Neural Circuits: How Brain Networks Generate Behavior
Author: Ukrainian Psychological Hub · Published: October 12, 2026 · Editorial Policy
A neural circuit is a set of connected neurons whose interactions transform signals into coordinated nervous-system activity. Circuits receive information, modify it through synapses and internal dynamics, and influence other neurons, muscles, glands, and ultimately behavior. The same circuit may participate in different actions depending on context, learning, neuromodulation, and the state of the wider nervous system. Neural circuits are therefore studied as living networks with measurable connections, changing activity, and testable effects on behavior.
Neuroscientists use the term at several scales. A spinal reflex can be described as a small circuit; a cortical microcircuit can contain diverse interconnected cell types; a motor-control system can include cortex, basal ganglia, thalamus, brainstem, cerebellum, and spinal cord. What unites these cases is the explanatory question: how do interactions among connected neural elements produce a reproducible transformation or an organized pattern of activity? The basic anatomical and functional definition is set out in Neuroscience (NCBI Bookshelf), while Luo's review of neuronal circuit architectures explains why patterns of connectivity are central to understanding computation and behavior.
What is a neural circuit?
A neural circuit consists of neurons connected directly or indirectly through chemical synapses, electrical coupling, and intervening neural pathways. Researchers identify its elements by anatomy, cellular properties, activity, and effects on a defined operation. A neuron can belong to many overlapping functional circuits. A circuit is also affected by glial support, blood supply, extracellular chemistry, hormones, and sensory input, even though a wiring diagram typically centers on neurons and synapses.
The smallest useful example is a reflex pathway. Sensory neurons detect a stimulus, interneurons may transform the incoming signal, and motor neurons influence muscular contraction. A more complex circuit integrates multiple inputs, generates activity internally, remembers recent events through changing synaptic efficacy, and feeds results back into its own processing. Networks for cognition, perception, emotion, and voluntary movement often extend across numerous anatomical regions. The boundary of a circuit is therefore chosen to answer a question; it is not always a permanent biological border.
A helpful distinction separates a neuron's physical connections from a circuit's momentary operation. Anatomical connectivity specifies possible routes through which signals can travel. Functional activity describes which neurons are active together under particular conditions. Causal investigation tests whether a route or activity pattern makes a difference to an outcome. A complete circuit explanation integrates all three, as discussed in the 2025 review of brain network organization.
From individual neurons to distributed behavior
The parts of a circuit
A neuron integrates signals at its dendrites and cell body, produces electrical activity according to its membrane properties, and communicates through its axon and synaptic terminals. Excitatory synaptic inputs tend to increase the probability of certain postsynaptic responses; inhibitory inputs tend to decrease it, depending on receptor type, membrane potential, location, and physiological context. In cortex and many other structures, distinct classes of inhibitory interneurons regulate the timing, gain, and selectivity of activity. Neuromodulators such as dopamine, acetylcholine, and norepinephrine alter how cells and synapses respond over different time scales. No neurotransmitter has one fixed psychological meaning across all receptor systems and circuits.
Circuit nodes can be individual cells, specific cell types, functional ensembles, or entire regions. Connections differ in strength, conduction delay, direction, reliability, and plasticity. Convergent connections permit multiple sources to influence one target. Divergent connections distribute output to multiple targets. Local inhibition may sharpen competition, while recurrent excitation or inhibition can shape persistent, oscillatory, or sequential activity. These motifs are found in different combinations throughout the nervous system; their roles depend on cellular parameters and the surrounding network. Luo (2021) reviews common architectures across species and brain regions.
Why connections matter as much as cells
Imagine two networks containing the same neurons but arranged in different ways. One might transmit a transient input directly toward an output, while another routes signals through recurrent loops and maintains a pattern after the original input has disappeared. Both contain neurons capable of generating action potentials; their network organization gives them different capacities. This is why knowing that a region is active during a task does not by itself reveal the computation it performs.
Cells can also be reused. A neuron participating in one movement may help organize another, because its partners, inputs, and excitability vary from one behavioral state to the next. Recent research on neuronal ensembles supports the importance of repeatedly coordinated populations as intermediate functional units. Ensemble membership and stability are empirical questions, and an observed cluster of coactive cells must be distinguished from proof of a unique, permanently assigned function.
