Dataset Information

Bottlenecks, Modularity, and the Neural Control of Behavior.

ABSTRACT: In almost all animals, the transfer of information from the brain to the motor circuitry is facilitated by a relatively small number of neurons, leading to a constraint on the amount of information that can be transmitted. Our knowledge of how animals encode information through this pathway, and the consequences of this encoding, however, is limited. In this study, we use a simple feed-forward neural network to investigate the consequences of having such a bottleneck and identify aspects of the network architecture that enable robust information transfer. We are able to explain some recently observed properties of descending neurons-that they exhibit a modular pattern of connectivity and that their excitation leads to consistent alterations in behavior that are often dependent upon the desired behavioral state of the animal. Our model predicts that in the presence of an information bottleneck, such a modular structure is needed to increase the efficiency of the network and to make it more robust to perturbations. However, it does so at the cost of an increase in state-dependent effects. Despite its simplicity, our model is able to provide intuition for the trade-offs faced by the nervous system in the presence of an information processing constraint and makes predictions for future experiments.

SUBMITTER: Nande A

PROVIDER: S-EPMC9020368 | biostudies-literature | 2022

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Bottlenecks, Modularity, and the Neural Control of Behavior.

Nande Anjalika A Dubinkina Veronika V Ravasio Riccardo R Zhang Grace H GH Berman Gordon J GJ

Frontiers in behavioral neuroscience 20220406

In almost all animals, the transfer of information from the brain to the motor circuitry is facilitated by a relatively small number of neurons, leading to a constraint on the amount of information that can be transmitted. Our knowledge of how animals encode information through this pathway, and the consequences of this encoding, however, is limited. In this study, we use a simple feed-forward neural network to investigate the consequences of having such a bottleneck and identify aspects of the ...[more]

PMID: 35464140

Similar Datasets

Project description:Phase variation occurs in many pathogenic and commensal bacteria and is a major generator of genetic variability. A putative advantage of phase variation is to counter reductions in variability imposed by nonselective bottlenecks during transmission. Genomes of Campylobacter jejuni, a widespread food-borne pathogen, contain multiple phase-variable loci whose rapid, stochastic variation is generated by hypermutable simple sequence repeat tracts. These loci can occupy a vast number of combinatorial expression states (phasotypes) enabling populations to rapidly access phenotypic diversity. The imposition of nonselective bottlenecks can perturb the relative frequencies of phasotypes, changing both within-population diversity and divergence from the initial population. Using both in vitro testing of C. jejuni populations and a simple stochastic simulation of phasotype change, we observed that single-cell bottlenecks produce output populations of low diversity but with bimodal patterns of either high or low divergence. Conversely, large bottlenecks allow divergence only by accumulation of diversity, while interpolation between these extremes is observed in intermediary bottlenecks. These patterns are sensitive to the genetic diversity of initial populations but stable over a range of mutation rates and number of loci. The qualitative similarities of experimental and in silico modeling indicate that the observed patterns are robust and applicable to other systems where localized hypermutation is a defining feature. We conclude that while phase variation will maintain bacterial population diversity in the face of intermediate bottlenecks, narrow transmission-associated bottlenecks could produce host-to-host variation in bacterial phenotypes and hence stochastic variation in colonization and disease outcomes.IMPORTANCE Transmission and within-host spread of pathogenic organisms are associated with selective and nonselective bottlenecks that significantly reduced population diversity. In several bacterial pathogens, hypermutable mechanisms have evolved that mediate high-frequency reversible switching of specific phenotypes, such as surface structures, and hence counteract bottleneck-associated reductions in population diversity. Here, we investigated how combinations of hypermutable simple sequence repeats interact with nonselective bottlenecks by using a stochastic computer model and experimental data for Campylobacter jejuni, a food-borne pathogen. We find that bottleneck size qualitatively alters the output populations, with large bottlenecks maintaining population diversity while small bottlenecks produce dramatic shifts in the prevalence of particular variants. We conclude that narrow bottlenecks are capable of producing host-to-host variation in repeat-controlled bacterial phenotypes, leading to a potential for stochastic person-to-person variations in disease outcome for C. jejuni and other organisms with similar hypermutable mechanisms.

Project description:A long-standing goal in artificial intelligence is creating agents that can learn a variety of different skills for different problems. In the artificial intelligence subfield of neural networks, a barrier to that goal is that when agents learn a new skill they typically do so by losing previously acquired skills, a problem called catastrophic forgetting. That occurs because, to learn the new task, neural learning algorithms change connections that encode previously acquired skills. How networks are organized critically affects their learning dynamics. In this paper, we test whether catastrophic forgetting can be reduced by evolving modular neural networks. Modularity intuitively should reduce learning interference between tasks by separating functionality into physically distinct modules in which learning can be selectively turned on or off. Modularity can further improve learning by having a reinforcement learning module separate from sensory processing modules, allowing learning to happen only in response to a positive or negative reward. In this paper, learning takes place via neuromodulation, which allows agents to selectively change the rate of learning for each neural connection based on environmental stimuli (e.g. to alter learning in specific locations based on the task at hand). To produce modularity, we evolve neural networks with a cost for neural connections. We show that this connection cost technique causes modularity, confirming a previous result, and that such sparsely connected, modular networks have higher overall performance because they learn new skills faster while retaining old skills more and because they have a separate reinforcement learning module. Our results suggest (1) that encouraging modularity in neural networks may help us overcome the long-standing barrier of networks that cannot learn new skills without forgetting old ones, and (2) that one benefit of the modularity ubiquitous in the brains of natural animals might be to alleviate the problem of catastrophic forgetting.

Dataset Information

Bottlenecks, Modularity, and the Neural Control of Behavior.

Publications

Bottlenecks, Modularity, and the Neural Control of Behavior.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets