Understanding how a nervous system wires itself from birth to adulthood is a fundamental challenge in developmental neuroscience. We present DevoTG, a temporal graph framework that applies Temporal Graph Neural Networks (TGNs) to two complementary representations of C. elegans neural development: a Continuous-Time Dynamic Graph (CTDG) of cell division events derived from cell lineage data, and a Discrete-Time Dynamic Graph (DTDG) of the developing synaptic connectome spanning eight reconstructed electron-microscopy datasets. On the lineage prediction task, our TGN achieves a mean test AUC of 0.839 +/- 0.007 (5 seeds; validation AUC 0.937 +/- 0.001), outperforming a static GNN with the identical architecture by 26 AUC points (0.577 +/- 0.080), demonstrating that temporal memory is the decisive factor. Applied to the connectome DTDG, DevoTG identifies three connection stability classes (stable, developmental, and variable) across 225 neurons and 858 to 2,496 connections over development (L1 birth to adult), providing a temporal-graph-theoretic complement to the individual-variability classification of Witvliet et al. Analysis of hub command interneurons AVA, AVB, and AVE reveals their persistent centrality and how their integration roles are progressively reinforced across larval stages. Accompanying interactive visualizations (3D animated networks, centrality heatmaps, and a spatiotemporal lineage graph) make developmental dynamics accessible for biological hypothesis generation. DevoTG is open-source and designed for extension to other developing nervous systems. Code is publicly available at https://github.com/DevoLearn/DevoGraph/tree/main/DevoTG.

One classic idea from the cybernetics literature is the Every Good Regulator Theorem (EGRT). The EGRT provides a means to identify good regulation, or the conditions under which an agent (regulator) can match the dynamical behaviour of a system. We re-evaluate and recast the EGRT in a modern context to provide insight into how intelligent autonomous learning systems might utilize a compressed global representation (world model). One-to-one mappings between a regulator (R) and the corresponding system (S) provide a reduced representation that preserves useful variety to match all possible outcomes of a system. The EGRT also extends to second-order cybernetics, where an internal model (M) observes the behaviour of S and supervises an S-R closed-loop mapping. Secondarily, we demonstrate how physical phenomena such as temporal criticality, non-normal denoising and alternating procedural acquisition can recast behaviour as statistical mechanics and yield regulatory relationships. These diverse physical systems challenge the notion of tightly coupled good regulation when applied to non-uniform and out-of-distribution phenomena. Overall, we aim to recast the EGRT as a potential approach for developing world models that adapt and respond to a wide range of task environments. This article is part of the theme issue 'World models in natural and artificial intelligence'.

One advantage to the embodiment approach is that bodies and their geometry are a key element in conferring intelligence on a computational agent. We present an approach that draws from the phylogenetic approach to cognition and the embryogenesis of biological forms. The eco-evo-devo approach allows us to construct various morphological and behavioral configurations by utilizing a phylogenetic approach to cognition. This is compared and contrasted with the cybernetic approach to cognition, which is a more standard approach to implementing computational agents. To demonstrate this, the biocybernetic Braitenberg Vehicle approach is subject to a phylogenetic analysis, which reveals an evolutionary development (evo-devo) model of embodiment origins. Computational agents achieve embodiment through a genotype-to-phenotype mapping where phylogenetic diversification amongst agents is driven by mutational and recombination events. Cognitive abilities are built upon common ancestry and refined by taxonomic-specific genotypic configurations. We then compare our approach to the cybernetic approach to cognition and how our approach differs. To reconcile some of these incompatibilities, we discuss dual inheritance models that further enable an eco-evo-devo approach to computational intelligence. Overall, our phylogenetic approach unifies the existing literature on evolutionary and developmental robotics, in addition to more clearly articulating these mechanisms in biological context.

Simulating development has a number of benefits for understanding the acquisition of intelligent behavior. Yet development is not simply a generic form of generativity or emergence. Development is an inherently embodied and interactive process that unfolds in a limited time span. While embodied development provides a basis for grounding intelligent behavior, it also serves as a means to differentiate various behaviors with respect to the origins of phenotypic characteristics. Agentive development requires both morphogenetic and behavioral acquisitions which can be dependent upon one another. On the other hand, development requires innate and invariant features which interact with but are distinct from the environment. A set of models are proposed that define critical period acquisition in an embodied developmental context. Critical periods are periods of enhanced acquisition that shape future learning and experience. Leading to contingencies that affect the ability of an agent to integrate environmental information, critical periods also require the acquisition of the phenotype itself. leading to contingencies that affect the ability of an agent to integrate environmental information. This work is presented in light of the Developmental Neurosimulation paradigm, in addition to understanding agentive development from a biologically-inspired perspective.

