Selection of cortical dynamics for motor behaviour by the basal ganglia

Chaotic recurrent neural networks for brain modelling: A review

Even in the absence of external stimuli, the brain is spontaneously active. Indeed, most cortical activity is internally generated by recurrence. Both theoretical and experimental studies suggest that chaotic dynamics characterize this spontaneous activity. While the precise function of brain chaotic activity is still puzzling, we know that chaos confers many advantages. From a computational perspective, chaos enhances the complexity of network dynamics. From a behavioural point of view, chaotic activity could generate the variability required for exploration. Furthermore, information storage and transfer are maximized at the critical border between order and chaos. Despite these benefits, many computational brain models avoid incorporating spontaneous chaotic activity due to the challenges it poses for learning algorithms. In recent years, however, multiple approaches have been proposed to overcome this limitation. As a result, many different algorithms have been developed, initially within the reservoir computing paradigm. Over time, the field has evolved to increase the biological plausibility and performance of the algorithms, sometimes going beyond the reservoir computing framework. In this review article, we examine the computational benefits of chaos and the unique properties of chaotic recurrent neural networks, with a particular focus on those typically utilized in reservoir computing. We also provide a detailed analysis of the algorithms designed to train chaotic RNNs, tracing their historical evolution and highlighting key milestones in their development. Finally, we explore the applications and limitations of chaotic RNNs for brain modelling, consider their potential broader impacts beyond neuroscience, and outline promising directions for future research.

A TAN-dopamine interaction mechanism based computational model of basal ganglia in action selection

The basal ganglia (BG) plays a key role in action selection. Physiological experiments have suggested that the reciprocal interaction between tonically active neurons (TANs) and dopamine (DA) is closely related to reward-based behaviors. However, the functional role of TAN-DA interaction in action selection remains unclear. In this study, a cortico-BG model including TAN-DA interaction mechanism is developed to explore the action selection mechanism of BG. The results show that in the default case, direct, indirect, and hyperdirect pathways are responsible for promoting, suppressing, and stopping the formation of stimulus-action associations, respectively. In the case of reinforcement learning, a single rewarded action is selected according to the combination of the TAN-DA dependent reinforcement mechanism and Hebbian mechanism with a gradual transfer from the former to the latter. Besides, a longer exploratory phase occurs when switching the reward to a new action because additional trials are required to overcome the habituation previously induced by the Hebbian mechanism. In the Parkinsonian state, the reinforcement mechanism is disrupted, and the resting tremor occurs due to dopamine deficiency. Although the model's performance significantly improves due to the levodopa treatment, it is still inferior to the healthy state. This phenomenon is consistent with the experimental results and is explained theoretically via the TAN pause duration and phasic DA release. Furthermore, the model's performances in multi-action selection further verify the rationality of the TAN-DA-dependent reinforcement mechanism. Our work provides a more complete framework for studying the action selection mechanism of basal ganglia.

Brain-inspired biomimetic robot control: a review

Complex robotic systems, such as humanoid robot hands, soft robots, and walking robots, pose a challenging control problem due to their high dimensionality and heavy non-linearities. Conventional model-based feedback controllers demonstrate robustness and stability but struggle to cope with the escalating system design and tuning complexity accompanying larger dimensions. In contrast, data-driven methods such as artificial neural networks excel at representing high-dimensional data but lack robustness, generalization, and real-time adaptiveness. In response to these challenges, researchers are directing their focus to biological paradigms, drawing inspiration from the remarkable control capabilities inherent in the human body. This has motivated the exploration of new control methods aimed at closely emulating the motor functions of the brain given the current insights in neuroscience. Recent investigation into these Brain-Inspired control techniques have yielded promising results, notably in tasks involving trajectory tracking and robot locomotion. This paper presents a comprehensive review of the foremost trends in biomimetic brain-inspired control methods to tackle the intricacies associated with controlling complex robotic systems.

