Searching for principles of brain computation

The geometry of efficient codes: How rate-distortion trade-offs distort the latent representations of generative models

Living organisms rely on internal models of the world to act adaptively. These models, because of resource limitations, cannot encode every detail and hence need to compress information. From a cognitive standpoint, information compression can manifest as a distortion of latent representations, resulting in the emergence of representations that may not accurately reflect the external world or its geometry. Rate-distortion theory formalizes the optimal way to compress information while minimizing such distortions, by considering factors such as capacity limitations, the frequency and the utility of stimuli. However, while this theory explains why the above factors distort latent representations, it does not specify which specific distortions they produce. To address this question, here we investigate how rate-distortion trade-offs shape the latent representations of images in generative models, specifically Beta Variational Autoencoders ([Formula: see text]-VAEs), under varying constraints of model capacity, data distributions, and task objectives. By systematically exploring these factors, we identify three primary distortions in latent representations: prototypization, specialization, and orthogonalization. These distortions emerge as signatures of information compression, reflecting the model's adaptation to capacity limitations, data imbalances, and task demands. Additionally, our findings demonstrate that these distortions can coexist, giving rise to a rich landscape of latent spaces, whose geometry could differ significantly across generative models subject to different constraints. Our findings contribute to explain how the normative constraints of rate-distortion theory shape the geometry of latent representations of generative models of artificial systems and living organisms.

Biological computation through recurrence

One of the defining features of living systems is their adaptability to changing environmental conditions. This requires organisms to extract temporal and spatial features of their environment, and use that information to compute the appropriate response. In the last two decades, a growing body of work, mainly coming from the machine learning and computational neuroscience fields, has shown that such complex information processing can be performed by recurrent networks. Temporal computations arise in these networks through the interplay between the external stimuli and the network's internal state. In this article we review our current understanding of how recurrent networks can be used by biological systems, from cells to brains, for complex information processing. Rather than focusing on sophisticated, artificial recurrent architectures such as long short-term memory (LSTM) networks, here we concentrate on simpler network structures and learning algorithms that can be expected to have been found by evolution. We also review studies showing evidence of naturally occurring recurrent networks in living organisms. Lastly, we discuss some relevant evolutionary aspects concerning the emergence of this natural computation paradigm.

A Brain-Inspired Approach for Probabilistic Estimation and Efficient Planning in Precision Physical Interaction

This article presents a novel structure of spiking neural networks (SNNs) to simulate the joint function of multiple brain regions in handling precision physical interactions. This task desires efficient movement planning while considering contact prediction and fast radial compensation. Contact prediction demands the cognitive memory of the interaction model, and we novelly propose a double recurrent network to imitate the hippocampus, addressing the spatiotemporal property of the distribution. Radial contact response needs rich spatial information, and we use a cerebellum-inspired module to achieve temporally dynamic prediction. We also use a block-based feedforward network to plan movements, behaving like the prefrontal cortex. These modules are integrated to realize the joint cognitive function of multiple brain regions in prediction, controlling, and planning. We present an appropriate controller and planner to generate teaching signals and provide a feasible network initialization for reinforcement learning, which modifies synapses in accordance with reality. The experimental results demonstrate the validity of the proposed method.

Neurodynamical Computing at the Information Boundaries of Intelligent Systems

Artificial intelligence has not achieved defining features of biological intelligence despite models boasting more parameters than neurons in the human brain. In this perspective article, we synthesize historical approaches to understanding intelligent systems and argue that methodological and epistemic biases in these fields can be resolved by shifting away from cognitivist brain-as-computer theories and recognizing that brains exist within large, interdependent living systems. Integrating the dynamical systems view of cognition with the massive distributed feedback of perceptual control theory highlights a theoretical gap in our understanding of nonreductive neural mechanisms. Cell assemblies-properly conceived as reentrant dynamical flows and not merely as identified groups of neurons-may fill that gap by providing a minimal supraneuronal level of organization that establishes a neurodynamical base layer for computation. By considering information streams from physical embodiment and situational embedding, we discuss this computational base layer in terms of conserved oscillatory and structural properties of cortical-hippocampal networks. Our synthesis of embodied cognition, based in dynamical systems and perceptual control, aims to bypass the neurosymbolic stalemates that have arisen in artificial intelligence, cognitive science, and computational neuroscience.