The history of science has presented multiple ways to understand progressive and adaptive processes. In the field of cybernetics, this was understood to be teleological, or purposeful improvement over time. Early cyberneticists proposed a means to classify behaviors in a way that distinguishes between random behaviors and more controlled behaviors. Therefore, we begin by providing a history of teleological phenomena, exploring its multitude of current forms.We then take our own perspective, revising the behavioral typology presented in Rosenbleuth et.al in several ways, exploring alternatives to their stated examples. In the process, the concept of cybernetic imperatives is introduced, which provides an alternative to goal-directed, purposeful behavior. Cybernetic imperatives describe behaviors observed in a wide range of processes and real-world dynamics, building up from random processes to refined and so-called purposeful behaviors. To conclude, we consider how cybernetic imperatives lead to behavioral and regulatory complexity.

Naturalistic cognition in human performance is defined by dynamical responses to stimuli. Allostasis Machines (AMs) are characterized by an internal model and corresponding output trajectory characterizing a generalized response to stresses and sudden changes. The effects of the environment on the internal model are collectively known as perturbations, with a generalized response analogous to allostatic load. AMs consist of a sensory input, an internal model, a source of environmental perturbation, and an dynamical output that represents the response to perturbation over time. These dynamical output trajectories characterize this response either by recovering from perturbation (well-matched, ergodic), or drifting to a new stable state (accommodative, non-ergodic). We construct a quantitative model of AMs and consider their behaviors in a variety of scenarios, including isolated, serial, and new state perturbations. Control-theoretic strategies and multi-scale information processing can also be employed to provide AM models with more sophisticated feedback and control mechanisms. Understanding the difference between well-matched responses (stably matching environmental states) and allostatic drift (hysteretic responses to perturbation) clarifies how nonlinear responses produce continuous stability.

As development varies greatly across the tree of life, it may seem difficult to suggest a model that proposes a single mechanism for understanding collective cell behaviors and the coordination of tissue formation. Here we propose a mechanism called differentiation waves, which unify many disparate results involving developmental systems from across the tree of life. We demonstrate how a relatively simple model of differentiation proceeds not from function-related molecular mechanisms, but from so-called differentiation waves. A phenotypic model of differentiation waves is introduced, and its relation to molecular mechanisms is proposed. These waves contribute to a differentiation tree, which is an alternate way of viewing cell lineage and local action of the molecular factors. We construct a model of differentiation wave-related molecular mechanisms (genome, epigenome, and proteome) based on bioinformatic data from the nematode Caenorhabditis elegans. To validate this approach across different modes of development, we evaluate protein expression across different types of development by comparing Caenorhabditis elegans with several model organisms: fruit flies (Drosophila melanogaster), yeast (Saccharomyces cerevisiae), and mouse (Mus musculus). Inspired by gene regulatory networks, two Models of Interactive Contributions (fully-connected MICs and ordered MICs) are used to suggest potential genomic contributions to differentiation wave-related proteins. This, in turn, provides a framework for understanding differentiation and development.

Across the scientific literature, information measurement in the nervous system is posed as a problem of information processing internal to the brain by constructs such as neuronal populations, sensory surprise, or cognitive models. Application of information theory in the nervous system has focused on measuring phenomena such as capacity and integration. Yet the ecological perspective suggests that information is a product of active perception and interactions with the environment. Here, we propose Gibsonian Information (GI), relevant to both the study of cognitive agents and single cell systems that exhibit cognitive behaviors. We propose a formal model of GI that characterizes how agents extract environmental information in a dynamic fashion. GI demonstrates how sensory information guides information processing within individual nervous system representations of motion and continuous multisensory integration, as well as representations that guide collective behaviors. GI is useful for understanding first-order sensory inputs in terms of agent interactions with naturalistic contexts and simple internal representations and can be extended to cybernetic or symbolic representations. Statistical affordances, or clustered information that is spatiotemporally dependent perceptual input, facilitate extraction of GI from the environment. As a quantitative accounting of perceptual information, GI provides a means to measure a generalized indicator of nervous system input and can be characterized by three scenarios: disjoint distributions, contingent action, and coherent movement. By applying this framework to a variety of specific contexts, including a four-channel model of multisensory embodiment, we demonstrate how GI is essential to understanding the full scope of cognitive information processing.