Bridging flexible goal-directed cognition and consciousness: The Goal-Aligning Representation Internal Manipulation theory

Goal-directed manipulation of internal representations is a key element of human flexible behaviour, while consciousness is commonly associated with higher-order cognition and human flexibility. Current perspectives have only partially linked these processes, thus preventing a clear understanding of how they jointly generate flexible cognition and behaviour. Moreover, these limitations prevent an effective exploitation of this knowledge for technological scopes. We propose a new theoretical perspective that extends our 'three-component theory of flexible cognition' toward higher-order cognition and consciousness, based on the systematic integration of key concepts from Cognitive Neuroscience and AI/Robotics. The theory proposes that the function of conscious processes is to support the alignment of representations with multi-level goals. This higher alignment leads to more flexible and effective behaviours. We analyse here our previous model of goal-directed flexible cognition (validated with more than 20 human populations) as a starting GARIM-inspired model. By bridging the main theories of consciousness and goal-directed behaviour, the theory has relevant implications for scientific and technological fields. In particular, it contributes to developing new experimental tasks and interpreting clinical evidence. Finally, it indicates directions for improving machine learning and robotics systems and for informing real-world applications (e.g., in digital-twin healthcare and roboethics).

Invariant neural dynamics drive commands to control different movements

It has been proposed that the nervous system has the capacity to generate a wide variety of movements because it reuses some invariant code. Previous work has identified that dynamics of neural population activity are similar during different movements, where dynamics refer to how the instantaneous spatial pattern of population activity changes in time. Here, we test whether invariant dynamics of neural populations are actually used to issue the commands that direct movement. Using a brain-machine interface (BMI) that transforms rhesus macaques' motor-cortex activity into commands for a neuroprosthetic cursor, we discovered that the same command is issued with different neural-activity patterns in different movements. However, these different patterns were predictable, as we found that the transitions between activity patterns are governed by the same dynamics across movements. These invariant dynamics are low dimensional, and critically, they align with the BMI, so that they predict the specific component of neural activity that actually issues the next command. We introduce a model of optimal feedback control (OFC) that shows that invariant dynamics can help transform movement feedback into commands, reducing the input that the neural population needs to control movement. Altogether our results demonstrate that invariant dynamics drive commands to control a variety of movements and show how feedback can be integrated with invariant dynamics to issue generalizable commands.

The contribution of the basal ganglia and cerebellum to motor learning: A neuro-computational approach

Motor learning involves a widespread brain network including the basal ganglia, cerebellum, motor cortex, and brainstem. Despite its importance, little is known about how this network learns motor tasks and which role different parts of this network take. We designed a systems-level computational model of motor learning, including a cortex-basal ganglia motor loop and the cerebellum that both determine the response of central pattern generators in the brainstem. First, we demonstrate its ability to learn arm movements toward different motor goals. Second, we test the model in a motor adaptation task with cognitive control, where the model replicates human data. We conclude that the cortex-basal ganglia loop learns via a novelty-based motor prediction error to determine concrete actions given a desired outcome, and that the cerebellum minimizes the remaining aiming error.

Striatonigrostriatal Spirals in Addiction

A biological reward system is integral to all animal life and humans are no exception. For millennia individuals have investigated this system and its influences on human behavior. In the modern day, with the US facing an ongoing epidemic of substance use without an effective treatment, these investigations are of paramount importance. It is well known that basal ganglia contribute to rewards and are involved in learning, approach behavior, economic choices, and positive emotions. This review aims to elucidate the physiological role of striatonigrostriatal (SNS) spirals, as part of basal ganglia circuits, in this reward system and their pathophysiological role in perpetuating addiction. Additionally, the main functions of neurotransmitters such as dopamine and glutamate and their receptors in SNS circuits will be summarized. With this information, the claim that SNS spirals are crucial intermediaries in the shift from goal-directed behavior to habitual behavior will be supported, making this circuit a viable target for potential therapeutic intervention in those with substance use disorders.