A Brain-Inspired Approach for Collision-Free Movement Planning in the Small Operational Space

In a small operational space, e.g., mesoscale or microscale, we need to control movements carefully because of fragile objects. This article proposes a novel structure based on spiking neural networks to imitate the joint function of multiple brain regions in visual guiding in the small operational space and offers two channels to achieve collision-free movements. For the state sensation, we simulate the primary visual cortex to directly extract features from multiple input images and the high-level visual cortex to obtain the object distance, which is indirectly measurable, in the Cartesian coordinates. Our approach emulates the prefrontal cortex from two aspects: multiple liquid state machines to predict distances of the next several steps based on the preceding trajectory and a block-based excitation-inhibition feedforward network to plan movements considering the target and prediction. Responding to "too close" states needs rich temporal information, and we leverage a cerebellar network for the subconscious reaction. From the viewpoint of the inner pathway, they also form two channels. One channel starts from state extraction to attraction movement planning, both in the camera coordinates, behaving visual-servo control. The other is the collision-avoidance channel, which calculates distances, predicts trajectories, and reacts to the repulsion, all in the Cartesian coordinates. We provide appropriate supervised signals for coarse training and apply reinforcement learning to modify synapses in accordance with reality. Simulation and experiment results validate the proposed method.

The Brain-Computer Metaphor Debate Is Useless: A Matter of Semantics

It is commonly assumed that usage of the word “computer” in the brain sciences reflects a metaphor. However, there is no single definition of the word “computer” in use. In fact, based on the usage of the word “computer” in computer science, a computer is merely some physical machinery that can in theory compute any computable function. According to this definition the brain is literally a computer; there is no metaphor. But, this deviates from how the word “computer” is used in other academic disciplines. According to the definition used outside of computer science, “computers” are human-made devices that engage in sequential processing of inputs to produce outputs. According to this definition, brains are not computers, and arguably, computers serve as a weak metaphor for brains. Thus, we argue that the recurring brain-computer metaphor debate is actually just a semantic disagreement, because brains are either literally computers or clearly not very much like computers at all, depending on one's definitions. We propose that the best path forward is simply to put the debate to rest, and instead, have researchers be clear about which definition they are using in their work. In some circumstances, one can use the definition from computer science and simply ask, what type of computer is the brain? In other circumstances, it is important to use the other definition, and to clarify the ways in which our brains are radically different from the laptops, smartphones, and servers that surround us in modern life.

The dimensionality of neural representations for control

Cognitive control allows us to think and behave flexibly based on our context and goals. At the heart of theories of cognitive control is a control representation that enables the same input to produce different outputs contingent on contextual factors. In this review, we focus on an important property of the control representation's neural code: its representational dimensionality. Dimensionality of a neural representation balances a basic separability/generalizability trade-off in neural computation. We will discuss the implications of this trade-off for cognitive control. We will then briefly review current neuroscience findings regarding the dimensionality of control representations in the brain, particularly the prefrontal cortex. We conclude by highlighting open questions and crucial directions for future research.

Inferring Neuronal Couplings From Spiking Data Using a Systematic Procedure With a Statistical Criterion

Recent remarkable advances in experimental techniques have provided a background for inferring neuronal couplings from point process data that include a great number of neurons. Here, we propose a systematic procedure for pre- and postprocessing generic point process data in an objective manner to handle data in the framework of a binary simple statistical model, the Ising or generalized McCulloch-Pitts model. The procedure has two steps: (1) determining time bin size for transforming the point process data into discrete-time binary data and (2) screening relevant couplings from the estimated couplings. For the first step, we decide the optimal time bin size by introducing the null hypothesis that all neurons would fire independently, then choosing a time bin size so that the null hypothesis is rejected with the strict criteria. The likelihood associated with the null hypothesis is analytically evaluated and used for the rejection process. For the second postprocessing step, after a certain estimator of coupling is obtained based on the preprocessed data set (any estimator can be used with the proposed procedure), the estimate is compared with many other estimates derived from data sets obtained by randomizing the original data set in the time direction. We accept the original estimate as relevant only if its absolute value is sufficiently larger than those of randomized data sets. These manipulations suppress false positive couplings induced by statistical noise. We apply this inference procedure to spiking data from synthetic and in vitro neuronal networks. The results show that the proposed procedure identifies the presence or absence of synaptic couplings fairly well, including their signs, for the synthetic and experimental data. In particular, the results support that we can infer the physical connections of underlying systems in favorable situations, even when using a simple statistical model.

