Inferring the role of inhibition in auditory processing of complex natural stimuli

Inferring the role of inhibition in auditory processing of complex natural stimuli Schinkel-Bielefeld David Shamma Butts

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.

Inferring the role of inhibition in auditory processing of complex natural stimuli 2012 Journal of Neurophysiology 10.1152/jn.01173.2011 Schinkel-Bielefeld2012-inferring-the-role-of-inhibition-in-auditory Schinkel-Bielefeld2012-inferring-the-role-of-inhibition-in-auditory.txt, the project's staged copy 2518e578179466d8b047c19cce5a1c0c60cbc4fc 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 how the interplay of excitation and inhibition shapes A1 responses to natural sounds, and whether it can be recovered from extracellular spikes alone Quotation withheld because this record is not cleared for redistribution. whether a model that separates excitatory from inhibitory contributions predicts A1 spike trains better than the linear receptive field, and recovers properties matching intracellular measurements Quotation withheld because this record is not cleared for redistribution. a nonlinear model whose inferred excitation and inhibition predict spikes better than the linear model and reproduce intensity tuning, onset and offset responses that were never explicitly built in Quotation withheld because this record is not cleared for redistribution. mechanism Quotation withheld because this record is not cleared for redistribution. excitatory-inhibitory interplay in primary 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 Intensity tuning arises because the excitatory/inhibitory balance shifts with sound level: at high intensities the filtered inputs land on the saturating part of the internal nonlinearities, so the excitatory lead no longer produces a large net drive. Quotation withheld because this record is not cleared for redistribution. mechanisms Onset responses are generated by excitation reaching the cell before inhibition, with the lagging inhibition cutting the response short. Quotation withheld because this record is not cleared for redistribution. mechanisms Offset responses are generated when the excitatory nonlinearity is amplified relative to the inhibitory one, so inhibition falls away before excitation does and firing rebounds. Quotation withheld because this record is not cleared for redistribution. mechanisms Sustained firing is generated when, at large filter outputs, the internal nonlinearity amplifies excitation more than inhibition. Quotation withheld because this record is not cleared for redistribution. transient onset responses arise from inhibition arriving later than excitation, and also from a difference in their nonlinear gains even when the latencies match Quotation withheld because this record is not cleared for redistribution. offset responses are described as rebound from inhibition, in the 15% of neurons that show them Quotation withheld because this record is not cleared for redistribution. and intensity tuning falls out of the separate nonlinearities, which shift the balance between excitation and inhibition as intensity and stimulus dynamics change Quotation withheld because this record is not cleared for redistribution. with the whole approach resting on approximating the mechanisms rather than the response properties Quotation withheld because this record is not cleared for redistribution. The model's excitatory and inhibitory terms cannot be claimed to correspond to actual synaptic currents or conductances. what this limits: every interpretation of the model's terms as physiology: THE LABELS FOR EACH TERM ARE MEANT TO DESCRIBE THE INTUITION BEHIND THE STRUCTURE OF THE MODEL, BUT WE CANNOT CLAIM THAT THESE TERMS NECESSARILY REPRESENT ACTUAL excitatory and inhibitory currents Quotation withheld because this record is not cleared for redistribution. The model's constrained structure sacrifices generality and may perform poorly for nonlinearities it does not embody. what this limits: the model's performance outside the nonlinearities it encodes: THIS LACK OF GENERALITY COULD HURT THE GNM PERFORMANCE WHEN CONSIDERING NONLINEAR PROCESSING OF FORMS NOT CAPTURED BY ITS STRUCTURE Quotation withheld because this record is not cleared for redistribution. The current model cannot represent separate offset-tuned inputs as an alternative source of offset responses. what this limits: the account of offset responses: the model cannot represent separate offset-tuned inputs as the alternative source, so that explanation is untested rather than excluded Quotation withheld because this record is not cleared for redistribution. The fit does not prove that the proposed nonlinear mechanisms actually produce the observed A1 nonlinearities. what this limits: the mechanistic reading of the model fit: ALTHOUGH THIS IS BY NO MEANS PROOF THAT SUCH NONLINEAR MECHANISMS ARE RESPONSIBLE FOR OBSERVED A1 NONLINEARITIES, IT DOES PROVIDE A UNIFIED FRAMEWORK Quotation withheld because this record is not cleared for redistribution. the fit is explicitly not proof that these mechanisms are responsible for the observed nonlinearities what this limits: the same scope as limitations[3], from the other