How neural circuits process information
Input, transformation, output, and feedback
A conventional circuit analysis begins with an input, identifies how that input changes neuronal activity, and traces possible outputs. Sensory circuits transform physical signals into patterns that downstream circuits can discriminate. Motor circuits combine goals, sensory conditions, and body feedback to coordinate action. Many pathways are recurrent: outputs affect other neurons that return signals to earlier stages, modifying processing while the action unfolds.
Feedback is particularly important because nervous systems operate continuously. When reaching for an object, information about the intended movement interacts with ongoing estimates of hand position, muscle activity, vision, and environmental resistance. Signals circulate among motor cortical areas, subcortical nuclei, brainstem structures, and spinal networks. The result depends on multiple linked operations, with no single region acting as a complete command center. The review by Arber and Costa describes the interaction of brainstem, basal ganglia, and wider motor systems.
A circuit may implement filtering, competition, amplification, integration over time, prediction, pattern completion, state transitions, or rhythmic coordination. These are descriptions of computations supported by particular experiments and models. Researchers must still establish exactly how the biological network performs the proposed operation.
Feedforward and recurrent architectures
In feedforward pathways, signals move through a sequence of processing stages, even when local feedback also exists. Such arrangements are useful models for studying information transformation and transmission. Recurrent pathways create loops: a population may influence itself indirectly through connected neurons. Depending on the balance of excitation, inhibition, conduction delays, and inputs, recurrent activity can stabilize a representation, support an oscillation, amplify a small signal, or destabilize a state.
The scientific value of these categories is that they generate testable predictions. Changing a recurrent connection should affect persistence or transitions in specified ways if the model is correct. The review of attractor and integrator networks explains how such architectures can support memory-like persistence and integration of noisy evidence. Attractor models are influential, although the strength of biological evidence differs across systems and tasks.
Parallel processing and population codes
Behavior rarely relies on a simple chain in which one neuron represents one idea and another neuron triggers one action. Many neurons respond to overlapping features, and the activity of a population can convey information that no single neuron's response adequately captures. Researchers investigate firing rates, spike timing, trial-to-trial variability, correlations, and coordinated trajectories across cells.
Correlations can improve or impair different aspects of encoding and downstream readout, depending on the task and the structure of shared variability. A stronger correlation is therefore not automatically evidence of better communication. Panzeri and colleagues review how structured population correlations shape information encoding, transmission, and readout. Decoding a stimulus from a recorded population shows that relevant information is present in those measurements under tested conditions; it does not establish that another brain region uses the identical decoder.
How neural circuits generate actions and behavioral patterns
Reflexes and spinal motor rhythms
Reflex circuits show how sensory input can evoke organized motor output with comparatively few intermediary elements. Yet even a reflex changes with posture, attention, movement, and descending control. The familiar knee-jerk reflex illustrates a tractable pathway, but it is not a model for all human actions.
Locomotion offers a second classic example. Experiments in several animal systems demonstrate spinal networks capable of generating patterned rhythmic motor activity. These networks interact continuously with proprioceptive sensory feedback and descending signals that help select movement speed and coordination. A 2026 review of central pattern generators updates the conceptual account of vertebrate locomotor circuitry. Such findings explain essential principles of organized movement while leaving room for important species-specific differences and the complexity of human walking.
Action selection and competing possibilities
When an animal or person chooses among possible actions, many circuits contribute: cortical systems integrate goals and sensory evidence; basal ganglia loops participate in selecting and shaping actions; brainstem and spinal systems help implement particular movement patterns; cerebellar interactions contribute to sensorimotor calibration and prediction. These contributions are distributed, state-dependent, and sensitive to learning. There is no complete one-region explanation of choosing, wanting, or deciding.
In experiments, investigators may record activity before an action, identify ensembles associated with alternative responses, and then perturb selected pathways. A good causal model specifies the behavior, the timing, the experimental control, and the circuit elements affected. The evidence for a circuit's role in a mouse initiating a movement supports a claim about that species and task; general claims about human motivation require additional evidence.
Learning and adaptive behavior
Neural circuits change through experience. Synaptic plasticity includes activity-dependent strengthening or weakening of connections, changes in inhibitory regulation, and longer-term adjustments to cellular or network properties. Learning can also change which existing pathways are engaged and how neuromodulatory signals influence them. Plasticity has multiple forms; it cannot be reduced to the slogan that frequently used connections always become stronger.