Fate of Duplicated Neural Structures

Statistical physics determines the abundance of different arrangements of matter depending on cost-benefit balances. Its formalism and phenomenology percolate throughout biological processes and set limits to effective computation. Under specific conditions, self-replicating and computationally complex patterns become favored, yielding life, cognition, and Darwinian evolution. Neurons and neural circuits sit at a crossroads between statistical physics, computation, and (through their role in cognition) natural selection. Can we establish a statistical physics of neural circuits? Such theory would tell what kinds of brains to expect under set energetic, evolutionary, and computational conditions. With this big picture in mind, we focus on the fate of duplicated neural circuits. We look at examples from central nervous systems, with stress on computational thresholds that might prompt this redundancy. We also study a naive cost-benefit balance for duplicated circuits implementing complex phenotypes. From this, we derive phase diagrams and (phase-like) transitions between single and duplicated circuits, which constrain evolutionary paths to complex cognition. Back to the big picture, similar phase diagrams and transitions might constrain I/O and internal connectivity patterns of neural circuits at large. The formalism of statistical physics seems to be a natural framework for this worthy line of research.

Learning spatiotemporal signals using a recurrent spiking network that discretizes time

Learning to produce spatiotemporal sequences is a common task that the brain has to solve. The same neurons may be used to produce different sequential behaviours. The way the brain learns and encodes such tasks remains unknown as current computational models do not typically use realistic biologically-plausible learning. Here, we propose a model where a spiking recurrent network of excitatory and inhibitory spiking neurons drives a read-out layer: the dynamics of the driver recurrent network is trained to encode time which is then mapped through the read-out neurons to encode another dimension, such as space or a phase. Different spatiotemporal patterns can be learned and encoded through the synaptic weights to the read-out neurons that follow common Hebbian learning rules. We demonstrate that the model is able to learn spatiotemporal dynamics on time scales that are behaviourally relevant and we show that the learned sequences are robustly replayed during a regime of spontaneous activity.

Regulation of gamma-frequency oscillation by feedforward inhibition: A computational modeling study

Throughout the brain, reciprocally connected excitatory and inhibitory neurons interact to produce gamma-frequency oscillations. The emergent gamma rhythm synchronizes local neural activity and helps to select which cells should fire in each cycle. We previously found that such excitation-inhibition microcircuits, however, have a potentially significant caveat: the frequency of the gamma oscillation and the level of selection (i.e., the percentage of cells that are allowed to fire) vary with the magnitude of the input signal. In networks with varying levels of brain activity, such a feature may produce undesirable instability on the time and spatial structure of the neural signal with a potential for adversely impacting important neural processing mechanisms. Here we propose that feedforward inhibition solves the latter instability problem of the excitation-inhibition microcircuit. Using computer simulations, we show that the feedforward inhibitory signal reduces the dependence of both the frequency of population oscillation and the level of selection on the magnitude of the input excitation. Such a mechanism can produce stable gamma oscillations with its frequency determined only by the properties of the feedforward network, as observed in the hippocampus. As feedforward and feedback inhibition motifs commonly appear together in the brain, we hypothesize that their interaction underlies a robust implementation of general computational principles of neural processing involved in several cognitive tasks, including the formation of cell assemblies and the routing of information between brain areas.

Stochasticity from function - Why the Bayesian brain may need no noise

An increasing body of evidence suggests that the trial-to-trial variability of spiking activity in the brain is not mere noise, but rather the reflection of a sampling-based encoding scheme for probabilistic computing. Since the precise statistical properties of neural activity are important in this context, many models assume an ad-hoc source of well-behaved, explicit noise, either on the input or on the output side of single neuron dynamics, most often assuming an independent Poisson process in either case. However, these assumptions are somewhat problematic: neighboring neurons tend to share receptive fields, rendering both their input and their output correlated; at the same time, neurons are known to behave largely deterministically, as a function of their membrane potential and conductance. We suggest that spiking neural networks may have no need for noise to perform sampling-based Bayesian inference. We study analytically the effect of auto- and cross-correlations in functional Bayesian spiking networks and demonstrate how their effect translates to synaptic interaction strengths, rendering them controllable through synaptic plasticity. This allows even small ensembles of interconnected deterministic spiking networks to simultaneously and co-dependently shape their output activity through learning, enabling them to perform complex Bayesian computation without any need for noise, which we demonstrate in silico, both in classical simulation and in neuromorphic emulation. These results close a gap between the abstract models and the biology of functionally Bayesian spiking networks, effectively reducing the architectural constraints imposed on physical neural substrates required to perform probabilistic computing, be they biological or artificial.