The connection between active perception and the limits of performance provide a path to understanding naturalistic behavior. We can take a comparative cognitive modeling perspective to understand the limits of this performance and the existence of superperformance. We will discuss two categories that are hypothesized to originate in terms of coevolutionary relationships and evolutionary trade offs: supersamplers and superplanners. Supersamplers take snapshots of their sensory world at a very high sampling rate. Examples include flies (vision) and frogs (audition) with ecological specializations. Superplanners internally store information to evaluate and act upon multiple features of spatiotemporal environments. Slow lorises and turtles provide examples of superplanning capabilities. The Gibsonian Information (GI) paradigm is used to evaluate sensory sampling and planning with respect to direct perception and its role in capturing environmental information content. By contrast, superplanners utilize internal models of the environment to compensate for normal rates of sensory sampling, and this relationship often exists as a sampling/planning tradeoff. Supersamplers and superplanners can exist in adversarial relationships, or longer-term as coevolutionary relationships. Moreover, the tradeoff between sampling and planning capacity can break down, providing relativistic regimes. We can apply the principles of superperformance to human augmentation technologies.

Complex networks can be used to analyze structures and systems in the embryo. Not only can we characterize growth and the emergence of form, but also differentiation. The process of differentiation from precursor cell populations to distinct functional tissues is of particular interest. These phenomena can be captured using a hypergraph consisting of nodes represented by cell type categories and arranged as a directed cyclic graph (lineage hypergraph) and a complex network (spatial hypergraph). The lineage hypergraph models the developmental process as an n-ary tree, which can model two or more descendent categories per division event. A lineage tree based on the mosaic development of the nematode C. elegans (2-ary tree), is used to capture this process. Each round of divisions produces a new set of categories that allow for exchange of cells between types. An example from single-cell morphogenesis based on the cyanobacterial species Nostoc punctiforme (multiple discontinuous 2-ary tree) is also used to demonstrate the flexibility of this method. This model allows for new structures to emerge (such as a connectome) while also demonstrating how precursor categories are maintained for purposes such as dedifferentiation or other forms of cell fate plasticity. To understand this process of divergent integration, we analyze the directed hypergraph and categorical models, in addition to considering the role of network fistulas (spaces that conjoin two functional modules) and spatial restriction.

Cognitive offloading occurs when environmental affordances expand cognitive capacity while facilitating spatial and social behaviors. Capacity-related constraints are also important, particularly as embodied agents come online during development. Vast differences in brain size and offloading capacity exist across the tree of life. We take from multiple perspectives to understand the proportional contributions of internal models (brain) and externalized processing (offloading) in developing embodied computational agents. As developing nervous systems scale with body size and/or functional importance, offloading is also driven by neural capacity. Cognitive capacity is ultimately determined by various innate and environmental constraints. We propose a similar model for computationally developing cognitive agents. A regulatory model of cognition is proposed as a means to build cognitive systems that interface with biologically-inspired substrates. Multiple tradeoffs result from energetic, innate, and informational constraints, and determine the proportion of internal to external information processing capacity. As growth of a biologically-inspired substrate accelerates or decelerates over developmental time, it changes the acquisitional capacity of the agent. Our agent’s capacity limitations determine externalization potential, which is characterized by three parameters and two mathematical functions. The neurosimulation approach to intelligence offloading can be applied to a broad range of agent-based models and Artificial Intelligences.

The embryological view of development is that coordinated gene expression, cellular physics and migration provides the basis for phenotypic complexity. This stands in contrast with the prevailing view of embodied cognition, which claims that informational feedback between organisms and their environment is key to the emergence of intelligent behaviours. We aim to unite these two perspectives as embodied cognitive morphogenesis, in which morphogenetic symmetry breaking produces specialized organismal subsystems which serve as a substrate for the emergence of autonomous behaviours. As embodied cognitive morphogenesis produces fluctuating phenotypic asymmetry and the emergence of information processing subsystems, we observe three distinct properties: acquisition, generativity and transformation. Using a generic organismal agent, such properties are captured through models such as tensegrity networks, differentiation trees and embodied hypernetworks, providing a means to identify the context of various symmetry-breaking events in developmental time. Related concepts that help us define this phenotype further include concepts such as modularity, homeostasis and 4E (embodied, enactive, embedded and extended) cognition. We conclude by considering these autonomous developmental systems as a process called connectogenesis, connecting various parts of the emerged phenotype into an approach useful for the analysis of organisms and the design of bioinspired computational agents.

One important feature of complex systems are problem domains that have many local minima and substructure. Biological systems manage these local minima by switching between different subsystems depending on their environmental or developmental context. Genetic Algorithms (GA) can mimic this switching property as well as provide a means to overcome problem domain complexity. However, standard GA requires additional operators that will allow for large-scale exploration in a stochastic manner. Gradient-free heuristic search techniques are suitable for providing an optimal solution in the discrete domain to such single objective optimization tasks, particularly compared to gradient-based methods which are noticeably slower. To do this, the authors turn to an optimization problem from the flight scheduling domain. The authors compare the performance of such common gradient-free heuristic search algorithms and propose variants of GAs. The Iterated Chaining (IC) method is also introduced, building upon traditional chaining techniques by triggering multiple local searches instead of the singular action of a mutation operator. The authors will show that the use of multiple local searches can improve performance on local stochastic searches, providing ample opportunity for application to a host of other problem domains. It is observed that the proposed GA variants have the least average cost across all benchmarks including the problem proposed and IC algorithm performs better than its constituents.

