A model for the neuronal substrate of dead reckoning and memory in arthropods: a comparative computational and behavioral study

Formalising the role of behaviour in neuroscience

We develop a mathematical approach to formally proving that certain neural computations and representations exist based on patterns observed in an organism's behaviour. To illustrate, we provide a simple set of conditions under which an ant's ability to determine how far it is from its nest would logically imply neural structures isomorphic to the natural numbers ℕ . We generalise these results to arbitrary behaviours and representations and show what mathematical characterisation of neural computation and representation is simplest while being maximally predictive of behaviour. We develop this framework in detail using a path integration example, where an organism's ability to search for its nest in the correct location implies representational structures isomorphic to two-dimensional coordinates under addition. We also study a system for processinga n b n strings common in comparative work. Our approach provides an objective way to determine what theory of a physical system is best, addressing a fundamental challenge in neuroscientific inference. These results motivate considering which neurobiological structures have the requisite formal structure and are otherwise physically plausible given relevant physical considerations such as generalisability, information density, thermodynamic stability and energetic cost.

A connectome of the Drosophila central complex reveals network motifs suitable for flexible navigation and context-dependent action selection

Flexible behaviors over long timescales are thought to engage recurrent neural networks in deep brain regions, which are experimentally challenging to study. In insects, recurrent circuit dynamics in a brain region called the central complex (CX) enable directed locomotion, sleep, and context- and experience-dependent spatial navigation. We describe the first complete electron microscopy-based connectome of the Drosophila CX, including all its neurons and circuits at synaptic resolution. We identified new CX neuron types, novel sensory and motor pathways, and network motifs that likely enable the CX to extract the fly's head direction, maintain it with attractor dynamics, and combine it with other sensorimotor information to perform vector-based navigational computations. We also identified numerous pathways that may facilitate the selection of CX-driven behavioral patterns by context and internal state. The CX connectome provides a comprehensive blueprint necessary for a detailed understanding of network dynamics underlying sleep, flexible navigation, and state-dependent action selection.

A Neurocomputational Model of Goal-Directed Navigation in Insect-Inspired Artificial Agents

Despite their small size, insect brains are able to produce robust and efficient navigation in complex environments. Specifically in social insects, such as ants and bees, these navigational capabilities are guided by orientation directing vectors generated by a process called path integration. During this process, they integrate compass and odometric cues to estimate their current location as a vector, called the home vector for guiding them back home on a straight path. They further acquire and retrieve path integration-based vector memories globally to the nest or based on visual landmarks. Although existing computational models reproduced similar behaviors, a neurocomputational model of vector navigation including the acquisition of vector representations has not been described before. Here we present a model of neural mechanisms in a modular closed-loop control—enabling vector navigation in artificial agents. The model consists of a path integration mechanism, reward-modulated global learning, random search, and action selection. The path integration mechanism integrates compass and odometric cues to compute a vectorial representation of the agent's current location as neural activity patterns in circular arrays. A reward-modulated learning rule enables the acquisition of vector memories by associating the local food reward with the path integration state. A motor output is computed based on the combination of vector memories and random exploration. In simulation, we show that the neural mechanisms enable robust homing and localization, even in the presence of external sensory noise. The proposed learning rules lead to goal-directed navigation and route formation performed under realistic conditions. Consequently, we provide a novel approach for vector learning and navigation in a simulated, situated agent linking behavioral observations to their possible underlying neural substrates.

An embodied biologically constrained model of foraging: from classical and operant conditioning to adaptive real-world behavior in DAC-X

Animals successfully forage within new environments by learning, simulating and adapting to their surroundings. The functions behind such goal-oriented behavior can be decomposed into 5 top-level objectives: 'how', 'why', 'what', 'where', 'when' (H4W). The paradigms of classical and operant conditioning describe some of the behavioral aspects found in foraging. However, it remains unclear how the organization of their underlying neural principles account for these complex behaviors. We address this problem from the perspective of the Distributed Adaptive Control theory of mind and brain (DAC) that interprets these two paradigms as expressing properties of core functional subsystems of a layered architecture. In particular, we propose DAC-X, a novel cognitive architecture that unifies the theoretical principles of DAC with biologically constrained computational models of several areas of the mammalian brain. DAC-X supports complex foraging strategies through the progressive acquisition, retention and expression of task-dependent information and associated shaping of action, from exploration to goal-oriented deliberation. We benchmark DAC-X using a robot-based hoarding task including the main perceptual and cognitive aspects of animal foraging. We show that efficient goal-oriented behavior results from the interaction of parallel learning mechanisms accounting for motor adaptation, spatial encoding and decision-making. Together, our results suggest that the H4W problem can be solved by DAC-X building on the insights from the study of classical and operant conditioning. Finally, we discuss the advantages and limitations of the proposed biologically constrained and embodied approach towards the study of cognition and the relation of DAC-X to other cognitive architectures.