Neural adaptation operates on several time scales. Brief changes in synaptic release or membrane excitability can alter processing immediately. Longer-lasting changes can follow repetition, prediction errors, injury, or development. Experience-dependent change also depends on sleep, physiological state, and the structure of the learning task. These are experimentally established classes of phenomena, but the exact cellular cause of a complex human habit cannot be inferred from behavior alone. For the psychological analysis of repeated actions, see The Habit Loop: Cue, Routine, Reward, and Scientific Limits.
Neural ensembles, circuit states, and brain networks
An ensemble is a group of neurons that exhibits a recurring coordinated pattern of activity under specified conditions. The ensemble concept connects cell-level physiology to larger circuit behavior. Ensembles may arise during sensory processing, motor planning, memory-related tasks, or spontaneous activity. They can be transient, overlapping, and sensitive to how investigators define a pattern in data.
A network state is a temporary configuration of activity across neurons or areas. A circuit can move among states as the organism wakes, attends, learns, sleeps, or acts. A recorded state may predict aspects of behavior without being the unique cause of that behavior. Experimental perturbation and model comparison are needed to establish mechanism.
At a still larger scale, researchers speak of brain networks such as thalamocortical pathways, corticostriatal loops, or functional networks identified in human imaging. A brain network in an fMRI paper often refers to statistical coupling among regional signals; a microcircuit in an electrophysiology study may refer to identified neurons and synapses. These meanings overlap conceptually but should not be treated as interchangeable measures. Kulkarni and Bassett (2025) emphasize the importance of linking network statistics to biological mechanisms and explanations of causation.
Structural, functional, and effective connectivity
Structural connectivity describes anatomical pathways or synaptic connections. Electron microscopy can resolve individual synaptic contacts in reconstructed tissue; diffusion MRI estimates properties of water movement that support probabilistic inference about larger white-matter pathways. These are very different levels of anatomical resolution, and diffusion tractography does not directly count synapses or prove a specific axonal pathway.
Functional connectivity describes statistical dependence between measured signals. Correlated firing, synchronized electrophysiological activity, or correlated blood-oxygen-level-dependent signals may indicate coordinated activity. Such associations can also reflect shared inputs, common task demands, physiological noise, or analytical choices. Functional coupling alone does not establish direct synapses, direction of information flow, or causal influence.
Effective connectivity refers to model-based estimates of directed influences among neural elements. Its validity depends on the experimental design and the assumptions of the chosen model. Directionality inferred from observational time series is weaker than well-controlled evidence from perturbation. These distinctions are central to interpreting whether a reported brain network is a discovered physical route, a statistical description, or a proposed mechanism.
How scientists discover and test neural circuits
Anatomy, tracing, and connectomics
Researchers map pathways with anatomical tracers, microscopy, molecular labeling, and electron microscopy. Cell-type mapping helps determine which populations send projections to others. Electron-microscopy connectomics can reconstruct enormous numbers of synaptic contacts and support graph-theoretic analyses of motifs and potential information routes. A connectome captures structural detail at the sampled resolution and is not a complete real-time simulation of function.
In 2024, the FlyWire consortium reconstructed an adult fruit fly brain containing 139,255 neurons and approximately 54.5 million synapses. This is a landmark anatomical dataset from one fly brain, with ongoing work needed to link particular routes to activity and behavior. The related network-statistics analysis reports structural organization and hypotheses about integration and signal broadcasting.
In 2025, the MICrONS consortium linked calcium imaging from about 75,000 neurons to electron-microscopy reconstruction of a mouse visual-cortex volume containing more than 200,000 cells and roughly half a billion synapses. This unusually detailed combination of structure and visual responses advances circuit analysis. It concerns a portion of a mouse's visual system under specific experimental conditions; it is not a completed map of an entire human brain or a full explanation of perception.
Electrical recordings
Intracellular and patch-clamp methods measure membrane voltage or membrane currents in selected cells. Extracellular electrodes measure voltage changes generated by nearby neuronal activity and, with suitable methods, permit estimates of spikes from multiple units. High-density recording probes extend coverage across many sites, but spike sorting and assignment to individual neurons introduce uncertainty. Local field potentials reflect summed extracellular electrical activity from multiple sources and require cautious interpretation.
EEG measures electrical potential differences at the scalp, generated largely by coordinated transmembrane currents in neural populations. MEG records associated magnetic fields. Both provide useful temporal information, but identifying the precise cellular generator or synaptic pathway from these measurements is an inverse problem. Neither records every neuron in a network.