Neuromatch Academy (https://neuromatch.io/academy) was designed as an online summer school to cover the basics of computational neuroscience in three weeks. The materials cover dominant and emerging computational neuroscience tools, how they complement one another, and specifically focus on how they can help us to better understand how the brain functions. An original component of the materials is its focus on modeling choices, i.e. how do we choose the right approach, how do we build models, and how can we evaluate models to determine if they provide real (meaningful) insight. This meta-modeling component of the instructional materials asks what questions can be answered by different techniques, and how to apply them meaningfully to get insight about brain function.

What role does phenotypic complexity play in the systems-level function of an embodied agent? The organismal phenotype is a topologically complex structure that interacts with a genotype, developmental physics, and an informational environment. Using this observation as inspiration, we utilize a type of embodied agent that exhibits layered representational capacity: meta-brain models. Meta-brains are used to demonstrate how phenotypes process information and exhibit self-regulation from development to maturity. We focus on two candidate structures that potentially explain this capacity: folding and layering. As layering and folding can be observed in a host of biological contexts, they form the basis for our representational investigations. First, an innate starting point (genomic encoding) is described. The generative output of this encoding is a differentiation tree, which results in a layered phenotypic representation. Then we specify a formal meta-brain model of the gut, which exhibits folding and layering in development in addition to different degrees of representation of processed information. This organ topology is retained in maturity, with the potential for additional folding and representational drift in response to inflammation. Next, we consider topological remapping using the developmental Braitenberg Vehicle (dBV) as a toy model. During topological remapping, it is shown that folding of a layered neural network can introduce a number of distortions to the original model, some with functional implications. The paper concludes with a discussion on how the meta-brains method can assist us in the investigation of enactivism, holism, and cognitive processing in the context of biological simulation.

We propose a new way to quantitatively characterize information: Gibsonian Information (GI). GI provides a means to characterize how agents extract information from direct perceptual signals. In this paper, we characterize GI quantitatively, and contrast this with rival approaches to quantitative information. Our formulation differs from existing approaches to measuring information in two ways. The first involves an emphasis on sensory processing and the dynamic evolution of such interactions. More broadly, GI also provides a means to measure a generalized indicator of nervous system input, and can be characterized in terms of multisensory integration and collective behavior. Overall, GI enables a differential system between both motion (information) and random noise/stasis (non-information) that can potentially be applied to a wide range of problem domains.

Connecting brain and behavior is a longstanding issue in the areas of behavioral science, artificial intelligence, and neurobiology. As is standard among models of artificial and biological neural networks, an analogue of the fully mature brain is presented as a blank slate. However, this does not consider the realities of biological development and developmental learning. Our purpose is to model the development of an artificial organism that exhibits complex behaviors. We introduce three alternate approaches to demonstrate how developmental embodied agents can be implemented. The resulting developmental Braitenberg vehicles (dBVs) will generate behaviors ranging from stimulus responses to group behavior that resembles collective motion. We will situate this work in the domain of artificial brain networks along with broader themes such as embodied cognition, feedback, and emergence. Our perspective is exemplified by three software instantiations that demonstrate how a BV-genetic algorithm hybrid model, a multisensory Hebbian learning model, and multi-agent approaches can be used to approach BV development. We introduce use cases such as optimized spatial cognition (vehicle-genetic algorithm hybrid model), hinges connecting behavioral and neural models (multisensory Hebbian learning model), and cumulative classification (multi-agent approaches). In conclusion, we consider future applications of the developmental neurosimulation approach.

As a biochemical process, direct cellular reprogramming is slow and complex. The early stages of this process is the most critical determinant of successful phenotypic conversion. This study provides insight into the statistical signatures that describe temporal structure in the reprogramming process. We examine two sources of variation in reprogramming cells: clonal instances from various tissues of origin and rate of expansion between these lines. Our analytical strategy involved modeling the potential of populations to reprogram, and then applying statistical models to capture this potential in action. This two-fold approach utilizes both conventional and novel techniques that allow us to infer and confirm a host of properties that define the phenomenon. These results can be summarized in a number of ways, and essentially suggest that reprogramming is organized around changes in gene expression phenotype (phases) which happens sporadically across a cellular population (bursts).