Dead Reckoning in the Desert Ant: A Defence of Connectionist Models

Dead reckoning is a feature of the navigation behaviour shown by several creatures, including the desert ant. Recent work by C. Randy Gallistel shows that some connectionist models of dead reckoning face important challenges. These challenges are thought to arise from essential features of the connectionist approach, and have therefore been taken to show that connectionist models are unable to explain even the most primitive of psychological phenomena. I show that Gallistel’s challenges are successfully met by one recent connectionist model, proposed by Ulysses Bernardet, Sergi Bermúdez i Badia, and Paul F.M.J. Verschure. The success of this model suggests that there are ways to implement dead reckoning with neural circuits that fall within the bounds of what many people regard as neurobiologically plausible, and so that the wholesale dismissal of the connectionist modelling project remains premature.

A biologically based model for the integration of sensory-motor contingencies in rules and plans: a prefrontal cortex based extension of the Distributed Adaptive Control architecture

Intelligence depends on the ability of the brain to acquire and apply rules and representations. At the neuronal level these properties have been shown to critically depend on the prefrontal cortex. Here we present, in the context of the Distributed Adaptive Control architecture (DAC), a biologically based model for flexible control and planning based on key physiological properties of the prefrontal cortex, i.e. reward modulated sustained activity and plasticity of lateral connectivity. We test the model in a series of pertinent tasks, including multiple T-mazes and the Tower of London that are standard experimental tasks to assess flexible control and planning. We show that the model is both able to acquire and express rules that capture the properties of the task and to quickly adapt to changes. Further, we demonstrate that this biomimetic self-contained cognitive architecture generalizes to planning. In addition, we analyze the extended DAC architecture, called DAC 6, as a model that can be applied for the creation of intelligent and psychologically believable synthetic agents.

iqr: a tool for the construction of multi-level simulations of brain and behaviour

The brain is the most complex system we know of. Despite the wealth of data available in neuroscience, our understanding of this system is still very limited. Here we argue that an essential component in our arsenal of methods to advance our understanding of the brain is the construction of artificial brain-like systems. In this way we can encompass the multi-level organisation of the brain and its role in the context of the complete embodied real-world and real-time perceiving and behaving system. Hence, on the one hand, we must be able to develop and validate theories of brains as closing the loop between perception and action, and on the other hand as interacting with the real world. Evidence is growing that one of the sources of the computational power of neuronal systems lies in the massive and specific connectivity, rather than the complexity of single elements. To meet these challenges-multiple levels of organisation, sophisticated connectivity, and the interaction of neuronal models with the real-world-we have developed a multi-level neuronal simulation environment, iqr. This framework deals with these requirements by directly transforming them into the core elements of the simulation environment itself. iqr provides a means to design complex neuronal models graphically, and to visualise and analyse their properties on-line. In iqr connectivity is defined in a flexible, yet compact way, and simulations run at a high speed, which allows the control of real-world devices-robots in the broader sense-in real-time. The architecture of iqr is modular, providing the possibility to write new neuron, and synapse types, and custom interfaces to other hardware systems. The code of iqr is publicly accessible under the GNU General Public License (GPL). iqr has been in use since 1996 and has been the core tool for a large number of studies ranging from detailed models of neuronal systems like the cerebral cortex, and the cerebellum, to robot based models of perception, cognition and action to large-scale real-world systems. In addition, iqr has been widely used over many years to introduce students to neuronal simulation and neuromorphic control. In this paper we outline the conceptual and methodological background of iqr and its design philosophy. Thereafter we present iqr's main features and computational properties. Finally, we describe a number of projects using iqr, singling out how iqr is used for building a "synthetic insect".

Which coordinate system for modelling path integration?

Path integration is a navigation strategy widely observed in nature where an animal maintains a running estimate, called the home vector, of its location during an excursion. Evidence suggests it is both ancient and ubiquitous in nature, and has been studied for over a century. In that time, canonical and neural network models have flourished, based on a wide range of assumptions, justifications and supporting data. Despite the importance of the phenomenon, consensus and unifying principles appear lacking. A fundamental issue is the neural representation of space needed for biological path integration. This paper presents a scheme to classify path integration systems on the basis of the way the home vector records and updates the spatial relationship between the animal and its home location. Four extended classes of coordinate systems are used to unify and review both canonical and neural network models of path integration, from the arthropod and mammalian literature. This scheme demonstrates analytical equivalence between models which may otherwise appear unrelated, and distinguishes between models which may superficially appear similar. A thorough analysis is carried out of the equational forms of important facets of path integration including updating, steering, searching and systematic errors, using each of the four coordinate systems. The type of available directional cue, namely allothetic or idiothetic, is also considered. It is shown that on balance, the class of home vectors which includes the geocentric Cartesian coordinate system, appears to be the most robust for biological systems. A key conclusion is that deducing computational structure from behavioural data alone will be difficult or impossible, at least in the absence of an analysis of random errors. Consequently it is likely that further theoretical insights into path integration will require an in-depth study of the effect of noise on the four classes of home vectors.