Lamarckian Evolution and the Baldwin Effect in Evolutionary Neural Networks
By P. Valdivieso, M. G. Arenas, Francisco Javier García Castellano, J. J. M. Guervós, A. Prieto, Víctor Manuel Rivas Santos, G. Romero. 2006
22 citations recorded by the source corpus.
Hybrid neuro-evolutionary algorithms may be inspired on Darwinian or Lamarckian evolu- tion. In the case of Darwinian evolution, the Baldwin effect, that is, the progressive incorporation of learned characteristics to the genotypes, can be observed and leveraged to improve the search. The purpose of this paper is to carry out an exper- imental study into how learning can improve G-Prop genetic search. Two ways of combining learning and genetic search are explored: one exploits the Baldwin effect, while the other uses a Lamarckian strategy. Our experiments show that using a Lamarckian op- erator makes the algorithm find networks with a low error rate, and the smallest size, while using the Bald- win effect obtains MLPs with the smallest error rate, and a larger size, taking longer to reach a solution. Both approaches obtain a lower average error than other BP-based algorithms like RPROP, other evolu- tionary methods and fuzzy logic based methods
Subfields: Artificial Intelligence
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Yoshua Bengio
- Pieter Abbeel
- Quoc V. Le
- Trevor Darrell
- Ruslan Salakhutdinov
- Mohit Bansal
- Chelsea Finn
Citations in the corpus
- Population Based Training of Neural Networks (2017, background): “Other similar works to ours are Lamarckian evolutionary algorithms in which parameters are inherited whilst hyperparameters are evolved (Castillo et al., 2006) or where the hyperparameters, initial weights, and architecture of networks are evolved, but the parameters are learned (Castillo et al.,…”
- Meta-learning by the Baldwin effect (2018, background): “As in earlier work we evolve inductive bias in the form of the initial parameters P and hyperparameters h of a learning algorithm [1, 6] but in addition provide a way to evolve agents for few-shot data-efficient learning on a task distribution.”