Inevitable evolutionary temporal elements in neural processing: a study based on evolutionary simulations

Abstract

Recent studies have suggested that some neural computational mechanisms are based on the fine temporal structure of spiking activity. However, less effort has been devoted to investigating the evolutionary aspects of such mechanisms. In this paper we explore the issue of temporal neural computation from an evolutionary point of view, using a genetic simulation of the evolutionary development of neural systems. We evolve neural systems in an environment with selective pressure based on mate finding, and examine the temporal aspects of the evolved systems. In repeating evolutionary sessions, there was a significant increase during evolution in the mutual information between the evolved agent's temporal neural representation and the external environment. In ten different simulated evolutionary sessions, there was an increased effect of time-related neural ablations on the agents' fitness. These results suggest that in some fitness landscapes the emergence of temporal elements in neural computation is almost inevitable. Future research using similar evolutionary simulations may shed new light on various biological mechanisms.

Figure 1. An example of development of a 4 cell agent.

A) The development process begins with a chromosome. B) The chromosome is translated into a gene-protein network expressed in a zygote. C) The gene-protein network triggers mitosis events, producing 4 different cells. The network is the same in all cells, but the concentrations are different. D) The cells migrate and differentiate into a neuron, motor cell, and two sensory cells. E) Neurite sprouting events occur. Some proteins are marked by k_type as ones that cannot diffuse from neurite to soma. Therefore, their instances are separated in the neurites, with the same connectivity. F) After the axon is guided by external protein concentrations, target selection events occur, causing the axons to synapse. A synapse is formed, allowing proteins marked k_type as synapse-diffusible to move from one cell to another.

Figure 2. Behavioral development during evolution.

Red: Proportion of reproduction triggered by agent contacts (as opposed to reproductions initiated by the system when the number of agents was too low). Black: Proportion of agents that developed a basic network (as defined in the text). Blue: Proportion of agent death events triggered by the system because of crowding (as opposed to deaths due to completing the life span period). The values are average proportions measured every 5 generations.

Figure 3. Static estimation of the mutual information between the agent's neural representation and the environment.

The agent (arrowed) is moved between the center of s₁ and s₂ in a random order while its neural activity is recorded. The agent is pinned to the center of s₁ and s₂ and cannot move freely. s₁ and s₂ are two different environments containing agents of the same sex (grey) or opposite sex (white) at random locations.

Figure 4. Best estimated mutual information with static environment values of randomly selected agents during evolution.

A) The rate based measure has higher values than the other measures, but no significant correlation was observed with generation (P =  0.7087, r =  2×10⁻²). B) Measure based on cross correlation combined with lag . C) Measure based on cross correlation alone . D) Measure based on lag alone . All values are based on Spearman's Rank Correlation Test made on 340 randomly chosen agents from the same evolutionary session. Please note the different axis in A.

Figure 5. Dynamic estimation of the mutual information between the agent's neural representation and the environment.

The agent (arrowed) moved in a single environment freely. The external environment was defined as d_l or d_r when there were more agents of the opposite sex on its left or right respectively.

Figure 6. Best estimated mutual information with dynamic environment values of randomly selected agents during evolution.

A) The rate based measure has higher values than the other measures, with a significant correlation . B) Measure based on cross correlation combined with lag . C) Measure based on cross correlation alone . D) Measure based on lag alone . All values are based on Spearman's Rank Correlation Test made on randomly chosen agents from the same evolutionary session. Since the cross correlation based measures reached a plateau earlier than the others, their statistical tests were based on the first 2700 generations whereas the other tests were based on the entire evolutionary session. Please note the different axis in A and B.

Figure 7. Ablation Effect correlated with generations.

Each sample is the average ablation effect measured for a population of 100 agents in 100 generation bins. In each generation one agent was chosen randomly for this experiment. (P = 5.0×10⁻³, r = 0.63, N = 20, Spearman's Rank Correlation Test).