Calcium and voltage imaging
Optical methods use indicators whose fluorescence changes with calcium concentration or membrane voltage. Calcium imaging can sample activity patterns across large populations but is an indirect proxy for spiking, with kinetics and noise that vary by sensor and preparation. Voltage indicators provide different temporal and physiological information, with their own trade-offs. The ability to observe hundreds or thousands of cells can reveal coactivity, sequences, and candidate ensembles, but imaging alone usually establishes association rather than behavioral necessity.
MRI, fMRI, and diffusion MRI
Structural MRI delineates anatomy on a macroscopic scale. Diffusion MRI models local diffusion properties and is used to estimate large-scale white-matter organization. Functional MRI measures blood-oxygen-level-dependent signals influenced by neurovascular responses to neural activity. A BOLD response has different temporal characteristics from action potentials and is affected by vascular and physiological variables.
Activation in an fMRI contrast therefore indicates a measured task-related difference under the statistical model, not direct observation of a thought or a demonstration that one area independently causes a complex experience. The problem of reverse inference arises when a psychological state is asserted solely because a broadly involved brain region is active.
Lesions, stimulation, and optogenetics
Lesion studies can show consequences of damage to neural tissue and help identify which systems contribute to a function. However, natural lesions often affect multiple tissues or connections, and recovery can reorganize function. Transcranial magnetic stimulation perturbs activity in targeted cortical regions and their connected networks, offering time-sensitive evidence when appropriate controls are included. It does not selectively activate a single identified synapse.
Optogenetics introduces light-sensitive proteins into selected cells and uses light to alter their activity. In animal research this allows temporally precise testing of the influence of particular populations or projections. Deisseroth's methodological review documents its importance, while a review of optogenetic interpretation describes how imposed synchrony, unintended effects, and circuit compensation can complicate causal claims. Activating a circuit and observing behavior demonstrates influence under that intervention; it does not prove that the same artificially induced pattern ordinarily causes the behavior in an unperturbed organism.
What does a strong causal explanation require?
A rigorous explanation links several kinds of evidence: identified elements and connections, neuronal activity during the relevant behavior, interventions that change predicted features of the system, and models that account for competing results. Necessity and sufficiency claims are specific to the test. If silencing a population disrupts an action, the population may be required under those conditions, but the disruption could also reflect interference with a larger network. If stimulating a population triggers a response, that response may be possible through the stimulated pathway without establishing that the pathway normally acts alone.
Timing matters. Perturbing a circuit during sensory detection may have a different consequence from perturbing it during decision formation, memory consolidation, or motor execution. The same network can also respond differently depending on recent experience, arousal, and ongoing behavior. Replication across preparations and convergence of methods provide stronger grounds for generalization than one striking stimulation result.
A circuit mechanism should explain more than a correlation. Ideally, it predicts how behavior will change when synaptic strengths, cell types, feedback loops, or external inputs are varied. It should identify the relevant species, task, measurement, and uncertainty. Models that successfully decode behavior from signals are valuable but require separate evidence before being interpreted as models of the brain's actual biological readout.
What do circuit studies reveal about cognition and emotion?
Human thinking, memory, emotion, and social behavior emerge through interactions among multiple neural systems and the body in its environment. Neural circuit research helps explain particular operations: maintaining activity over a delay, integrating evidence, learning associations, selecting actions, adapting to changing sensory conditions, and coordinating physiological responses. These findings inform broader psychological models but do not fully explain a person's lived meaning, values, or social circumstances.
For example, threat-related behavior involves interactions among sensory pathways, learning systems, autonomic and endocrine mechanisms, cortical regulation, and prior experience. The amygdala participates in several relevant forms of learning and salience processing, but the complete human experience of fear is not located in one anatomical structure. Readers interested in a clinically focused synthesis can consult Anxiety and the Brain: Threat Processing, Learning, and What Neuroimaging Can Show, which has a distinct clinical and psychological search intent.
Likewise, a reward circuit does not carry an invariant signal equivalent to pleasure or addiction. Reinforcement, incentive motivation, learning, movement, and expectation involve partially overlapping loops and neuromodulatory effects. The broad circuit principles here provide foundations for more specialized analyses of reward prediction and neurotransmitter function. These topics have separate canonical articles within this knowledge network. For a specific example of neural learning computations, see Reward Prediction Error: How the Brain Learns From Outcomes.
Neural circuits and mental health
Circuit-level findings can improve scientific accounts of symptoms and intervention mechanisms. Altered network dynamics, plasticity, or connectivity are studied in relation to numerous psychiatric and neurological conditions. Yet a group difference in average imaging signals cannot automatically be used as a diagnostic classifier for an individual, and one disorder rarely corresponds to a single malfunctioning circuit.
