Millisecond encoding precision of auditory cortex neurons

Millisecond encoding precision of auditory cortex neurons Kayser Logothetis Panzeri

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Millisecond encoding precision of auditory cortex neurons 2010 Proceedings of the National Academy of Sciences 10.1073/pnas.1012656107 Kayser2010-millisecond-encoding-precision-of-auditory-c Kayser2010-millisecond-encoding-precision-of-auditory-c.txt, the project's staged copy efc54f8504fe40ad6f741790fe035df9d0305a00 not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them whether individual auditory cortex neurons carry stimulus information in spike times precise to a few milliseconds, and what feature of the spike train carries it Quotation withheld because this record is not cleared for redistribution. how precisely must spike times be registered before stimulus information is lost Quotation withheld because this record is not cleared for redistribution. shows that registering spikes coarser than a few milliseconds loses information, that the carrier is fast modulation of firing rate rather than spike-time correlations, and that the precision depends on the stimulus dynamics Quotation withheld because this record is not cleared for redistribution. existence Quotation withheld because this record is not cleared for redistribution. millisecond spike-timing precision in auditory cortex Quotation withheld because this record is not cleared for redistribution. electrophysiology electrophysiology Quotation withheld because this record is not cleared for redistribution. mechanisms Faster stimulus dynamics drive more rapid membrane depolarizations, whose steeper threshold crossings produce more precisely timed spikes, which in turn carry stimulus information at finer temporal precision. Quotation withheld because this record is not cleared for redistribution. effective precision is degraded by shuffling spikes between neighbouring bins, which lowers resolution without changing the statistical dimensionality of the data Quotation withheld because this record is not cleared for redistribution. direct mutual information is used instead of decoding precisely because a decoder can miss information a neuron actually carries Quotation withheld because this record is not cleared for redistribution. and the two are then run against each other on the same data, showing the method itself shifts the apparent coding timescale Quotation withheld because this record is not cleared for redistribution. the stimulus is chosen to make the test hardest - random chords have a correlation time of a few milliseconds and are the case where precise timing was argued not to survive Quotation withheld because this record is not cleared for redistribution. a second stimulus class with a much slower correlation structure is run so that stimulus dynamics becomes a variable rather than a constant Quotation withheld because this record is not cleared for redistribution. the peak-width measure is explicitly declared insufficient before the information analysis is introduced, because a spike could carry information by where it falls within a peak Quotation withheld because this record is not cleared for redistribution. the information estimate is built to assume nothing about which acoustic features drive the cell Quotation withheld because this record is not cleared for redistribution. significance of information loss is established per neuron against an artificial response carrying no information at that precision Quotation withheld because this record is not cleared for redistribution. the rate-versus-correlation question is settled by computing the information available to an observer who sees only the time-dependent rate Quotation withheld because this record is not cleared for redistribution. the precision bound is bracketed from both sides rather than reported as a single figure Quotation withheld because this record is not cleared for redistribution. the discrepancy with earlier work is attributed to two separable method differences, and the experiment that would separate them is specified Quotation withheld because this record is not cleared for redistribution. Because coding precision tracks stimulus dynamics, the reported precision is a lower bound and could be finer with faster stimuli. what this limits: the precision figure, with its direction: OUR RESULTS MAY RATHER PROVIDE A LOWER BOUND ON THE EFFECTIVE PRECISION, LEAVING THE POSSIBILITY THAT AUDITORY CORTEX NEURONS MIGHT ENCODE SENSORY INFORMATION AT THE (SUB) MILLISECOND SCALE Quotation withheld because this record is not cleared for redistribution. The higher precision found here relative to earlier work may be an artefact of the continuous-sequence stimulation paradigm rather than a general property. what this limits: the comparison with earlier reports, which is the paper's novelty: Quotation withheld Quotation withheld because this record is not cleared for redistribution. The context dependence of coding time scales was not directly tested and is left to future work. what this limits: the generality of the time scales: FUTURE STUDIES SHOULD DIRECTLY INVESTIGATE THE CONTEXT DEPENDENCY OF NEURAL CODING TIME SCALES Quotation withheld because this record is not cleared for redistribution. the precision reported is a lower bound, since faster stimuli might reveal finer coding duplicate of: limitations[0] · what this limits: the same scope as limitations[0] - faster stimuli could reveal finer coding Quotation withheld because this record is not cleared for redistribution. coding precision was much more variable across neurons for natural sounds, and the population loss was smaller than for random chords what this limits: the natural-sound results specifically: THE PROPORTIONAL INFORMATION LOST BY REDUCING THE EFFECTIVE TEMPORAL PRECISION WAS SMALLER THAN THE RESULTS OBTAINED DURING STIMULATION WITH RANDOM CHORDS Quotation withheld because this record is not cleared for redistribution. the