Active inference through whiskers

Rodents use whisking to probe actively their environment and to locate objects in space, hence providing a paradigmatic biological example of active sensing. Numerous studies show that the control of whisking has anticipatory aspects. For example, rodents target their whisker protraction to the distance at which they expect objects, rather than just reacting fast to contacts with unexpected objects. Here we characterize the anticipatory control of whisking in rodents as an active inference process. In this perspective, the rodent is endowed with a prior belief that it will touch something at the end of the whisker protraction, and it continuously modulates its whisking amplitude to minimize (proprioceptive and somatosensory) prediction errors arising from an unexpected whisker-object contact, or from a lack of an expected contact. We will use the model to qualitatively reproduce key empirical findings about the ways rodents modulate their whisker amplitude during exploration and the scanning of (expected or unexpected) objects. Furthermore, we will discuss how the components of active inference model can in principle map to the neurobiological circuits of rodent whisking.

Bidirectional interaction between visual and motor generative models using Predictive Coding and Active Inference

In this work, we build upon the Active Inference (AIF) and Predictive Coding (PC) frameworks to propose a neural architecture comprising a generative model for sensory prediction, and a distinct generative model for motor trajectories. We highlight how sequences of sensory predictions can act as rails guiding learning, control and online adaptation of motor trajectories. We furthermore inquire the effects of bidirectional interactions between the motor and the visual modules. The architecture is tested on the control of a simulated robotic arm learning to reproduce handwritten letters.

Internal manipulation of perceptual representations in human flexible cognition: A computational model

Executive functions represent a set of processes in goal-directed cognition that depend on integrated cortical-basal ganglia brain systems and form the basis of flexible human behaviour. Several computational models have been proposed for studying cognitive flexibility as a key executive function and the Wisconsin card sorting test (WCST) that represents an important neuropsychological tool to investigate it. These models clarify important aspects that underlie cognitive flexibility, particularly decision-making, motor response, and feedback-dependent learning processes. However, several studies suggest that the categorisation processes involved in the solution of the WCST include an additional computational stage of category representation that supports the other processes. Surprisingly, all models of the WCST ignore this fundamental stage and they assume that decision making directly triggers actions. Thus, we propose a novel hypothesis where the key mechanisms of cognitive flexibility and goal-directed behaviour rely on the acquisition of suitable representations of percepts and their top-down internal manipulation. Moreover, we propose a neuro-inspired computational model to operationalise this hypothesis. The capacity of the model to support cognitive flexibility was validated by systematically reproducing and interpreting the behaviour exhibited in the WCST by young and old healthy adults, and by frontal and Parkinson patients. The results corroborate and further articulate the hypothesis that the internal manipulation of representations is a core process in goal-directed flexible cognition.

Thalamic control of cortical dynamics in a model of flexible motor sequencing

The neural mechanisms that generate an extensible library of motor motifs and flexibly string them into arbitrary sequences are unclear. We developed a model in which inhibitory basal ganglia output neurons project to thalamic units that are themselves bidirectionally connected to a recurrent cortical network. We model the basal ganglia inhibitory patterns as silencing some thalamic neurons while leaving others disinhibited and free to interact with cortex during specific motifs. We show that a small number of disinhibited thalamic neurons can control cortical dynamics to generate specific motor output in a noise-robust way. Additionally, a single "preparatory" thalamocortical network can produce fast cortical dynamics that support rapid transitions between any pair of learned motifs. If the thalamic units associated with each sequence component are segregated, many motor outputs can be learned without interference and then combined in arbitrary orders for the flexible production of long and complex motor sequences.