direction: THERE IS NO GUARANTEE THAT FOCUSING ON INTERNAL NONLINEAR MECHANISMS WOULD NECESSARILY DESCRIBE THE OBSERVED RESPONSE NONLINEARITIES, WE FOUND THAT THEY DID Quotation withheld because this record is not cleared for redistribution. the model cannot distinguish rebound from a separate set of offset-tuned inputs, which more recent work reports what this limits: the rebound account of offset responses: THE GNM IN ITS CURRENT FORM DOES NOT ADDRESS THE POSSIBILITY OF SEPARATE OFFSET-TUNED INPUTS Quotation withheld because this record is not cleared for redistribution. some intensity tuning may be inherited from the inputs to A1 rather than generated there what this limits: the claim that the tuning is generated in A1: IT IS POSSIBLE THAT SOME ASPECTS OF INTENSITY TUNING ARE INHERITED FROM THE INPUTS TO A1 Quotation withheld because this record is not cleared for redistribution. and the structure that makes it efficient also costs it the generality of methods that search for stimulus dimensions without assuming a form what this limits: the generality of the model class: it FOCUSES ON WHAT IS EXPECTED TO BE A DOMINANT mechanism rather than searching the space of stimulus dimensions, which is what makes it efficient and what bounds it Quotation withheld because this record is not cleared for redistribution. with the intensity tuning measured here agreeing only qualitatively with the same property measured using pure tones what this limits: any comparison of these intensity-tuning values with pure-tone measurements: Quotation withheld, DUE TO THE EFFECT OF STIMULUS CONTEXT ON A1 RESPONSES Quotation withheld because this record is not cleared for redistribution. cotuned excitation and inhibition, and the timing and gain differences between them, imported as the model's structure Quotation withheld because this record is not cleared for redistribution. a sentence database, used as the natural stimulus set Quotation withheld because this record is not cleared for redistribution. the linear receptive field itself, and the maximum likelihood machinery used to fit it Quotation withheld because this record is not cleared for redistribution. spike-triggered covariance and maximally informative dimension analysis, which identify additional stimulus features that contribute to the computation underlying the neuronal response Quotation withheld because this record is not cleared for redistribution. the equaliser calibrated before each experiment to achieve flat gain for the acoustic properties of that earphone insertion eliminates: a spectral tilt introduced by how the earphone happened to sit - re-calibrating per experiment means the stimulus spectrum is the one intended rather than the one the fitting produced · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. THE STANDARD MODEL STRUCTURALLY CANNOT SEE THE PHENOMENON: the linear receptive field assumes responses are an average over the neuron's many inputs, and SUCH AVERAGING NECESSARILY HIDES the effects of excitation and inhibition where their tuning overlaps Quotation withheld because this record is not cleared for redistribution. so a property that intracellular work established is invisible to the measure the field uses extracellularly - not undetected, but hidden by construction Quotation withheld because this record is not cleared for redistribution. AND THE FIT IS DECLARED NOT TO BE PROOF: there is no guarantee that focusing on internal mechanisms would describe the observed nonlinearities, and this is BY NO MEANS PROOF that they are responsible Quotation withheld because this record is not cleared for redistribution. AN OUT-OF-DESIGN PREDICTION USED AS EVIDENCE: the model has NO EXPLICIT MECHANISM for intensity tuning and predicts it better than the linear model and as well as a model built for it Quotation withheld because this record is not cleared for redistribution. THREE modeling APPROACHES COMPARED, WITH THE TRADE-OFF NAMED: searching for extra stimulus features is general but leaves it unclear what mechanism is being captured, and this model's structure costs generality while buying data efficiency Quotation withheld because this record is not cleared for redistribution. with the weakness of the first approach stated in its own terms - no constraints are introduced on the type of computation, so the higher-order description is hard to tie to known processing Quotation withheld because this record is not cleared for redistribution. THE SAME PROPERTY MEASURED WITH TWO STIMULI GIVES DIFFERENT NUMBERS, AND THE PAPER SAYS WHY IT SHOULD: intensity tuning agrees qualitatively but NOT quantitatively with the pure-tone measurement, because context and stimulus dynamics change the excitatory-inhibitory balance Quotation withheld because this record is not cleared for redistribution. and a new measure is introduced specifically so that models can be compared on it, rather than only on spike prediction Quotation withheld because this record is not cleared for redistribution. A RIVAL MECHANISM FOR THE SAME DATA, NAMED AND LEFT OPEN: offset responses are described here as rebound from inhibition, while more recent work finds a distinct set of offset-tuned inputs Quotation withheld because this record is not cleared for