difference from previous reports may itself be an artefact of presenting sounds continuously rather than in isolation duplicate of: limitations[1] · what this limits: the same scope as limitations[1] - continuous versus isolated presentation may explain the difference from previous work Quotation withheld because this record is not cleared for redistribution. the biophysical account of why faster stimuli give more precise spikes is offered as a simple hypothesis what this limits: the mechanistic account: A SIMPLE HYPOTHESIS IS THAT PRECISION OF SPIKE TIMES IS DETERMINED BY THE SLOPE OF THE MEMBRANE VOLTAGE WHEN CROSSING THE FIRING THRESHOLD - offered as a hypothesis, not shown Quotation withheld because this record is not cleared for redistribution. the link to behaviour is explicitly a temptation rather than a result what this limits: the link to behaviour: IT IS TEMPTING TO HYPOTHESIZE THAT SUCH MILLISECOND-PRECISE SINGLE NEURON-ENCODING SERVES AS NEURAL SUBSTRATE FOR RECENTLY REPORTED BEHAVIORAL CORRELATES Quotation withheld because this record is not cleared for redistribution. Shannon's mutual information, defined over a set of stimuli and a set of responses, is imported as the direct measure of what a spike train tells an observer about the sound Quotation withheld because this record is not cleared for redistribution. Shannon mutual information estimated directly, chosen because it bounds all the information a response could carry rather than what a particular decoder can extract Quotation withheld because this record is not cleared for redistribution. the finding that response precision scales with stimulus speed, imported as the prediction that the auditory answer should depend on stimulus dynamics Quotation withheld because this record is not cleared for redistribution. each stimulus repeated many times - 55 repeats on average, range 39 to 70 eliminates: a spike-timing precision estimate built on too few trials to separate reliable timing from chance alignment · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. temporal coding - as spike-time correlations carrying information beyond the rate versus as fast modulation of the rate itself; the paper finds the second and so the dichotomy dissolves Quotation withheld because this record is not cleared for redistribution. precision of the neural code - defined operationally as the coarsest resolution that still yields all the information a response carries Quotation withheld because this record is not cleared for redistribution. response precision - as PSTH peak width versus as information loss under degraded resolution; the first is declared insufficient and the second is what the claims rest on Quotation withheld because this record is not cleared for redistribution. coding timescale - as a property of the neuron versus as a property of the neuron given a stimulus; the same cells are more precise for faster sounds Quotation withheld because this record is not cleared for redistribution. millisecond precision - established at the auditory periphery and subcortex, but contested at cortex; the paper places cortex between the two Quotation withheld because this record is not cleared for redistribution. a sound presented in silence versus embedded in a continuous stream of similar sounds - two paradigms that may give different answers about the same neurons Quotation withheld because this record is not cleared for redistribution. claim_type: existence a second reading reaching the same value. Recorded: existence — note: a second reading reaching the same value from a different sentence. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. method: single-unit recording in passively listening macaques during long continuous random-chord and natural-sound sequences, with direct Shannon information estimates and a spike-shuffling procedure that degrades effective precision a second reading DIFFERING from the recorded value. Recorded: electrophysiology — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. phenomenon: temporal precision of the neural code in auditory cortex a second reading DIFFERING from the recorded value. Recorded: millisecond spike-timing precision in auditory cortex — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. contradicts → previous studies reporting 10- to 20-ms coarse coding precision in auditory cortex (refs. 9 and 11) Quotation withheld because this record is not cleared for redistribution. builds-on → Desbordes et al. (ref. 19), PSTH autocorrelogram measure of response precision Quotation withheld because this record is not cleared for redistribution. extends → ref. 18, response precision scaling with stimulus speed in visual thalamus of anesthetized cats Quotation withheld because this record is not cleared for redistribution. contradicts → the conclusion that millisecond-precise spike timing adds nothing beyond spike rate on the 10-20 ms scale in auditory cortex Quotation withheld because this record is not cleared for redistribution. critiques → decoding-based analyses, which understate the information carried at fine precision differs in: THE AXIS IS THE TEMPORAL SENSITIVITY OF THE DECODER. 'Quotation withheld' A published estimate of how much information a neuron carries is bounded by the resolution of the tool used to read it - so a lower number is evidence about the decoder as much as about the neuron. Quotation withheld because this record is not cleared for redistribution. critiques → paradigms presenting brief isolated sounds separated by silence, which may themselves produce the coarser precision reported differs in: THE AXIS IS THE STIMULATION PARADIGM, AND THE PAPER TURNS IT ON ITSELF. 'Quotation withheld' The disagreement with the literature is attributed to the paper's own method BEFORE ANYONE ELSE CAN ATTRIBUTE IT - and its [4] makes the complementary point about OTHER people's decoders, so one paper supplies both halves of why the numbers differ. Quotation withheld because this record is not cleared for redistribution. extends → established millisecond precision at peripheral and subcortical auditory stages, upward to cortex Quotation withheld because this record is not cleared for redistribution. supports → findings in rat somatosensory cortex and retina that sensory codes require millisecond resolution Quotation withheld because this record is not cleared for redistribution. extends → the visual thalamus result that response precision is higher for faster stimuli, to auditory cortex in the alert primate Quotation withheld because this record is not cleared for redistribution. time-dependent firing rate —contrasted-with→ higher-order correlations time-dependent firing rate higher-order correlations strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. temporal precision —operationalised-as→ mutual information temporal precision mutual information strength of the relation: 0.95 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'Shannon information [I(S;R)]', not 'mutual information'. Quotation withheld because this record is not cleared for redistribution. stimulus dynamics —causes→ temporal precision stimulus dynamics temporal precision strength of the relation: 0.8 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the results 'SUGGEST' precision 'MAY depend' on stimulus dynamics; causes is hedged. Quotation withheld because this record is not cleared for redistribution. decoding —contrasted-with→ mutual information decoding mutual information strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. random chords —contrasted-with→ natural sounds random chords natural sounds strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. temporal precision —operationalised-as→ PSTH autocorrelogram temporal precision PSTH autocorrelogram strength of the relation: 0.9 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'response precision', not 'temporal precision'. Quotation withheld because this record is not cleared for redistribution. effective precision —causes→ information per spike effective precision information per spike strength of the relation: 0.95 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'response precision', not 'effective precision'. Quotation withheld because this record is not cleared for redistribution. membrane voltage —causes→ spike timing membrane voltage spike timing strength of the relation: 0.7 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but 'a simple hypothesis' is that precision is determined by the voltage slope; causes is hedged. Quotation withheld because this record is not cleared for redistribution. mutual information mutual information Quotation withheld because this record is not cleared for redistribution. bootstrap bootstrap Quotation withheld because this record is not cleared for redistribution. median median Quotation withheld because this record is not cleared for redistribution. p-value p-value Quotation withheld because this record is not cleared for redistribution. autocorrelation autocorrelation Quotation withheld because this record is not cleared for redistribution. standard error of the mean standard error of the mean Quotation withheld because this record is not cleared for redistribution. bias correction bias correction Quotation withheld because this record is not cleared for redistribution. Shannon information Shannon information Quotation withheld because this record is not cleared for redistribution. bootstrap procedure bootstrap procedure Quotation withheld because this record is not cleared for redistribution. limited sampling bias limited sampling bias Quotation withheld because this record is not cleared for redistribution. autocorrelation autocorrelation Quotation withheld because this record is not cleared for redistribution. Registering auditory cortex spike times at a resolution coarser than a few milliseconds significantly reduces the stimulus information that can be recovered. Quotation withheld because this record is not cleared for redistribution. The effective coding resolution in this dataset lies between 1 ms and roughly 4-6 ms, since degrading precision from 1 to 2 ms cost no information but 4-6 ms did. Quotation withheld because this record is not cleared for redistribution. Nearly all of the stimulus information resides in the finely timed modulation of firing rate, with higher-order correlations between spike times contributing negligibly. Quotation withheld because this record is not cleared for redistribution. Response precision and the cost of coarsening it were greater for the fast-varying random chords than for natural sounds, so coding precision tracks the stimulus's own time scale. Quotation withheld because this record is not cleared for redistribution. Linear decoding showed smaller information losses when precision was degraded, so it would have indicated a coarser encoding timescale than direct estimates. Quotation withheld because this record is not cleared for redistribution. median PSTH autocorrelogram half-width was 5 ms for random chords, indicating precision finer than 10 ms Quotation withheld because this record is not cleared for redistribution. information loss was 5%, 11% and 20% at effective precisions of 4, 6 and 12 ms Quotation withheld because this record is not cleared for redistribution. loss was significant for 7% of neurons at 4 ms and 46% at 12 ms Quotation withheld because this record is not cleared for redistribution. no neuron lost information going from 1 ms to 2 ms bins, so the precision is coarser than 1 ms but finer than about 4 to 6 ms Quotation withheld because this record is not cleared for redistribution. information per spike fell from 4.3 bits at 1 ms to 3.6 and 2.6 bits at 6 and 12 ms Quotation withheld because this record is not cleared for redistribution. the time-dependent firing rate accounted for 99% of the total information, so higher-order spike correlations add essentially nothing Quotation withheld because this record is not cleared for redistribution. for natural sounds the proportional information loss was smaller than for random chords Quotation withheld because this record is not cleared for redistribution. EVIDENCE CORRECTED (RE-GROUNDED - THE POINTER WAS WRONG, NOT THE CLAIM, AND THE AUTOMATIC CANDIDATE WAS WRONG TOO) · but a subset still showed significant loss, 11% at 6 ms and 17% at 12 ms EVIDENCE CORRECTED: RE-GROUNDED - THE POINTER WAS WRONG, NOT THE CLAIM, AND THE AUTOMATIC CANDIDATE WAS WRONG TOO. The old evidence was the sentence immediately BEFORE the one carrying the numbers. the project's candidate generator proposed 'the information loss amounted…' - A DIFFERENT QUANTITY: the MEDIAN INFORMATION LOSS, not the FRACTION OF NEURONS with a significant loss. Taking it would have made the claim say 20% where the paper says 17%. The right sentence is the next one and gives 11 and 17% exactly as the claim does. A GOOD CANDIDATE GENERATOR IS A BAD EDITOR, in the project's own words. Quotation withheld because this record is not cleared for redistribution. so millisecond timing carries information about natural sounds that a 10 ms spike count cannot recover Quotation withheld because this record is not cleared for redistribution. linear decoding showed smaller losses than direct information, and would have implied a coarser coding timescale from the same data Quotation withheld because this record is not cleared for redistribution. precision was finer and information loss larger for the faster-varying stimulus Quotation withheld because this record is not cleared for redistribution. cortical precision is high but coarser than at peripheral and subcortical auditory stages Quotation withheld because this record is not cleared for redistribution. Shannon (mutual) information Shannon (mutual) information Quotation withheld because this record is not cleared for redistribution. precision of the neural code precision of the neural code Quotation withheld because this record is not cleared for redistribution. random chords random chords Quotation withheld because this record is not cleared for redistribution. precision of the neural code precision of the neural code Quotation withheld because this record is not cleared for redistribution. Shannon information Shannon information Quotation withheld because this record is not cleared for redistribution. shuffling procedure shuffling procedure Quotation withheld because this record is not cleared for redistribution. random chords random chords Quotation withheld because this record is not cleared for redistribution. effective temporal precision effective temporal precision Quotation withheld because this record is not cleared for redistribution. information loss information loss Quotation withheld because this record is not cleared for redistribution. response precision response precision Quotation withheld because this record is not cleared for redistribution. statistical dimensionality statistical dimensionality Quotation withheld because this record is not cleared for redistribution. effective precision effective precision Quotation withheld because this record is not cleared for redistribution. stimulus information stimulus information Quotation withheld because this record is not cleared for redistribution. correlation time scale correlation time scale Quotation withheld because this record is not cleared for redistribution. information per spike information per spike Quotation withheld because this record is not cleared for redistribution. 5 ms — population median half-width at half-height of the PSTH autocorrelogram (response precision) during random chord stimulation Quotation withheld because this record is not cleared for redistribution. 5, 11, and 20% — median information loss at effective precisions of 4, 6, and 12 ms relative to 1 ms precision Quotation withheld because this record is not cleared for redistribution. 4.3 bits/spike — median information per spike at 1 ms precision for neurons with significant loss at 12 ms, dropping to 3.6 and 2.6 bits/spike at 6 and 12 ms Quotation withheld because this record is not cleared for redistribution. 99 ± 0.2% — share of total stimulus information captured by the time-dependent firing rate alone at precisions below 48 ms Quotation withheld because this record is not cleared for redistribution. 46% — proportion of random-chord neurons with statistically significant information loss at 12 ms effective precision (7% already at 4 ms) Quotation withheld because this record is not cleared for redistribution. 16 ms — median PSTH-derived response precision during natural sound stimulation Quotation withheld because this record is not cleared for redistribution. 5 ms — population median PSTH autocorrelogram half-width for random chords Quotation withheld because this record is not cleared for redistribution. 5, 11, and 20% — median information loss at 4, 6 and 12 ms effective precision Quotation withheld because this record is not cleared for redistribution. 4.3 bits/spike — information per spike at 1 ms precision, falling to 2.6 at 12 ms Quotation withheld because this record is not cleared for redistribution. 99 ± 0.2% — share of total information available from the time-dependent rate alone Quotation withheld because this record is not cleared for redistribution. n = 41 and n = 31 — neurons recorded for random chords and for natural sounds value source note: THE VALUE DRAWS ON THE PAPER BEYOND THIS QUOTE: part of it is printed elsewhere in the paper, not in this quote ('responses of 41 neurons to a sequence of pseudorandom tones…'). Quotation withheld because this record is not cleared for redistribution. average 55 repeats — stimulus repetitions per recording, range 39 to 70 Quotation withheld because this record is not cleared for redistribution. 40-s and 52-s — durations of the random chord and natural sound sequences Quotation withheld because this record is not cleared for redistribution. three adult rhesus monkeys — the sample Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1073/pnas.1012656107.