Basal Ganglia and Thalamic Contributions to Language Function: Insights from A Parallel Distributed Processing Perspective

Cerebral representations are encoded as patterns of activity involving billions of neurons. Parallel distributed processing (PDP) across these neuronal populations provides the basis for a number of emergent properties: 1) processing occurs and knowledge (long term memories) is stored (as synaptic connection strengths) in exactly the same networks; 2) networks have the capacity for setting into stable attractor states corresponding to concepts, symbols, implicit rules, or data transformations; 3) networks provide the scaffold for the acquisition of knowledge but knowledge is acquired through experience; 4) PDP networks are adept at incorporating the statistical regularities of experience as well as frequency and age of acquisition effects; 5) networks enable content-addressable memory; 6) because knowledge is distributed throughout networks, they exhibit the property of graceful degradation; 7) networks intrinsically provide the capacity for inference. This paper details the features of the basal ganglia and thalamic systems (recurrent and distributed connectivity) that support PDP. The PDP lens and an understanding of the attractor trench dynamics of the basal ganglia provide a natural explanation for the peculiar dysfunctions of Parkinson's disease and the mechanisms by which dopamine deficiency is causal. The PDP lens, coupled with the fact that the basal ganglia of humans bears strong homology to the basal ganglia of lampreys and the central complex of arthropods, reveals that the fundamental function of the basal ganglia is computational and involves the reduction of the vast dimensionality of a complex multi-dimensional array of sensorimotor input into the optimal choice from a small repertoire of behavioral options - the essence of reactive intention (automatic responses to sensory input). There is strong evidence that the sensorimotor basal ganglia make no contributions to cognitive or motor function in humans but can cause serious dysfunction when pathological. It appears that humans, through the course of evolution, have developed cortical capacities (working memory and volitional and reactive attention) for managing sensory input, however complex, that obviate the need for the basal ganglia. The functions of the dorsal tier thalamus, however, even viewed with an understanding of the properties of population encoded representations, remain somewhat more obscure. Possibilities include the enabling of attractor state constellations that optimize function by taking advantage of simultaneous input from multiple cortical areas; selective engagement of cortical representations; and support of the gamma frequency synchrony that enables binding of the multiple network representations that comprise a full concept representation.

A Basal Ganglia Computational Model to Explain the Paradoxical Sensorial Improvement in the Presence of Huntington's Disease

The basal ganglia (BG) represent a critical center of the nervous system for sensorial discrimination. Although it is known that Huntington's disease (HD) affects this brain area, it still remains unclear how HD patients achieve paradoxical improvement in sensorial discrimination tasks. This paper presents a computational model of the BG including the main nuclei and the typical firing properties of their neurons. The BG model has been embedded within an auditory signal detection task. We have emulated the effect that the altered levels of dopamine and the degree of HD affectation have in information processing at different layers of the BG, and how these aspects shape transient and steady states differently throughout the selection task. By extracting the independent components of the BG activity at different populations, it is evidenced that early and medium stages of HD affectation may enhance transient activity in the striatum and the substantia nigra pars reticulata. These results represent a possible explanation for the paradoxical improvement that HD patients present in discrimination task performance. Thus, this paper provides a novel understanding on how the fast dynamics of the BG network at different layers interact and enable transient states to emerge throughout the successive neuron populations.

A Robust Model of Gated Working Memory

Gated working memory is defined as the capacity of holding arbitrary information at any time in order to be used at a later time. Based on electrophysiological recordings, several computational models have tackled the problem using dedicated and explicit mechanisms. We propose instead to consider an implicit mechanism based on a random recurrent neural network. We introduce a robust yet simple reservoir model of gated working memory with instantaneous updates. The model is able to store an arbitrary real value at random time over an extended period of time. The dynamics of the model is a line attractor that learns to exploit reentry and a nonlinearity during the training phase using only a few representative values. A deeper study of the model shows that there is actually a large range of hyperparameters for which the results hold (e.g., number of neurons, sparsity, global weight scaling) such that any large enough population, mixing excitatory and inhibitory neurons, can quickly learn to realize such gated working memory. In a nutshell, with a minimal set of hypotheses, we show that we can have a robust model of working memory. This suggests this property could be an implicit property of any random population, that can be acquired through learning. Furthermore, considering working memory to be a physically open but functionally closed system, we give account on some counterintuitive electrophysiological recordings.

