Presented at Network Neuroscience 2022.
The structural and functional emergence of simple connectomes can yield great diversity and complexity, even in their tiniest forms. To better understand how network phenotypes lead to the emergence of complex outputs, we assume three requirements: a mechanism for direct perception, an internal network that processes the signal between sensory input and action output, and an embodied phenotype that generates said action. A variant of the developmental Braitenberg Vehicle (dBV) [1] can be used to model input (sensory information), processing of the I/O signal (internal network), and output (effector action). The dBV provides a simple integrated sensor-to-effector system that typically characterizes linear to combinatorial relations between sensor and effector.
But what is the potential role of a network in such a simple system? We propose a rather provocative hypothesis: that increasing functional complexity is marked by maximizing the circuitousness (or path length) between sensor and effector. This assumes the goal of internal networks is to convolve the signal between perception and action as much as possible. By coupling our dBV model with biological Rube Goldberg Machines [2] as the I/O network, nonlinear directed graphs can be produced that demonstrate non-optimal phenotypes and output behaviors. We will walk through the means by which these non-optimal components might be generated developmentally, particularly in contrast with optimality criteria for network growth..
Modeling increased phenotypic complexity in a suboptimal manner favors the emergence of redundancies, which in turn serve as the interface for new phenotypic components and modular subdivision of the internal network. Both of these are generated by the developmental process, as increasing sensory experience intersects with changing network parameter values (e.g. growth in network diameter and increasing path length from sensor to effector). Maximizing circuitous path lengths is also consequential to the production of novel behaviors and ambiguities. As the network phenotype grows in complexity, the behavioral outputs may become more diverse and more unpredictable. This provides not only a mapping between stimulus and behavior, but also between embodied perception and network architectures of varying complexity.
References: [1] Dvoretskii, S. Gong, Z. Gupta, A. Parent, J. and Alicea, B. (2022). Braitenberg Ve...
The structural and functional emergence of simple connectomes can yield great diversity and complexity, even in their tiniest forms. To better understand how network phenotypes lead to the emergence of complex outputs, we assume three requirements: a mechanism for direct perception, an internal network that processes the signal between sensory input and action output, and an embodied phenotype that generates said action. A variant of the developmental Braitenberg Vehicle (dBV) [1] can be used to model input (sensory information), processing of the I/O signal (internal network), and output (effector action). The dBV provides a simple integrated sensor-to-effector system that typically characterizes linear to combinatorial relations between sensor and effector.
But what is the potential role of a network in such a simple system? We propose a rather provocative hypothesis: that increasing functional complexity is marked by maximizing the circuitousness (or path length) between sensor and effector. This assumes the goal of internal networks is to convolve the signal between perception and action as much as possible. By coupling our dBV model with biological Rube Goldberg Machines [2] as the I/O network, nonlinear directed graphs can be produced that demonstrate non-optimal phenotypes and output behaviors. We will walk through the means by which these non-optimal components might be generated developmentally, particularly in contrast with optimality criteria for network growth..
Modeling increased phenotypic complexity in a suboptimal manner favors the emergence of redundancies, which in turn serve as the interface for new phenotypic components and modular subdivision of the internal network. Both of these are generated by the developmental process, as increasing sensory experience intersects with changing network parameter values (e.g. growth in network diameter and increasing path length from sensor to effector). Maximizing circuitous path lengths is also consequential to the production of novel behaviors and ambiguities. As the network phenotype grows in complexity, the behavioral outputs may become more diverse and more unpredictable. This provides not only a mapping between stimulus and behavior, but also between embodied perception and network architectures of varying complexity.
References: [1] Dvoretskii, S. Gong, Z. Gupta, A. Parent, J. and Alicea, B. (2022). Braitenberg Ve...