redistribution. and the population is heterogeneous - only 60% are clear onset cells and 15% have a clear offset response, so one mechanism cannot stand for the class Quotation withheld because this record is not cleared for redistribution. THE STIMULUS CHOICE MAKES CERTAIN MODEL PROPERTIES RELEVANT: speech is made of discrete syllables separated by gaps, which is what makes onset, offset and sustained responses matter to overall performance Quotation withheld because this record is not cleared for redistribution. and the experiment that would settle it is specified - tones ramped in intensity at different speeds, compared against the model or measured intracellularly Quotation withheld because this record is not cleared for redistribution. with an alternative source acknowledged: some intensity tuning may be INHERITED from the inputs to A1 rather than computed there Quotation withheld because this record is not cleared for redistribution. AND THE VALIDATION IS AGREEMENT WITH A DIFFERENT METHOD: properties inferred from extracellular spikes are consistent with previous intracellular observations Quotation withheld because this record is not cleared for redistribution. the class membership is assigned by a property bundle - tonotopic organization, latency and simple frequency tuning verify a neuron as belonging to the area Quotation withheld because this record is not cleared for redistribution. claim_type: method a second reading DIFFERING from the recorded value. Recorded: mechanism — 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. method: a generalised nonlinear model with separate filters and nonlinearities for putative excitatory and inhibitory elements, fitted to spikes evoked by continuous speech 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: cotuned excitation and inhibition, whose effects cancel in any measure that averages over a neuron's inputs a second reading DIFFERING from the recorded value. Recorded: excitatory-inhibitory interplay in primary 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. builds-on → the generalized nonlinear model (GNM) framework of Butts et al. (2011) Quotation withheld because this record is not cleared for redistribution. builds-on → the A1 recording dataset and procedures of David et al. (2009) Quotation withheld because this record is not cleared for redistribution. extends → the multilinear / input-nonlinearity model of A1 intensity tuning (Ahrens et al. 2008b) Quotation withheld because this record is not cleared for redistribution. critiques → the linear spectrotemporal receptive field, which averages over inputs and therefore cannot represent cotuned excitation and inhibition differs in: THE AXIS IS WHETHER INHIBITION IS COTUNED WITH EXCITATION. 'Quotation withheld.' A model that averages over inputs cannot represent two opposing drives with the same tuning - so the objection is about what the FORMALISM can hold, not about whether its fits are good. Quotation withheld because this record is not cleared for redistribution. extends → intracellular findings of cotuned excitation and inhibition, to extracellular recordings and natural stimuli Quotation withheld because this record is not cleared for redistribution. distinguishes-from → feature-search methods such as spike-triggered covariance and maximally informative dimensions, which are more general and less interpretable differs in: THE AXIS IS WHETHER THE NONLINEARITY IS DISCOVERED FROM THE DATA. 'One approach has been to identify additional stimulus features (in addition to the linear STRF) that contribute to the computation underlying the neuronal response, for example using SPIKE-TRIGGERED COVARIANCE (STC) or MAXIMALLY INFORMATIVE DIMENSION (MID) ANALYSIS.' Its twin at [6] carries the other family, which builds in nonlinearities chosen in advance. NOTE: the target also calls these methods 'more general and less interpretable' AND THIS SPAN SAYS NEITHER - the axis recorded here is the one the sentence supports, and the target's extra claim remains unanchored. Quotation withheld because this record is not cleared for redistribution. distinguishes-from → multilinear models that add components for specific known nonlinearities differs in: THE AXIS IS WHETHER THE NONLINEARITY IS DISCOVERED OR SPECIFIED IN ADVANCE. 'A second approach has taken an opposite tack, by incorporating model components to capture specific nonlinear elements of A1 processing, such as intensity tuning and dependence on stimulus context. This approach has been CAST AS A MULTILINEAR MODEL.' One family searches for features the data reveal; the other builds in nonlinearities the modeller already believes in - so the two can fit the same neuron and disagree about what it does. Quotation withheld because this record is not cleared for redistribution. supports → the account in which relative timing of excitation and inhibition produces transient onset responses Quotation withheld because this record is not cleared for redistribution. qualifies → its own rebound account of offset responses, against a report of separate offset-tuned inputs Quotation withheld because this record is not cleared for redistribution. excitation —contrasted-with→ inhibition excitation inhibition