CoChat
Sign Up
Cited by 315Neural and Behavioral Psychology StudiesVisual perception and processing mechanismsNeural dynamics and brain function

Reward value-based gain control: divisive normalization in parietal cortex

Open access
Louie, K., Grattan, L. E., Glimcher, P. W.·2011·Journal of Neuroscience
DOI 10.1523/jneurosci.1237-11.2011PMC PMC3285508
Open this paper in CoChat — free accountOpen PDF

Abstract

The representation of value is a critical component of decision making. Rational choice theory assumes that options are assigned absolute values, independent of the value or existence of other alternatives. However, context-dependent choice behavior in both animals and humans violates this assumption, suggesting that biological decision processes rely on comparative evaluation. Here we show that neurons in the monkey lateral intraparietal cortex encode a relative form of saccadic value, explicitly dependent on the values of the other available alternatives. Analogous to extra-classical receptive field effects in visual cortex, this relative representation incorporates target values outside the response field and is observed in both stimulus-driven activity and baseline firing rates. This context-dependent modulation is precisely described by divisive normalization, indicating that this standard form of sensory gain control may be a general mechanism of cortical computation. Such normalization in decision circuits effectively implements an adaptive gain control for value coding and provides a possible mechanistic basis for behavioral context-dependent violations of rationality.

AI snapshot

Written from the paper’s full text.

TL;DR

Neurons in monkey lateral intraparietal cortex (LIP) encode saccadic value relative to the values of other available alternatives, not as absolute values. Full divisive normalization best describes this context-dependent modulation, suggesting a canonical cortical gain-control mechanism for value coding.

Key findings

  • In a two-target task, LIP activity increased with reward inside the response field (Vin) and decreased with reward outside it (Vout): 12/16 neurons (75.0%) were positively correlated with Vin, and 18/27 (66.7%) were negatively correlated with Vout; population mean correlations were 0.33 and −0.21, respectively.
  • In a three-target task, stimulus-driven activity was suppressed by summed extra-RF value: 33/62 neurons (53.2%) were modulated, with 30/33 showing suppression; population mean correlation was −0.155 (p = 2.6×10−6). Activity in the absence of an RF target was also suppressed: 27/62 (43.6%) modulated, 21/27 suppressed; mean correlation −0.087 (p = 0.0019).
  • Full divisive normalization best fit individual neurons (MED-RF: 53/62, 85.5%; HIGH-RF: 16/17, 94.1%; LOW-RF: 26/41, 61.9%) and the combined population dataset (R² = 0.961) versus fractional (0.730), difference (0.601), and simple normalization (0.219); it also had the lowest cross-validated mean squared error (0.0393).
  • Suppression depended on extra-RF target value but not relative target distance (ANCOVA: value p = 0.004; distance p = 0.058; interaction p = 0.793), and behavioral measures such as fixation breaks, error rates, and reaction times showed no consistent dependence on Vout.

Methods

  • Three male rhesus monkeys performed two-target and three-target saccade tasks with water reward magnitudes assigned to targets inside or outside the LIP neuron’s response field. Two-target task: 27 LIP neurons (monkey D n = 17; monkey W n = 10), with a subset (n = 16) also tested with RF value varied. Three-target task: 62 neurons under MED-RF (monkey W n = 31; monkey B n = 31), plus 45 neurons under HIGH-RF or LOW-RF conditions.
  • Single-unit electrophysiology in LIP; activity analyzed mainly 0–1000 ms after peripheral target presentation and normalized to peak firing in single-target RF mapping trials.
  • Analyses included Pearson correlations, multiple regression with Vout and saccade latency, ANCOVA on a suppression index, and model fitting of fractional, difference, simple normalization, and full normalization models using AIC and leave-one-out cross-validation.

