What does it take to evolve behaviorally complex organisms?

Abstract

What genotypic features explain the evolvability of organisms that have to accomplish many different tasks? The genotype of behaviorally complex organisms may be more likely to encode modular neural architectures because neural modules dedicated to distinct tasks avoid neural interference, i.e., the arrival of conflicting messages for changing the value of connection weights during learning. However, if the connection weights for the various modules are genetically inherited, this raises the problem of genetic linkage: favorable mutations may fall on one portion of the genotype encoding one neural module and unfavorable mutations on another portion encoding another module. We show that this can prevent the genotype from reaching an adaptive optimum. This effect is different from other linkage effects described in the literature and we argue that it represents a new class of genetic constraints. Using simulations we show that sexual reproduction can alleviate the problem of genetic linkage by recombining separate modules all of which incorporate either favorable or unfavorable mutations. We speculate that this effect may contribute to the taxonomic prevalence of sexual reproduction among higher organisms . In addition to sexual recombination, the problem of genetic linkage for behaviorally complex organisms may be mitigated by entrusting evolution with the task of finding appropriate modular architectures and learning with the task of finding the appropriate connection weights for these architectures.

Learning what to ignore: Memetic climbing in topology and weight space

“…tasks in human neurobiology, simultaneously. They found that for this to work, the two tasks needed to be performed by largely separate networks, otherwise networks would evolve that could solve only one of the tasks [9]. Similarly, De Nardi et al found that to evolve successful helicopter control it was crucial to enforce some modularity-it was necessary to keep the yaw stabilization module separate from the networks that controlled other aspects of the helicopter's flight [10].…”

Ecological Neural Networks for Object Recognition and Generalization

“…The advantage of modular over nonmodular networks for learning the What and Where task has been confirmed by Di Ferdinando et al ([2]; see also [1]) who use a genetic algorithm [4] to evolve the network architecture in a population of neural networks starting from randomly generated architectures. The individual networks learn the What and Where task during their life using the backpropagation procedure and the networks with the best performance (least error) at the end of their life are selected for reproduction.…”

Genetic Representation and Evolvability of Modular Neural Controllers

“…It has been hypothesized that the reason for the higher evolvability of modular representations is that they change the effects of mutations [29], [61]. In order to analyze the effects of mutations, we subjected the networks of the last generation to mutations and compared the fitness of 1,000 mutant networks to the fitness of the original networks.…”

“…a genotype-phenotype map that translates modular genomes into modular neural networks). Experiments in a robotic cleaning task [25], [26], [27] and a visual discrimination problem [28], [29] indicated that the solutions found using modular mapping performed as well as hand-designed modular network structures and were better than non-modular networks. A limitation of these experiments is that they relied on simple, non-plastic, feed-forward network architectures and direct encoding of the network topology and weights, which restricts scalability.…”