strength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span concerns co-tuned excitation and inhibition; contrasted-with is loose. Quotation withheld because this record is not cleared for redistribution. STRF —operationalised-as→ stimulus STRF stimulus strength of the relation: 1 · note: sounds matching it drive the strongest responses, which is the intuition the whole field runs on Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the STRF describes stimulus selectivity; operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. GNM —causes→ intensity tuning GNM intensity tuning strength of the relation: 1 · note: AN OUT-OF-DESIGN PREDICTION USED AS EVIDENCE: NO EXPLICIT MECHANISM for it, and it predicts it better than the linear model and as well as a model built for it Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the model REPRODUCED A1 intensity tuning through its nonlinearities; causes is loose. Quotation withheld because this record is not cleared for redistribution. input —causes→ intensity tuning input intensity tuning strength of the relation: 1 · note: with the alternative source acknowledged - some of it may be INHERITED from the inputs rather than computed here Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span says intensity tuning may be inherited from inputs; causes states a possibility. Quotation withheld because this record is not cleared for redistribution. inhibition —causes→ onset inhibition onset strength of the relation: 1 · note: inhibition delayed relative to excitation, as intracellular work suggested - AND a gain difference can do it with matched latencies Audit 2026-09-28: DOUBTFUL - read against its span and kept, but onset responses arise from the interplay of excitation and inhibition; causes is loose. Quotation withheld because this record is not cleared for redistribution. maximum likelihood estimation maximum likelihood estimation Quotation withheld because this record is not cleared for redistribution. cross-validation cross-validation Quotation withheld because this record is not cleared for redistribution. t-test t-test Quotation withheld because this record is not cleared for redistribution. paired t-test paired t-test Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. principal components analysis principal components analysis Quotation withheld because this record is not cleared for redistribution. median median Quotation withheld because this record is not cleared for redistribution. entropy entropy Quotation withheld because this record is not cleared for redistribution. explained variance (r-squared) explained variance (r-squared) Quotation withheld because this record is not cleared for redistribution. principal components analysis principal components analysis Quotation withheld because this record is not cleared for redistribution. k -means clustering k -means clustering Quotation withheld because this record is not cleared for redistribution. The nonlinear GNM predicts A1 spike trains during speech significantly better than the linear STRF-based model across the recorded population. Quotation withheld because this record is not cleared for redistribution. The excitation and inhibition inferred from extracellular data are cotuned: they peak at the same frequency in 44% of neurons and within half an octave in a further 35%. Quotation withheld because this record is not cleared for redistribution. Inhibition is more broadly tuned in frequency than excitation in the large majority of cells. Quotation withheld because this record is not cleared for redistribution. Nonmonotonic intensity tuning emerges from the model even though no intensity-tuning term was built into it, matching a model that fits intensity tuning explicitly. Quotation withheld because this record is not cleared for redistribution. The GNM also reproduces neuronal responses to sound onsets and offsets better than the linear model. Quotation withheld because this record is not cleared for redistribution. Among neurons with sufficient data, 60% were onset cells with little sustained firing and 24% responded with strong sustained activity. Quotation withheld because this record is not cleared for redistribution. Because it is a linear measure, the spectrotemporal receptive field cannot detect overlapping excitation and inhibition. Quotation withheld because this record is not cleared for redistribution. THE STANDARD MODEL STRUCTURALLY CANNOT SEE THE PHENOMENON: averaging over a neuron's inputs NECESSARILY HIDES excitation and inhibition where their tuning overlaps WHOSE RESULT THIS IS: studies show - THE SUBJECT OF THIS RESULT IS AN INDEFINITE STUDY, NOT THIS PAPER. 