Computational neuroscience is an interdisciplinary field that uses mathematical models, computer simulations, statistical methods and theoretical analysis to understand how nervous systems process information. It sits at the intersection of neuroscience, computer science, mathematics, physics, psychology and engineering. Rather than studying neural systems only through biological observation, computational neuroscience asks what computational principles could explain observed neural activity and behaviour. Researchers may model individual neurons, networks of neurons, brain regions, sensory systems, learning mechanisms or entire behavioural tasks.
The field can investigate questions ranging from how an ion channel influences the firing of a single neuron to how populations of neurons represent objects, navigate environments or make decisions. Computational neuroscience therefore concerns both the mechanisms implemented by nervous systems and the mathematical principles that describe their information processing.
At the cellular level, computational neuroscience often models how neurons transform electrical and chemical inputs into output signals. Neuron models vary in complexity. The Hodgkin-Huxley model describes membrane dynamics using differential equations representing ionic conductances and helped explain the generation of action potentials. Simpler integrate-and-fire models abstract away much of the biophysical detail and represent a neuron as accumulating input until a threshold is reached. Intermediate models can capture adaptation, bursting and other behaviours without modelling every molecular mechanism.
These models allow researchers to investigate how changes in membrane properties, synaptic input or network connectivity affect neuronal activity. The appropriate model depends on the research question: a detailed biophysical model may be useful for studying cellular mechanisms, while a simpler model may be preferable when analysing a large network containing thousands or millions of neurons.
Synapses are another major focus because neurons rarely operate in isolation. A synapse transmits information from one neuron to another and can be excitatory, inhibitory or modulatory depending on the neurotransmitter, receptor and cellular context. Computational models can represent synaptic effects as changes in membrane current, conductance or firing probability. Synaptic plasticity describes changes in synaptic strength over time and is central to theories of learning and memory.
Models such as spike-timing-dependent plasticity alter synaptic strength according to the temporal relationship between pre-synaptic and post-synaptic activity. Other models describe long-term potentiation, long-term depression or homeostatic plasticity. Computational experiments allow researchers to test whether particular plasticity rules can produce stable learning, memory formation, pattern recognition or other observed behaviours.
At the network level, computational neuroscience examines how collections of neurons generate patterns of activity and perform computations. Recurrent neural networks can represent feedback connections, while feedforward networks model directional information flow. Attractor networks can stabilise activity around particular patterns and have been used to model associative memory. Population coding investigates how information may be represented collectively across many neurons rather than by a single specialised neuron.
Sparse coding proposes that representations may involve relatively small subsets of active neurons, while distributed coding allows information to be represented across many cells. Oscillatory activity, synchronisation and temporal coding are additional areas of investigation. These models help bridge the gap between cellular mechanisms and observable behaviour by examining how network organisation gives rise to information-processing properties.
Computational neuroscience also studies how brains represent and process sensory information. Visual neuroscience, for example, includes models of receptive fields, motion detection, object recognition and hierarchical processing. Auditory models investigate frequency representation, temporal patterns and sound localisation. Motor-control models examine how neural systems generate coordinated movements despite uncertainty and changing physical conditions. Decision-making models can represent how evidence accumulates over time and how neural activity relates to choices.
Reinforcement-learning theory has been especially influential because some mathematical mechanisms resemble aspects of biological reward processing. However, computational models are hypotheses about mechanisms rather than proof that the brain literally implements the same equations. Multiple computational models can sometimes reproduce similar behavioural observations while making different predictions about underlying neural processes.
Computational neuroscience differs from artificial intelligence, although the fields have influenced each other extensively. Artificial neural networks were historically inspired partly by biological neurons, but modern machine-learning systems often use mathematical abstractions that differ substantially from biological neural circuits. Computational neuroscience generally aims to explain biological observations, whereas machine learning primarily aims to construct systems that perform useful computational tasks. The relationship is nevertheless bidirectional.
Ideas from neuroscience have influenced neural-network architectures, reinforcement learning and learning algorithms, while machine-learning methods provide tools for analysing neural recordings and developing hypotheses about neural representations. Computational neuroscience also differs from neuroinformatics, which focuses heavily on organising, sharing and analysing neuroscience data, although computational modelling is often an important part of neuroinformatics research.
A major strength of computational neuroscience is that models make explicit predictions. A verbal hypothesis about how neurons might process information can be translated into equations or algorithms, simulated under controlled conditions and compared with experimental observations. Researchers can manipulate variables that would be difficult or impossible to isolate experimentally and examine how changes propagate through a model. However, modelling involves simplification, and a model can produce plausible behaviour for the wrong mechanistic reasons.
Biological nervous systems contain enormous numbers of interacting components, and many parameters remain uncertain. Neural data are also noisy, variable across individuals and strongly dependent on experimental conditions. Consequently, successful computational neuroscience requires iterative interaction between theory, simulation and empirical measurement rather than treating computational output as a direct description of biological truth.
Computational neuroscience has contributed to understanding neural coding, sensory processing, learning, memory, motor control, decision-making and network dynamics, while also providing conceptual foundations for technologies such as neuromorphic computing and biologically inspired artificial intelligence. Its central question is not simply how much information a brain contains, but how physical neural systems transform signals into representations, decisions and behaviour.
By connecting biological mechanisms with mathematical descriptions, the field provides a framework for testing explanations at multiple scales, from ion channels and individual neurons to large networks and behaviour. It remains an active research area because many fundamental questions about neural computation are unresolved, including how distributed representations arise, how memories are stabilised, how consciousness relates to neural activity and how complex cognition emerges from interacting biological circuits.