Gated spiking neural network using Iterative Free-Energy Optimization and rank-order coding for structure learning in memory sequences (INFERNO GATE)

We present a framework based on iterative free-energy optimization with spiking neural networks for modeling the fronto-striatal system (PFC-BG) for the generation and recall of audio memory sequences. In line with neuroimaging studies carried out in the PFC, we propose a genuine coding strategy using the gain-modulation mechanism to represent abstract sequences based solely on the rank and location of items within them. Based on this mechanism, we show that we can construct a repertoire of neurons sensitive to the temporal structure in sequences from which we can represent any novel sequences. Free-energy optimization is then used to explore and to retrieve the missing indices of the items in the correct order for executive control and compositionality. We show that the gain-modulation mechanism permits the network to be robust to variabilities and to have long-term dependencies as it implements a gated recurrent neural network. This model, called Inferno Gate, is an extension of the neural architecture Inferno standing for Iterative Free-Energy Optimization of Recurrent Neural Networks with Gating or Gain-modulation. In experiments performed with an audio database of ten thousand MFCC vectors, Inferno Gate is capable of encoding efficiently and retrieving chunks of fifty items length. We then discuss the potential of our network to model the features of working memory in the PFC-BG loop for structural learning, goal-direction and hierarchical reinforcement learning.

Different Dopaminergic Dysfunctions Underlying Parkinsonian Akinesia and Tremor

Although the occurrence of Parkinsonian akinesia and tremor is traditionally associated to dopaminergic degeneration, the multifaceted neural processes that cause these impairments are not fully understood. As a consequence, current dopamine medications cannot be tailored to the specific dysfunctions of patients with the result that generic drug therapies produce different effects on akinesia and tremor. This article proposes a computational model focusing on the role of dopamine impairments in the occurrence of akinesia and resting tremor. The model has three key features, to date never integrated in a single computational system: (a) an architecture constrained on the basis of the relevant known system-level anatomy of the basal ganglia-thalamo-cortical loops; (b) spiking neurons with physiologically-constrained parameters; (c) a detailed simulation of the effects of both phasic and tonic dopamine release. The model exhibits a neural dynamics compatible with that recorded in the brain of primates and humans. Moreover, it suggests that akinesia might involve both tonic and phasic dopamine dysregulations whereas resting tremor might be primarily caused by impairments involving tonic dopamine release and the responsiveness of dopamine receptors. These results could lead to develop new therapies based on a system-level view of the Parkinson's disease and targeting phasic and tonic dopamine in differential ways.

Know Your Body Through Intrinsic Goals

The first “object” that newborn children play with is their own body. This activity allows them to autonomously form a sensorimotor map of their own body and a repertoire of actions supporting future cognitive and motor development. Here we propose the theoretical hypothesis, operationalized as a computational model, that this acquisition of body knowledge is not guided by random motor-babbling, but rather by autonomously generated goals formed on the basis of intrinsic motivations. Motor exploration leads the agent to discover and form representations of the possible sensory events it can cause with its own actions. When the agent realizes the possibility of improving the competence to re-activate those representations, it is intrinsically motivated to select and pursue them as goals. The model is based on four components: (1) a self-organizing neural network, modulated by competence-based intrinsic motivations, that acquires abstract representations of experienced sensory (touch) changes; (2) a selector that selects the goal to pursue, and the motor resources to train to pursue it, on the basis of competence improvement; (3) an echo-state neural network that controls and learns, through goal-accomplishment and competence, the agent's motor skills; (4) a predictor of the accomplishment of the selected goals generating the competence-based intrinsic motivation signals. The model is tested as the controller of a simulated simple planar robot composed of a torso and two kinematic 3-DoF 2D arms. The robot explores its body covered by touch sensors by moving its arms. The results, which might be used to guide future empirical experiments, show how the system converges to goals and motor skills allowing it to touch the different parts of own body and how the morphology of the body affects the formed goals. The convergence is strongly dependent on competence-based intrinsic motivations affecting not only skill learning and the selection of formed goals, but also the formation of the goal representations themselves.

