Neural code
How is information encoded in the activity of neurons?
Summary
Why it is open
Neurons signal with action potentials, or spikes, and many features of their activity have been shown to carry information about stimuli [9]. Yet in 2017 Stefano Panzeri and colleagues wrote that "there is still no consensus about the neural code for most sensory stimuli in most areas of the nervous system" [9]. Part of the reason, they argue, is that it is often unclear whether the brain uses the information that such a feature carries [9].
The problem is several questions. One is the grain of the code. Adam Jacobs and colleagues contrast coarse codes, such as the number of spikes during a stimulus, which are robust to jitter in spike times but carry little information per cell, with fine codes, in which the individual spike is the unit and downstream neurons must resolve the structure of the spike train [8]. A second is whether information lies in single cells or in populations [1, 5], and a third is how sparse the code is [4, 7].
Some researchers question the framing itself. In a 2019 target article, published with open peer commentary, Romain Brette argues that the coding metaphor is "often inappropriate and misleading": a neural code depends on experimental details that the coding variable does not carry, and spikes are timed actions that mediate coupling in a distributed dynamical system, not observables tied to an experiment [11].
No body sets a bar for a resolution. Panzeri and colleagues propose one working criterion: the neural code is the set of response features that carry sensory information which the animal uses to drive behaviour, identified by combining recordings during behaviour with interventions on neural activity [9].
What is known
Rate or timing in cortex (1993–1998). William Softky and Christof Koch found that the firing of neurons in visual areas V1 and MT of awake macaque monkeys was nearly as irregular as a random process, whereas model neurons summing many small, independent inputs fired far more regularly [2]. Only fast dendritic nonlinearities or strong synchronisation of inputs brought the models close to the real variability, and the neuron then acted more as a millisecond-resolution detector of coincidences than as an integrator [2]. In slices of rat neocortex, Zachary Mainen and Terrence Sejnowski found that constant stimuli produced imprecise spike trains, while fluctuating stimuli resembling synaptic activity produced spike timing reproducible to less than 1 millisecond [3]. Michael Shadlen and William Newsome showed in a model that balanced excitation and inhibition reproduces the variability, inferred that cortical neurons are "unlikely to transmit information in the temporal pattern of spike discharge", and proposed rate codes in column-like ensembles of 50–100 neurons [5]. Whether cortical columns share one computation is the separate problem of canonical cortical computation.
Timing at the periphery (2004–2009). Roland Johansson and Ingvars Birznieks found that the relative timing of the first spikes in ensembles of human tactile afferents, the nerve fibres from the fingertip, conveys the direction of fingertip force and the shape of the touched surface faster than "the fastest possible rate code" [6]. In mouse retina, Jacobs and colleagues recorded the output cells an animal uses to solve a visual task and decoded their spike trains optimally: a spike-count code fell short of the animal's performance, as did a code treating each spike as independent, while a code that took account of the time since the previous spike did not [8].
Populations and sparse codes (1986–2005). Apostolos Georgopoulos, Andrew Schwartz and Ronald Kettner found that single neurons in the arm area of monkey motor cortex are only broadly tuned to movement direction, but their population vector, the sum of each cell's preferred direction weighted by its change in activity, pointed in the direction of movement [1]. Bruno Olshausen and David Field showed that a learning algorithm that seeks sparse codes for natural images develops receptive fields like those of simple cells in primary visual cortex: localised, oriented and selective for scale [4]. Rodrigo Quian Quiroga and colleagues found neurons in the human medial temporal lobe that responded to strikingly different pictures of a given person, landmark or object, sometimes even to the written name, which they read as an "invariant, sparse and explicit code" [7].
Large-scale recordings (2017–2025). Neuropixels probes, described by James Jun and colleagues in 2017, have 384 recording channels on a 10-mm shank; two probes recorded more than 700 well-isolated neurons at once from five structures in an awake mouse [10]. Carsen Stringer and colleagues recorded more than 10,000 neurons in mouse visual cortex: spontaneous activity encoded a high-dimensional state, partly related to the mouse's ongoing behaviour and present across the forebrain, and visual stimuli added their representation in orthogonal dimensions [12]. In 2024 Jason Manley and colleagues, imaging up to 1 million neurons across mouse cortex at cellular resolution, reported what they call an unbounded scaling of dimensionality with the number of neurons; half the variance lay in 16 dimensions correlated with behaviour, and the further dimensions had no immediate behavioural or sensory correlates [13]. In 2025 the International Brain Laboratory reported 621,733 neurons recorded with 699 Neuropixels probes in 139 mice, covering 279 brain areas during one decision task; responses correlated with impending movement "almost everywhere in the brain" [14]. In the same programme, Charles Findling and colleagues found the mice's estimate of a prior probability encoded in at least 20% to 30% of brain regions, a result they read as consistent with Bayesian inference involving loops between areas [15]; see the Bayesian brain.
Main approaches
Rate codes. Information lies in spike counts, averaged over time or over a pool of similar neurons [5]. Strongest evidence: balanced excitation and inhibition reproduce cortical variability, and an ensemble of 100 neurons gives a reliable estimate of rate within one interspike interval of 10–50 milliseconds [5]; population vectors in monkey motor cortex predict movement direction [1]. Main difficulty: in mouse retina spike counts fall short of behaviour [8], and human tactile signals are used faster than rate codes readily explain [6].
Temporal codes. The timing of spikes, relative to each other or to an event, carries information [6, 8]. Strongest evidence: sub-millisecond reliability in rat cortical slices [3], first-spike order in human tactile afferents [6] and the retinal decoding results [8]. Main difficulty: in Shadlen and Newsome's model, detailed information about the timing of a neuron's inputs cannot be recovered from its output spikes [5], and information present in timing does not show that the brain reads it [9]. Synchrony as a signal that features belong to one object is treated on the page on the binding problem.
Population and sparse codes. Information lies in the joint activity of many neurons [1], possibly in a sparse code [4, 7]. Strongest evidence: the motor population vector [1], sparse-code receptive fields [4] and invariant single neurons in humans [7]. Main difficulty: the 1996 result shows only that maximising sparseness is sufficient to produce such receptive fields in a model [4], and for any population code, information that an experimenter's decoder finds need not be what the brain uses [9].
Codes defined by use. A feature counts as code only if the information it carries is read out for behaviour, tested by relating sensory and choice information in recordings and by intervening on activity [9]. Strongest evidence: comparing decoded codes with behaviour can rule codes out [8]. Main difficulty: Brette argues that the coding metaphor as such is incompatible with both the causal structure of the brain and the representational requirements of cognition [11].
Open sub-questions
- Fine codes in cortex. Does a code that accounts for the time since each neuron's previous spike, which did not fall short of mouse behaviour in the retina [8], also outperform spike counts when decoding visual cortex during the same task?
- First spikes downstream. Do neurons in somatosensory cortex preserve the relative first-spike order of tactile afferents [6] within the time available for adjusting grip?
- Intersection at scale. In the International Brain Laboratory data, which regions carry stimulus information that also predicts the mouse's choice, in Panzeri et al.'s sense [9, 14]?
- Behaviour in sensory codes. Is the behavioural signal in visual cortex [12] orthogonal to the stimulus for decoders read from downstream areas, or does it bias them?
- Dimensionality and method. Does the growth of dimensionality with neuron number [13] hold in electrophysiological recordings with spike timing resolution?
- How sparse. What fraction of human medial temporal lobe neurons respond to more than one unrelated concept, and does it fit an invariant, sparse code [7]?
Work on this in Papex
If you have a testable answer to one of the sub-questions above, you can draft it as a private hypothesis in Papex, or prepare the draft from Claude or ChatGPT with the Papex connector. You review the exact version, sign it and decide whether to publish it; a public record that names this problem is listed on this page. A Papex record shows who signed which version and when. It is not peer review and does not mean the hypothesis is correct.