Brain stimulation, pharmacological intervention, psychotherapy, learning, rehabilitation, and environmental changes may influence neural function through different mechanisms. Clinical effectiveness must be established by appropriate patient studies, outcomes, safety evidence, and guidance specific to an intervention. Laboratory demonstration that stimulating neurons changes an animal's behavior is a mechanistic result, not a treatment recommendation.
Trauma-related symptoms offer one example of the difficulty of moving from group-level circuitry to individual clinical explanation. The English Hub article How PTSD Affects the Brain: Neurobiology, Evidence, and Misconceptions covers the condition-specific evidence, while the present article addresses general circuit architecture and causal inference.
Brain networks, artificial neural networks, and AI
Artificial neural networks borrow selected mathematical ideas from neuroscience, such as connected processing units, adjustable weights, recurrent computation, and distributed representations. Their components do not reproduce the full biology of neurons, glia, synapses, neurotransmitters, developmental processes, bodily regulation, or lived environments. Artificial-network performance can inspire testable hypotheses about computation, and neuroscience can motivate new modeling architectures; resemblance between the two does not establish that a machine and a brain use identical mechanisms.
A successful machine-learning model may approximate a behavioral output or predict neural measurements while leaving several biological mechanisms compatible with the result. A stronger connection requires experimentally discriminating predictions about real neural circuits. Circuit neuroscience also constrains computational theories by supplying facts about cellular diversity, connectivity, dynamics, plasticity, and physiology.
Common misconceptions about neural circuits
Myth: Each behavior has one dedicated circuit
Some behaviors depend on comparatively specialized pathways, but most complex behaviors recruit multiple overlapping networks. A circuit can serve different tasks, and different circuits may partly compensate for one another. A named pathway is an investigative unit, not necessarily an exclusive anatomical module for a psychological category.
Myth: Brain regions communicate only along one fixed route
The anatomy constrains which routes are physically possible, while activity, receptor states, short-term synaptic effects, and neuromodulation change which pathways influence processing at a given moment. Signals can also spread through recurrent loops and parallel pathways. The resulting flow depends on state and task.
Myth: A colorful brain scan shows a neural circuit firing
Conventional human fMRI depicts statistically modeled differences in BOLD signals at a macroscopic scale. It does not record action potentials in identified cells or map synapses directly. Neuroimaging results gain explanatory power through convergence with anatomy, electrophysiology, lesions, behavior, and interventions.
Myth: More connectivity always means a better brain
Connectivity is organized for particular functional requirements. Excessive coupling may degrade selectivity, flexibility, or stability; weak coupling may impair communication. Strength, direction, cell type, timing, and network organization are often more informative than the total number of edges in a graph. Even favorable connectivity patterns are not universal across tasks.
Myth: An optogenetic experiment proves a single neuron causes a human emotion
Cell-specific perturbations can establish causal influence in particular experimental systems. Generalization from manipulated animal circuits to subjective human experiences requires independent behavioral and human evidence. The intervention itself may produce neural patterns that differ from those occurring naturally.
Frequently asked questions
Are neural circuits the same as neural networks?
The terms overlap but are used at different levels. A neural circuit commonly emphasizes identifiable biological elements and their connections, often in relation to a specific operation. A brain network can refer to a larger distributed system or a statistical pattern of regional coupling. An artificial neural network is a computational architecture whose nodes and weights are mathematical constructs.
How many neurons make a neural circuit?
There is no universal minimum or maximum. A simple reflex pathway can involve very few neurons; other circuit models span enormous populations and several brain regions. The useful scale is defined by the mechanism being studied and the methods available to identify relevant elements.
How do neural circuits communicate?
Neurons exchange information through chemical synapses, electrical coupling in some circuits, and activity transmitted along axons. Signals are transformed by cellular excitability, synaptic transmission, receptor action, inhibition, recurrent connectivity, and neuromodulation. Hormonal and body-state signals can also change circuit function.
Can neural circuits change?
Yes. Development, learning, aging, injury, and changing physiological conditions influence circuit structure and dynamics. Plasticity includes changes to synaptic function, structural connections, network recruitment, and excitability. The scale, direction, and consequences of a particular change must be measured in context.
Do neural circuits control thoughts and feelings?