Evolutionary aspects of reservoir computing

Reservoir computing (RC) is a powerful computational paradigm that allows high versatility with cheap learning. While other artificial intelligence approaches need exhaustive resources to specify their inner workings, RC is based on a reservoir with highly nonlinear dynamics that does not require a fine tuning of its parts. These dynamics project input signals into high-dimensional spaces, where training linear readouts to extract input features is vastly simplified. Thus, inexpensive learning provides very powerful tools for decision-making, controlling dynamical systems, classification, etc. RC also facilitates solving multiple tasks in parallel, resulting in a high throughput. Existing literature focuses on applications in artificial intelligence and neuroscience. We review this literature from an evolutionary perspective. RC's versatility makes it a great candidate to solve outstanding problems in biology, which raises relevant questions. Is RC as abundant in nature as its advantages should imply? Has it evolved? Once evolved, can it be easily sustained? Under what circumstances? (In other words, is RC an evolutionarily stable computing paradigm?) To tackle these issues, we introduce a conceptual morphospace that would map computational selective pressures that could select for or against RC and other computing paradigms. This guides a speculative discussion about the questions above and allows us to propose a solid research line that brings together computation and evolution with RC as test model of the proposed hypotheses. This article is part of the theme issue 'Liquid brains, solid brains: How distributed cognitive architectures process information'.

Leveraging heterogeneity for neural computation with fading memory in layer 2/3 cortical microcircuits

Complexity and heterogeneity are intrinsic to neurobiological systems, manifest in every process, at every scale, and are inextricably linked to the systems' emergent collective behaviours and function. However, the majority of studies addressing the dynamics and computational properties of biologically inspired cortical microcircuits tend to assume (often for the sake of analytical tractability) a great degree of homogeneity in both neuronal and synaptic/connectivity parameters. While simplification and reductionism are necessary to understand the brain's functional principles, disregarding the existence of the multiple heterogeneities in the cortical composition, which may be at the core of its computational proficiency, will inevitably fail to account for important phenomena and limit the scope and generalizability of cortical models. We address these issues by studying the individual and composite functional roles of heterogeneities in neuronal, synaptic and structural properties in a biophysically plausible layer 2/3 microcircuit model, built and constrained by multiple sources of empirical data. This approach was made possible by the emergence of large-scale, well curated databases, as well as the substantial improvements in experimental methodologies achieved over the last few years. Our results show that variability in single neuron parameters is the dominant source of functional specialization, leading to highly proficient microcircuits with much higher computational power than their homogeneous counterparts. We further show that fully heterogeneous circuits, which are closest to the biophysical reality, owe their response properties to the differential contribution of different sources of heterogeneity.

Inhibitory connectivity defines the realm of excitatory plasticity

Recent experiments demonstrate substantial volatility of excitatory connectivity in the absence of any learning. This challenges the hypothesis that stable synaptic connections are necessary for long-term maintenance of acquired information. Here we measure ongoing synaptic volatility and use theoretical modeling to study its consequences on cortical dynamics. We show that in the balanced cortex, patterns of neural activity are primarily determined by inhibitory connectivity, despite the fact that most synapses and neurons are excitatory. Similarly, we show that the inhibitory network is more effective in storing memory patterns than the excitatory one. As a result, network activity is robust to ongoing volatility of excitatory synapses, as long as this volatility does not disrupt the balance between excitation and inhibition. We thus hypothesize that inhibitory connectivity, rather than excitatory, controls the maintenance and loss of information over long periods of time in the volatile cortex.

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.

Toward an Integration of Deep Learning and Neuroscience

Neuroscience has focused on the detailed implementation of computation, studying neural codes, dynamics and circuits. In machine learning, however, artificial neural networks tend to eschew precisely designed codes, dynamics or circuits in favor of brute force optimization of a cost function, often using simple and relatively uniform initial architectures. Two recent developments have emerged within machine learning that create an opportunity to connect these seemingly divergent perspectives. First, structured architectures are used, including dedicated systems for attention, recursion and various forms of short- and long-term memory storage. Second, cost functions and training procedures have become more complex and are varied across layers and over time. Here we think about the brain in terms of these ideas. We hypothesize that (1) the brain optimizes cost functions, (2) the cost functions are diverse and differ across brain locations and over development, and (3) optimization operates within a pre-structured architecture matched to the computational problems posed by behavior. In support of these hypotheses, we argue that a range of implementations of credit assignment through multiple layers of neurons are compatible with our current knowledge of neural circuitry, and that the brain's specialized systems can be interpreted as enabling efficient optimization for specific problem classes. Such a heterogeneously optimized system, enabled by a series of interacting cost functions, serves to make learning data-efficient and precisely targeted to the needs of the organism. We suggest directions by which neuroscience could seek to refine and test these hypotheses.