'Previous studies reported', 'studies in guinea pig showed', 'recent studies have reported' - somebody else's work under a common noun rather than a name, with a citation in the sentence and no first-person marker. The quote is verbatim and the claim is true; only the attribution is wrong. FOUND BY MEASURING MY OWN CUE'S RECALL: the first sweep required a NAMED author as the subject ('Sperling (1960) demonstrated') and found 17. Reading eleven citation-bearing findings on 'Plakke2016' by hand showed four the cue could not see - 'A study examining recognition found', 'studies of typical human subjects have demonstrated', 'A follow up study found' - so the named cue's 17 was a floor and not a count. A CUE'S FALSE POSITIVES TELL YOU NOTHING ABOUT ITS FALSE NEGATIVES, and the only way to find these was to read a record's whole layer. Quotation withheld because this record is not cleared for redistribution. the linear receptive field is the standard first-order description, and sounds matching it drive the strongest responses Quotation withheld because this record is not cleared for redistribution. and the gap is named precisely - existing nonlinear models address some aspects and NONE has addressed the interplay of excitation and inhibition Quotation withheld because this record is not cleared for redistribution. THE VALIDATION IS AGREEMENT WITH A DIFFERENT METHOD: properties inferred from extracellular spikes are consistent with previous intracellular observations Quotation withheld because this record is not cleared for redistribution. only clusters with 90% isolation were used, and neurons were verified as belonging to the area by tonotopic organization, latency and simple frequency tuning Quotation withheld because this record is not cleared for redistribution. the stimulus was continuous speech from a sentence database, balanced across speaker sex Quotation withheld because this record is not cleared for redistribution. with a consequence of the upsampling disclosed - all stimulus power was below 8 kHz Quotation withheld because this record is not cleared for redistribution. and two sets of neurons heard different stimulus sets, which is disclosed rather than smoothed over Quotation withheld because this record is not cleared for redistribution. A NEW MEASURE IS INTRODUCED so that models can be compared on intensity tuning itself, not only on spike prediction Quotation withheld because this record is not cleared for redistribution. THE SAME PROPERTY MEASURED WITH TWO STIMULI GIVES DIFFERENT NUMBERS, AND THE PAPER SAYS WHY IT SHOULD Quotation withheld because this record is not cleared for redistribution. AN OUT-OF-DESIGN PREDICTION USED AS EVIDENCE: the model has no explicit mechanism for intensity tuning and predicts it better than the linear model Quotation withheld because this record is not cleared for redistribution. and the settling experiment is specified - tones ramped in intensity at different speeds, compared against the model or measured intracellularly Quotation withheld because this record is not cleared for redistribution. with an alternative source acknowledged - some intensity tuning may be inherited from the inputs to the area Quotation withheld because this record is not cleared for redistribution. THE STIMULUS CHOICE MAKES CERTAIN MODEL PROPERTIES RELEVANT: speech is made of syllables separated by gaps, so onset, offset and sustained responses matter to performance Quotation withheld because this record is not cleared for redistribution. transient onsets arise from delayed inhibition, AND from a gain difference even when the latencies are the same Quotation withheld because this record is not cleared for redistribution. only 60% of neurons are clear onset cells and 15% have a clear offset response WHOSE RESULT THIS IS: Qin et al. 2007 ), although more recent work found - THE EVIDENCE FOR THIS ROW ATTRIBUTES THE WORK TO SOMEBODY ELSE. The cited name is the subject of a result verb and the sentence carries no first-person marker, so the finding, quantity or definition is the CITED author's rather than this record's. The claim is true and the quote is verbatim - what is wrong is whose result the corpus says it is. Nothing edited, nothing withdrawn: the row is marked so a page can name the source. Where the record is a commentary or review, its claim relations layer usually already names the paper being discussed - the record knows at one layer and forgets at another. Quotation withheld because this record is not cleared for redistribution. A RIVAL MECHANISM NAMED AND LEFT OPEN: the model cannot address separate offset-tuned inputs, and could be extended to find them Quotation withheld because this record is not cleared for redistribution. with the motivation stated as a recognition - linear models MISS SIGNIFICANT ASPECTS of the processing Quotation withheld because this record is not cleared for redistribution. and the weakness of the feature-search approach stated in its own terms - no constraints on the type of computation, so the mechanism captured is unclear Quotation withheld because this record is not cleared for redistribution. the second approach adds components for specific known nonlinearities and can be estimated efficiently Quotation withheld because this record is not cleared for redistribution. THE TRADE-OFF NAMED IN BOTH DIRECTIONS: the structure could hurt performance for nonlinearities it does not have, and will be much more efficient with data when they match Quotation withheld because this record is not cleared for redistribution. AND THE FIT IS DECLARED NOT TO BE PROOF: no guarantee it would work, and BY NO MEANS PROOF that these mechanisms are responsible Quotation withheld because this record is not cleared for redistribution. the properties that were NOT explicitly modeled are exactly the ones offered as evidence