Guided self-organization in indirectly encoded and evolving topographic maps

Abstract

An important phenomenon seen in many areas of biological brains and recently in deep learning architectures is a process known as self-organization. For example, in the primary visual cortex, color and orientation maps develop based on lateral inhibitory connectivity patterns and Hebbian learning dynamics. These topographic maps, which are found in all sensory systems, are thought to be a key factor in enabling abstract cognitive representations. This paper shows for the first time that the Hypercube-based NeuroEvolution of Augmenting Topologies (HyperNEAT) method can be seeded to begin evolution with such lateral connectivity, enabling genuine self-organizing dynamics. The proposed approach draws on HyperNEAT's ability to generate a pattern of weights across the connectivity of an artificial neural network (ANN) based on a function of its geometry. Validating this approach, the afferent weights of an ANN self-organize in this paper to form a genuine topographic map of the input space for a simple line orientation task. Most interestingly, this seed can then be evolved further, providing a method to guide the self-organization of weights in a specific way, much as evolution likely guided the self-organizing trajectories of biological brains.

Distinctive properties of biological neural networks and recent advances in bottom-up approaches toward a better biologically plausible neural network

“…The artificial counterpart of the mutation-selection process, namely, the evolutionary algorithm (EA), has been applied in numerous domains for decades, and “neuroevolution” refers to the application of EA to neural networks (Yao and Liu, 1998 ; Stanley et al, 2019 ; Galván and Mooney, 2021 ). Although the neuroevolution scheme simplified or omitted numerous aspects of the biological evolution process, it successfully captured the essentials and performed well in rediscovering the BNN properties (Risi and Stanley, 2014 ) and optimizing the ANN architecture (Liang et al, 2018 ; Zoph et al, 2018 ). In addition to structural connectivity, network architecture comprises the functional features of a network, such as the activation function of each neuron and its hyperparameters or initial synaptic weights.…”

Born to learn: The inspiration, progress, and future of evolved plastic artificial neural networks

“…Adaptive ES-HyperNEAT allows each individual synaptic connection, rather than neuron, to be standard or modulatory, thus introducing further design flexibility. Risi and Stanley (2014) showed how adaptive HyperNEAT can be seeded to produce a specific lateral connectivity pattern, thereby allowing the weights to self-organize to form a topographic map of the input space. The study shows that evolution can be seeded with specific plasticity mechanisms that can facilitate the evolution of specific types of learning.…”

Neuroevolution in Games: State of the Art and Open Challenges

“…However, novel combinations of recent advances such as more advanced forms of local plasticity (e.g. neuromodulation [113]), hypothesis testing in distal reward learning [112], larger indirectly-encoded adaptive networks [91][92][93], methods that avoid deception inherent in evolving learning architectures [63,94], and learning of large behavioral repertoires [23], could allow the creating of learning networks for more complex domains such as games.…”