Key references
- Georgopoulos, A. P., Schwartz, A. B. & Kettner, R. E. (1986). Neuronal population coding of movement direction. Science 233(4771), 1416–1419. 10.1126/science.3749885 · Checked 15 Oct 2026
- Softky, W. R. & Koch, C. (1993). The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs. The Journal of Neuroscience 13(1), 334–350. 10.1523/JNEUROSCI.13-01-00334.1993 · Checked 15 Oct 2026
- Mainen, Z. F. & Sejnowski, T. J. (1995). Reliability of spike timing in neocortical neurons. Science 268(5216), 1503–1506. 10.1126/science.7770778 · Checked 15 Oct 2026
- Olshausen, B. A. & Field, D. J. (1996). Emergence of simple-cell receptive field properties by learning a sparse code for natural images. Nature 381(6583), 607–609. 10.1038/381607a0 · Checked 15 Oct 2026
- Shadlen, M. N. & Newsome, W. T. (1998). The variable discharge of cortical neurons: implications for connectivity, computation, and information coding. The Journal of Neuroscience 18(10), 3870–3896. 10.1523/JNEUROSCI.18-10-03870.1998 · Checked 15 Oct 2026
- Johansson, R. S. & Birznieks, I. (2004). First spikes in ensembles of human tactile afferents code complex spatial fingertip events. Nature Neuroscience 7(2), 170–177. 10.1038/nn1177 · Checked 15 Oct 2026
- Quiroga, R. Q., Reddy, L., Kreiman, G., Koch, C. & Fried, I. (2005). Invariant visual representation by single neurons in the human brain. Nature 435(7045), 1102–1107. 10.1038/nature03687 · Checked 15 Oct 2026
- Jacobs, A. L., Fridman, G., Douglas, R. M., Alam, N. M., Latham, P. E., Prusky, G. T. & Nirenberg, S. (2009). Ruling out and ruling in neural codes. Proceedings of the National Academy of Sciences 106(14), 5936–5941. 10.1073/pnas.0900573106 · Checked 15 Oct 2026
- Panzeri, S., Harvey, C. D., Piasini, E., Latham, P. E. & Fellin, T. (2017). Cracking the neural code for sensory perception by combining statistics, intervention, and behavior. Neuron 93(3), 491–507. 10.1016/j.neuron.2016.12.036 · Checked 15 Oct 2026
- Jun, J. J., Steinmetz, N. A., Siegle, J. H., Denman, D. J., Bauza, M., Barbarits, B. et al. (2017). Fully integrated silicon probes for high-density recording of neural activity. Nature 551(7679), 232–236. 10.1038/nature24636 · Checked 15 Oct 2026
- Brette, R. (2019). Is coding a relevant metaphor for the brain? Behavioral and Brain Sciences 42, e215 (published online 2018). 10.1017/S0140525X19000049 · Checked 15 Oct 2026
- Stringer, C., Pachitariu, M., Steinmetz, N., Reddy, C. B., Carandini, M. & Harris, K. D. (2019). Spontaneous behaviors drive multidimensional, brainwide activity. Science 364(6437), eaav7893. 10.1126/science.aav7893 · Checked 15 Oct 2026
- Manley, J., Lu, S., Barber, K., Demas, J., Kim, H., Meyer, D., Traub, F. M. & Vaziri, A. (2024). Simultaneous, cortex-wide dynamics of up to 1 million neurons reveal unbounded scaling of dimensionality with neuron number. Neuron 112(10), 1694–1709.e5. 10.1016/j.neuron.2024.02.011 · Checked 15 Oct 2026
- International Brain Laboratory, Angelaki, D., Benson, B., Benson, J., Birman, D., Bonacchi, N. et al. (2025). A brain-wide map of neural activity during complex behaviour. Nature 645(8079), 177–191. 10.1038/s41586-025-09235-0 · Checked 15 Oct 2026
- Findling, C., Hubert, F., International Brain Laboratory, Acerbi, L., Benson, B., Benson, J. et al. (2025). Brain-wide representations of prior information in mouse decision-making. Nature 645(8079), 192–200. 10.1038/s41586-025-09226-1 · Checked 15 Oct 2026
Source
Open, as listed by Wikipedia · Unsolved problems in neuroscience. Checked 29 Sep 2026.
The status is the source’s. Papex does not decide whether a problem is solved.
Records on this problem
No public record addresses this problem yet.