The architecture challenge: Future artificial-intelligence systems will require sophisticated architectures, and knowledge of the brain might guide their construction

In this commentary, we highlight a crucial challenge posed by the proposal of Lake et al. to introduce key elements of human cognition into deep neural networks and future artificial-intelligence systems: the need to design effective sophisticated architectures. We propose that looking at the brain is an important means of facing this great challenge.

Interplay of prefrontal cortex and amygdala during extinction of drug seeking

Extinction of Pavlovian conditioning is a complex process that involves brain regions such as the medial prefrontal cortex (mPFC), the amygdala and the locus coeruleus. In particular, noradrenaline (NA) coming from the locus coeruleus has been recently shown to play a different role in two subregions of the mPFC, the prelimbic (PL) and the infralimbic (IL) regions. How these regions interact in conditioning and subsequent extinction is an open issue. We studied these processes using two approaches: computational modelling and NA manipulation in a conditioned place preference paradigm (CPP) in mice. In the computational model, NA in PL and IL causes inputs arriving to these regions to be amplified, thus allowing them to modulate learning processes in amygdala. The model reproduces results from studies involving depletion of NA from PL, IL, or both in CPP. In addition, we simulated new experiments of NA manipulations in mPFC, making predictions on the possible results. We searched the parameters of the model and tested the robustness of the predictions by performing a sensitivity analysis. We also present an empirical experiment where, in accord with the model, a double depletion of NA from both PL and IL in CPP with amphetamine impairs extinction. Overall the proposed model, supported by anatomical, physiological, and behavioural data, explains the differential role of NA in PL and IL and opens up the possibility to understand extinction mechanisms more in depth and hence to aid the development of treatments for disorders such as addiction.

Reward Based Motor Adaptation Mediated by Basal Ganglia

It is widely accepted that the basal ganglia (BG) play a key role in action selection and reinforcement learning. However, despite considerable number of studies, the BG architecture and function are not completely understood. Action selection and reinforcement learning are facilitated by the activity of dopaminergic neurons, which encode reward prediction errors when reward outcomes are higher or lower than expected. The BG are thought to select proper motor responses by gating appropriate actions, and suppressing inappropriate ones. The direct striato-nigral (GO) and the indirect striato-pallidal (NOGO) pathways have been suggested to provide the functions of BG in the two-pathway concept. Previous models confirmed the idea that these two pathways can mediate the behavioral choice, but only for a relatively small number of potential behaviors. Recent studies have provided new evidence of BG involvement in motor adaptation tasks, in which adaptation occurs in a non-error-based manner. In such tasks, there is a continuum of possible actions, each represented by a complex neuronal activity pattern. We extended the classical concept of the two-pathway BG by creating a model of BG interacting with a movement execution system, which allows for an arbitrary number of possible actions. The model includes sensory and premotor cortices, BG, a spinal cord network, and a virtual mechanical arm performing 2D reaching movements. The arm is composed of 2 joints (shoulder and elbow) controlled by 6 muscles (4 mono-articular and 2 bi-articular). The spinal cord network contains motoneurons, controlling the muscles, and sensory interneurons that receive afferent feedback and mediate basic reflexes. Given a specific goal-oriented motor task, the BG network through reinforcement learning constructs a behavior from an arbitrary number of basic actions represented by cortical activity patterns. Our study confirms that, with slight modifications, the classical two-pathway BG concept is consistent with results of previous studies, including non-error based motor adaptation experiments, pharmacological manipulations with BG nuclei, and functional deficits observed in BG-related motor disorders.