Quotation withheld because this record is not cleared for redistribution. generalized nonlinear model (GNM) generalized nonlinear model (GNM) Quotation withheld because this record is not cleared for redistribution. onset-triggered average onset-triggered average Quotation withheld because this record is not cleared for redistribution. silence rate and sound rate silence rate and sound rate Quotation withheld because this record is not cleared for redistribution. effective time kernel effective time kernel Quotation withheld because this record is not cleared for redistribution. level-rate correlation coefficient level-rate correlation coefficient Quotation withheld because this record is not cleared for redistribution. STRF STRF Quotation withheld because this record is not cleared for redistribution. cotuned cotuned WHOSE RESULT THIS IS: studies show - THE SUBJECT OF THIS RESULT IS AN INDEFINITE STUDY, NOT THIS PAPER. 'Previous studies reported', 'studies in guinea pig showed', 'recent studies have reported' - somebody else's work under a common noun rather than a name, with a citation in the sentence and no first-person marker. The quote is verbatim and the claim is true; only the attribution is wrong. FOUND BY MEASURING MY OWN CUE'S RECALL: the first sweep required a NAMED author as the subject ('Sperling (1960) demonstrated') and found 17. Reading eleven citation-bearing findings on 'Plakke2016' by hand showed four the cue could not see - 'A study examining recognition found', 'studies of typical human subjects have demonstrated', 'A follow up study found' - so the named cue's 17 was a floor and not a count. A CUE'S FALSE POSITIVES TELL YOU NOTHING ABOUT ITS FALSE NEGATIVES, and the only way to find these was to read a record's whole layer. Quotation withheld because this record is not cleared for redistribution. GNM GNM Quotation withheld because this record is not cleared for redistribution. isolation isolation Quotation withheld because this record is not cleared for redistribution. cross-validated LL cross-validated LL Quotation withheld because this record is not cleared for redistribution. firing rate firing rate Quotation withheld because this record is not cleared for redistribution. sound pressure level sound pressure level Quotation withheld because this record is not cleared for redistribution. intensity tuning intensity tuning Quotation withheld because this record is not cleared for redistribution. prediction prediction Quotation withheld because this record is not cleared for redistribution. latency latency Quotation withheld because this record is not cleared for redistribution. gain gain Quotation withheld because this record is not cleared for redistribution. 94 — single units recorded extracellularly from primary auditory cortex of awake ferrets Quotation withheld because this record is not cleared for redistribution. P = 4 × 10 −12 — significance of the GNM's cross-validated log-likelihood advantage over the linear model across neurons Quotation withheld because this record is not cleared for redistribution. r 2 (GN) = 0.12 and r 2 (LN) = 0.07 — explained variance of the two models for the example neuron Quotation withheld because this record is not cleared for redistribution. 0.40 bits per spike / 0.30 bits per spike — cross-validated log-likelihood of the GNM versus the linear model for the example neuron Quotation withheld because this record is not cleared for redistribution. 83% — proportion of the stimulus used as training data, the rest held out for cross-validation Quotation withheld because this record is not cleared for redistribution. 40% — median increase in frequency-tuning width (in octaves) of inhibition relative to excitation Quotation withheld because this record is not cleared for redistribution. 60% / 24% — share of analysable neurons classed as onset cells versus sustained-response cells Quotation withheld because this record is not cleared for redistribution. 60% — the share of recorded neurons that are clear onset cells, against 15% with a clear offset response - the heterogeneity a single mechanism would have to cover WHOSE RESULT THIS IS: Qin et al. 2007 ), although more recent work found - THE EVIDENCE FOR THIS ROW ATTRIBUTES THE WORK TO SOMEBODY ELSE. The cited name is the subject of a result verb and the sentence carries no first-person marker, so the finding, quantity or definition is the CITED author's rather than this record's. The claim is true and the quote is verbatim - what is wrong is whose result the corpus says it is. Nothing edited, nothing withdrawn: the row is marked so a page can name the source. Where the record is a commentary or review, its claim relations layer usually already names the paper being discussed - the record knows at one layer and forgets at another. Quotation withheld because this record is not cleared for redistribution. 90% isolation — the spike-sorting criterion, so the analysed population is already selected on cluster quality Quotation withheld because this record is not cleared for redistribution. 8 kHz — the upper bound on stimulus power, a consequence of upsampling that the paper points out itself 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/10.1152/jn.01173.2011.