Computational Neuroscience is an interdisciplinary field that uses mathematical models, computational methods, algorithms and theoretical analysis to understand how nervous systems process information and generate behaviour. It connects neuroscience with computer science, mathematics, physics, engineering and statistics, operating at scales ranging from ion channels and individual neurons to neural circuits, brain regions and whole-brain systems. The central objective is not simply to simulate the brain, but to develop quantitative explanations of how biological neural systems operate.
Computational models can describe electrical activity in neurons, synaptic interactions, network dynamics, sensory processing, learning, memory, decision-making and motor control. Some models are biologically detailed, while others abstract away biological mechanisms to study computational principles. The field therefore includes both mechanistic modelling and theoretical investigation of what neural systems can compute, how they learn and how their structure constrains their function.
The intellectual roots of Computational Neuroscience extend through mathematical biology, cybernetics, information theory and theoretical neuroscience. Early mathematical descriptions of neurons include the Hodgkin-Huxley model, developed from experimental studies of the squid giant axon and published in the early 1950s. The McCulloch-Pitts neuron, proposed earlier, provided an influential abstract model of neural computation. Donald Hebb's work on learning and synaptic change also influenced computational theories of neural organisation.
During the latter twentieth century, increasing computing power enabled increasingly detailed neural simulations, while developments in dynamical systems, statistics and information theory provided new analytical tools. Computational Neuroscience emerged as a recognisable interdisciplinary field as researchers began systematically combining experimental neuroscience with quantitative models. The field subsequently expanded through large-scale neural recording, high-performance computing, machine learning and increasingly detailed simulations of neural circuits and brain systems.
A fundamental topic in Computational Neuroscience is the mathematical representation of neuronal activity. A neuron can be modelled at different levels of abstraction. The Hodgkin-Huxley model represents ionic currents through differential equations and can reproduce detailed aspects of action-potential generation. The integrate-and-fire family of models abstracts the neuron into membrane-potential dynamics followed by a threshold and reset mechanism, making large network simulations computationally cheaper.
Conductance-based models can represent synaptic and membrane currents, while rate-based models represent populations of neurons through average activity rather than individual spikes. These approaches illustrate an important modelling principle: greater biological detail can provide mechanistic insight but increases computational cost and parameter requirements. Model selection therefore depends on the scientific question. A model intended to study ion-channel dynamics requires different detail from one designed to investigate information flow through thousands or millions of neurons.
Computational Neuroscience also studies neural networks as dynamical systems. Neurons communicate through synapses whose strengths, delays and connection patterns influence network behaviour. Recurrent networks can generate oscillations, attractor states, synchronisation and other emergent dynamics. Researchers use differential equations, stochastic processes, graph theory and dynamical-systems analysis to investigate these phenomena. Neural population activity can be studied through firing-rate distributions, correlations, dimensionality reduction and latent-variable models. The connectivity structure of a network can also influence its computational capabilities.
Feed-forward circuits can transform sensory representations, while recurrent circuits can maintain information or generate temporal dynamics. Inhibitory and excitatory interactions contribute to stability and selectivity. Computational models therefore provide a framework for asking not only which neurons are active but how connectivity, timing, feedback and network dynamics collectively produce observable behaviour.
Learning and plasticity are major areas of Computational Neuroscience. Biological learning involves changes in synaptic efficacy and network organisation, and computational models attempt to explain how these changes can produce adaptation, memory and behavioural improvement. Hebbian learning describes a broad principle in which correlated neuronal activity can strengthen synaptic connections, while spike-timing-dependent plasticity models changes according to the relative timing of pre- and postsynaptic spikes.
Reinforcement-learning models examine how organisms can learn from rewards and penalties, and have been used to investigate relationships between behavioural learning and neural signals associated with reward prediction. Bayesian and probabilistic models provide another framework for understanding how nervous systems may represent uncertainty and combine sensory evidence. These theories do not imply that the brain literally implements a particular mathematical algorithm in the same way a computer program does; they provide hypotheses that can be tested against neural and behavioural observations.
Computational Neuroscience has applications to sensory systems, motor control, memory, attention, decision-making and clinical neuroscience. Visual neuroscience, for example, uses computational models to investigate how neural circuits extract information about edges, motion, orientation and objects. Auditory models can examine how temporal and spectral information is transformed by neural pathways. Motor-control models investigate how the brain and spinal systems coordinate movements despite sensory delays and uncertainty. Models of memory can explore how recurrent neural circuits store and retrieve patterns.
In clinical research, computational approaches can help analyse neural recordings, model disease mechanisms and investigate treatments or brain-machine interfaces. Computational Neuroscience also contributes concepts to artificial intelligence, particularly through research into learning, attention, representation and neural dynamics, although biological brains and artificial neural networks differ substantially in structure, energy use, learning mechanisms and embodiment.
A major methodological feature of Computational Neuroscience is the interaction between models and experimental data. A useful model should generate predictions that can be compared with observations rather than merely reproduce data that were already used to construct it. Researchers may fit model parameters to electrophysiological recordings, imaging data or behavioural measurements and then test the model against independent observations. Model comparison can determine whether different hypotheses make distinguishable predictions.
Computational methods can also reveal patterns in large neural datasets that would be difficult to identify manually. However, correlation does not establish a biological mechanism, and a model can fit experimental data for the wrong reasons if it contains too many adjustable parameters. Issues such as identifiability, overfitting, measurement noise, sampling bias and model assumptions therefore matter greatly. Computational Neuroscience combines quantitative sophistication with the requirement that models remain scientifically interpretable and experimentally testable.
Computational Neuroscience differs from artificial intelligence, neural network engineering and brain simulation, although the fields overlap substantially. AI generally seeks to construct systems that perform useful cognitive tasks, whereas Computational Neuroscience seeks to explain biological neural computation, though computational neuroscience ideas can influence AI. Whole-brain simulation is one possible application of computational modelling but represents only a subset of the field.
Neuroinformatics focuses more directly on organising, analysing and sharing neuroscience data, while Computational Neuroscience focuses on quantitative theories and models of neural function. The field's broader significance lies in providing a mathematical language for connecting neural mechanisms with information processing and behaviour. As recording technologies produce increasingly large and detailed datasets, computational methods are becoming essential for integrating observations across scales and for testing hypotheses about how biological neural systems transform information into perception, learning, decisions and action.