Dysfunctions of the basal ganglia-cerebellar-thalamo-cortical system produce motor tics in Tourette syndrome

Motor tics are a cardinal feature of Tourette syndrome and are traditionally associated with an excess of striatal dopamine in the basal ganglia. Recent evidence increasingly supports a more articulated view where cerebellum and cortex, working closely in concert with basal ganglia, are also involved in tic production. Building on such evidence, this article proposes a computational model of the basal ganglia-cerebellar-thalamo-cortical system to study how motor tics are generated in Tourette syndrome. In particular, the model: (i) reproduces the main results of recent experiments about the involvement of the basal ganglia-cerebellar-thalamo-cortical system in tic generation; (ii) suggests an explanation of the system-level mechanisms underlying motor tic production: in this respect, the model predicts that the interplay between dopaminergic signal and cortical activity contributes to triggering the tic event and that the recently discovered basal ganglia-cerebellar anatomical pathway may support the involvement of the cerebellum in tic production; (iii) furnishes predictions on the amount of tics generated when striatal dopamine increases and when the cortex is externally stimulated. These predictions could be important in identifying new brain target areas for future therapies. Finally, the model represents the first computational attempt to study the role of the recently discovered basal ganglia-cerebellar anatomical links. Studying this non-cortex-mediated basal ganglia-cerebellar interaction could radically change our perspective about how these areas interact with each other and with the cortex. Overall, the model also shows the utility of casting Tourette syndrome within a system-level perspective rather than viewing it as related to the dysfunction of a single brain area.

Tourette syndrome is a neuropsychiatric disorder characterized by vocal and motor tics. Tics represent a cardinal symptom traditionally associated with a dysfunction of the basal ganglia leading to an excess of the dopamine neurotransmitter. This view gives a restricted clinical picture and limits therapeutic approaches because it ignores the influence of altered interactions between the basal ganglia and other brain areas. In this respect, recent evidence supports a more articulated framework where cerebellum and cortex are also involved in tic production. Building on these data, we propose a computational model of the basal ganglia-cerebellar-thalamo-cortical network to investigate the specific mechanisms underlying motor tic production. The model reproduces the results of recent experiments and suggests an explanation of the system-level processes underlying tic production. Moreover, it furnishes predictions related to the amount of tics generated when there are dysfunctions in the basal ganglia-cerebellar-thalamo-cortical circuits. These predictions could be important in identifying new brain target areas for future therapies based on a system-level view of Tourette syndrome.

Iterative free-energy optimization for recurrent neural networks (INFERNO)

The intra-parietal lobe coupled with the Basal Ganglia forms a working memory that demonstrates strong planning capabilities for generating robust yet flexible neuronal sequences. Neurocomputational models however, often fails to control long range neural synchrony in recurrent spiking networks due to spontaneous activity. As a novel framework based on the free-energy principle, we propose to see the problem of spikes' synchrony as an optimization problem of the neurons sub-threshold activity for the generation of long neuronal chains. Using a stochastic gradient descent, a reinforcement signal (presumably dopaminergic) evaluates the quality of one input vector to move the recurrent neural network to a desired activity; depending on the error made, this input vector is strengthened to hill-climb the gradient or elicited to search for another solution. This vector can be learned then by one associative memory as a model of the basal-ganglia to control the recurrent neural network. Experiments on habit learning and on sequence retrieving demonstrate the capabilities of the dual system to generate very long and precise spatio-temporal sequences, above two hundred iterations. Its features are applied then to the sequential planning of arm movements. In line with neurobiological theories, we discuss its relevance for modeling the cortico-basal working memory to initiate flexible goal-directed neuronal chains of causation and its relation to novel architectures such as Deep Networks, Neural Turing Machines and the Free-Energy Principle.

Parkinson's disease as a system-level disorder

Traditionally, the basal ganglia have been considered the main brain region implicated in Parkinson’s disease. This single area perspective gives a restricted clinical picture and limits therapeutic approaches because it ignores the influence of altered interactions between the basal ganglia and other cerebral components on Parkinsonian symptoms. In particular, the basal ganglia work closely in concert with cortex and cerebellum to support motor and cognitive functions. This article proposes a theoretical framework for understanding Parkinson’s disease as caused by the dysfunction of the entire basal ganglia–cortex–cerebellum system rather than by the basal ganglia in isolation. In particular, building on recent evidence, we propose that the three key symptoms of tremor, freezing, and impairments in action sequencing may be explained by considering partially overlapping neural circuits including basal ganglia, cortical and cerebellar areas. Studying the involvement of this system in Parkinson’s disease is a crucial step for devising innovative therapeutic approaches targeting it rather than only the basal ganglia. Possible future therapies based on this different view of the disease are discussed.

