks, M., Differential signaling via the same axon of neocortical pyramidal neurons, PNAS, V 95, pp. 5323--5328, April 1998.Google ScholarCross Ref
- Mathias, K., Eshelman, L., and Schaffer, J.D., Niches in NK-Landscapes, Foundations of Genetic Algorithms 6, Morgan Kaufmann, San Francisco, CA, 2001.Google Scholar
- Sichtig, H., Schaffer, J.D., Laramee, C., SSNNS -- A suite of tools to explore spiking neural networks, Graduate student workshop, GECCO 2008. Google ScholarDigital Library
- Watts, L., Event-driven simulation of networks of spiking neurons. Advances in Neural Information, Processing Systems, 6:927 U934, 1994.Google Scholar
Index Terms
- A series of failed and partially successful fitness functions for evolving spiking neural networks
Recommendations
EvoNN: a customizable evolutionary neural network with heterogenous activation functions
GECCO '18: Proceedings of the Genetic and Evolutionary Computation Conference CompanionWhile some attention has been given to the process of using Evolutionary Computation (EC) to optimize the activation functions within hidden layers, available activation function sets have always been hard coded by the developers, and were immutable by ...
Fitness landscapes and graphs: multimodularity, ruggedness and neutrality
GECCO '09: Proceedings of the 11th Annual Conference Companion on Genetic and Evolutionary Computation Conference: Late Breaking PapersThe performances of evolutionary algorithms (genetics algorithms, genetic programming, etc.) or local search algotihms (Simulated annealing, tabu search, etc.) depends on the properties of seach space structure. One concept to analyse the search space ...
Training spiking neural networks with a multi-agent evolutionary robotics framework
GECCO '21: Proceedings of the Genetic and Evolutionary Computation ConferenceWe demonstrate the training of Spiking Neural Networks (SNN) in a novel multi-agent Evolutionary Robotics (ER) framework inspired by competitive evolutionary environments in nature. The topology of a SNN along with morphological parameters of the bot it ...
Comments