They participate in the biological processes that support perception, attention, memory, emotion, decisions, and bodily responses. These processes depend on distributed neural activity and are interpreted in relation to behavior, development, environment, and the person's circumstances. A scan or stimulation result does not establish a complete map from one circuit to a complex human mental state.
Can neuroscientists map every circuit in a human brain?
Current methods map selected anatomical and functional aspects at different scales, with trade-offs between coverage, spatial detail, temporal detail, and invasiveness. The 2024 adult fly connectome and 2025 mouse visual-cortex data represent major achievements, but neither constitutes a complete, dynamically validated human connectome.
Does neural circuit research explain consciousness?
It contributes experimentally testable evidence about activity patterns and network processes associated with perception, awareness, and changes in behavioral state. A complete theory of consciousness requires additional conceptual commitments and convergent evidence. No single mapped circuit has been established as a universal explanation of conscious experience.
Can damaged neural circuits recover?
Outcomes depend on the kind and location of damage, affected pathways, timing, rehabilitation, and individual factors. Surviving circuits may undergo plastic changes and support functional compensation. Recovery potential and effective intervention cannot be inferred from general statements about neuroplasticity alone.
One formal approach to understanding context-sensitive circuit computations is predictive processing in the brain.
Conclusion
Neural circuits provide the biological level at which individual neuronal mechanisms connect to organized perception, action, memory-related processing, and adaptive behavior. Their functions depend jointly on anatomical connections, cellular properties, changing network states, incoming information, and feedback. The most informative studies combine anatomical mapping, activity measurements, interventions, computational models, and carefully defined behavioral tests.
Recent connectomics makes the structural architecture of selected nervous systems visible at unprecedented detail. Its greatest scientific value emerges when those maps are integrated with live measurements and causal tests. The central question remains precise: which interconnected elements, operating through which mechanisms and under which conditions, contribute to a particular transformation of information or behavior?
Related Articles
A concrete example of action-related neural response properties and the distinction between a directly recorded neuron and a functional network is explained in Mirror Neurons: Discovery, Evidence, and Common Misconceptions.
References
Arber, S., & Costa, R. M. (2022). Networking brainstem and basal ganglia circuits for movement. Nature Reviews Neuroscience, 23, 342–360. https://doi.org/10.1038/s41583-022-00581-w
Deisseroth, K. (2015). Optogenetics: 10 years of microbial opsins in neuroscience. Nature Neuroscience, 18, 1213–1225. https://doi.org/10.1038/nn.4091
Dorkenwald, S., et al. (2024). Neuronal wiring diagram of an adult brain. Nature. https://doi.org/10.1038/s41586-024-07558-y
El Manira, A. (2026). Redefining the central pattern generator for vertebrate locomotion. Nature Reviews Neuroscience, 27, 327–344. https://doi.org/10.1038/s41583-026-01029-1
Khona, M., & Fiete, I. R. (2022). Attractor and integrator networks in the brain. Nature Reviews Neuroscience, 23, 744–766. https://doi.org/10.1038/s41583-022-00642-0
Kulkarni, S., & Bassett, D. S. (2025). Toward principles of brain network organization and function. Annual Review of Biophysics, 54, 353–378. https://doi.org/10.1146/annurev-biophys-030722-110624
Lin, A., et al. (2024). Network statistics of the whole-brain connectome of Drosophila. Nature. https://doi.org/10.1038/s41586-024-07968-y
Luo, L. (2021). Architectures of neuronal circuits. Science, 373, eabg7285. https://doi.org/10.1126/science.abg7285
MICrONS Consortium. (2025). Functional connectomics spanning multiple areas of mouse visual cortex. Nature, 640, 435–447. https://doi.org/10.1038/s41586-025-08790-w
Panzeri, S., Moroni, M., Safaai, H., & Harvey, C. D. (2022). The structures and functions of correlations in neural population codes. Nature Reviews Neuroscience, 23, 551–567. https://doi.org/10.1038/s41583-022-00606-4
Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10, 59–63. https://doi.org/10.1016/j.tics.2005.12.004
Purves, D., Augustine, G. J., Fitzpatrick, D., et al. (Eds.). (2001). Neural circuits. In Neuroscience (2nd ed.). NCBI Bookshelf. https://www.ncbi.nlm.nih.gov/books/NBK11154/
Yuste, R., Cossart, R., & Yaksi, E. (2024). Neuronal ensembles: Building blocks of neural circuits. Neuron, 112, 875–892. https://doi.org/10.1016/j.neuron.2023.12.008