Active Inference: A Process Theory

This article describes a process theory based on active inference and belief propagation. Starting from the premise that all neuronal processing (and action selection) can be explained by maximizing Bayesian model evidence-or minimizing variational free energy-we ask whether neuronal responses can be described as a gradient descent on variational free energy. Using a standard (Markov decision process) generative model, we derive the neuronal dynamics implicit in this description and reproduce a remarkable range of well-characterized neuronal phenomena. These include repetition suppression, mismatch negativity, violation responses, place-cell activity, phase precession, theta sequences, theta-gamma coupling, evidence accumulation, race-to-bound dynamics, and transfer of dopamine responses. Furthermore, the (approximately Bayes' optimal) behavior prescribed by these dynamics has a degree of face validity, providing a formal explanation for reward seeking, context learning, and epistemic foraging. Technically, the fact that a gradient descent appears to be a valid description of neuronal activity means that variational free energy is a Lyapunov function for neuronal dynamics, which therefore conform to Hamilton's principle of least action.

Goal-Directed Behavior and Instrumental Devaluation: A Neural System-Level Computational Model

Devaluation is the key experimental paradigm used to demonstrate the presence of instrumental behaviors guided by goals in mammals. We propose a neural system-level computational model to address the question of which brain mechanisms allow the current value of rewards to control instrumental actions. The model pivots on and shows the computational soundness of the hypothesis for which the internal representation of instrumental manipulanda (e.g., levers) activate the representation of rewards (or "action-outcomes", e.g., foods) while attributing to them a value which depends on the current internal state of the animal (e.g., satiation for some but not all foods). The model also proposes an initial hypothesis of the integrated system of key brain components supporting this process and allowing the recalled outcomes to bias action selection: (a) the sub-system formed by the basolateral amygdala and insular cortex acquiring the manipulanda-outcomes associations and attributing the current value to the outcomes; (b) three basal ganglia-cortical loops selecting respectively goals, associative sensory representations, and actions; (c) the cortico-cortical and striato-nigro-striatal neural pathways supporting the selection, and selection learning, of actions based on habits and goals. The model reproduces and explains the results of several devaluation experiments carried out with control rats and rats with pre- and post-training lesions of the basolateral amygdala, the nucleus accumbens core, the prelimbic cortex, and the dorso-medial striatum. The results support the soundness of the hypotheses of the model and show its capacity to integrate, at the system-level, the operations of the key brain structures underlying devaluation. Based on its hypotheses and predictions, the model also represents an operational framework to support the design and analysis of new experiments on the motivational aspects of goal-directed behavior.

Active inference and learning

This paper offers an active inference account of choice behaviour and learning. It focuses on the distinction between goal-directed and habitual behaviour and how they contextualise each other. We show that habits emerge naturally (and autodidactically) from sequential policy optimisation when agents are equipped with state-action policies. In active inference, behaviour has explorative (epistemic) and exploitative (pragmatic) aspects that are sensitive to ambiguity and risk respectively, where epistemic (ambiguity-resolving) behaviour enables pragmatic (reward-seeking) behaviour and the subsequent emergence of habits. Although goal-directed and habitual policies are usually associated with model-based and model-free schemes, we find the more important distinction is between belief-free and belief-based schemes. The underlying (variational) belief updating provides a comprehensive (if metaphorical) process theory for several phenomena, including the transfer of dopamine responses, reversal learning, habit formation and devaluation. Finally, we show that active inference reduces to a classical (Bellman) scheme, in the absence of ambiguity.