Limitations

  • Stated: the circuit mechanisms generating divisive normalization remain unknown, the biophysical implementation of the semisaturation term σ is unknown, and the causal role of normalized value coding in context-dependent choice remains to be explored.
  • Stated: visuospatial parameters may influence suppression when alternative targets are very close to the RF (distance effect p = 0.058, not significant here), and attention cannot be fully excluded as an alternative explanation, though the authors argue the parametric value effects are not easily reconciled with simple attentional modulation.
  • Evident: findings are restricted to macaque LIP, specific saccade tasks, and water reward magnitudes; some neurons were recorded under multiple value conditions, so across-condition analyses were restricted to population responses rather than individual-neuron comparisons.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1523/jneurosci.1237-11.2011

Cites

57 of 57 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

0 of 31 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    AIC values confirmed that the full divisive normalization model was the most likely of the examined value representations in the vast majority of LIP neurons (MED-RF: 53/62, 85.5%; HIGH-RF: 16/17, 94.1%; LOW-RF: 26/41, 61.9%).

    “Examination of the AIC values confirmed that the full divisive normalization model was the most likely of these underlying value representations in the vast majority of LIP neurons (MED-RF condition: n = 53/62, 85.5%; HIGH-RF condition: n = 16/17, 94.1%; LOW-RF conditions: n = 26/41, 61.9%).”
    Quote in the paperSubstantial claim
  • Findingown data

    The full normalization equation (Rmax = 3.86, σ = 1541 μl, β = 87.6 μl) characterized LIP responses in both RF target-present and RF target-absent conditions.

    “As seen in Figure 7, the full normalization equation (Rmax = 3.86, σ = 1541 µl, β = 87.6 µl) characterized LIP responses in both conditions, which is evident when the best-fit normalization model (black lines) is compared to the population responses (red and blue points).”
    Quote in the paperSubstantial claim
  • Findingown data

    AIC values confirmed that the full normalization model best explained the MED-RF condition data (AICfract = −31.8, AICdiff = −26.5, AICnorm_simp = −21.9, AICnorm_full = −70.9).

    “Examination of AIC values confirmed that the full normalization model best explained the MED-RF condition data (AICfract = −31.8, AICdiff = −26.5, AICnorm_simp = −21.9, AICnorm_full = −70.9).”
    Quote in the paperSubstantial claim
  • Findingown data

    In the three-target task, in RF target-absent trials, the average correlation coefficient remained significantly negative after saccade instruction.

    “In contrast, the average correlation coefficient in RF target-absent trials remained significantly negative.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.
  • Findingown data

    Many LIP neurons in the three-target task possessed motor-related activity: a presaccadic increase in firing rate selective for saccades into the RF.

    “Consistent with previous reports, many LIP neurons in this task possessed motor-related activity: a presaccadic increase in firing rate selective for saccades into the RF.”
    Quote in the paperSubstantial claim
  • May combine two claims. 37
  • Classified as probably not a substantial claim. 11

  • May combine two claims.

    Findingown data

    The simple normalization model failed to predict suppression below baseline levels observed in the absence of RF stimulation; only the full normalization model fully predicted the modulation of both RF-driven and baseline activity.

    “However, the simple normalization model failed to predict the suppression below baseline levels observed in the absence of RF stimulation; only the full normalization model fully predicted the modulation of both RF-driven and baseline activity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the three-target task, there was no discernible difference in the average relative position or in the relative position distributions of small and large value alternative targets.

    “However, examination of the relative target positions (distance from RF target) of small and large value alternative targets showed no discernible difference in the average relative position (distancesmall = 35.43 deg, distancelarge = 35.10 deg; p = 0.77, t-test) or the relative position distributions (p = 0.98, two-sample Kolmogorov-Smirnov test; Figure 4 inset).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the three-target task, the majority of LIP neurons were modulated by value context, with almost all showing suppression, and suppression was significant across the population.

    “The majority of LIP neurons were modulated by value context, with almost all neurons showing suppression (modulation: n = 33/62, 53.2%; suppression n = 30/33, 90.9%); furthermore, suppression was significant across the population of neurons (mean correlation −0.155; p = 2.6×10−6, t-test) and evident in the mean population activity (Figure 2C).”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the three-target task, the population of LIP neurons displayed Vout-dependent suppression in both RF target-present and RF target-absent conditions across much of the trial, with peak modulation approximately 200-300 ms after peripheral target onset.

    “As shown in Figure 3, the population of LIP neurons displayed Vout-dependent suppression in both RF target-present and RF target-absent conditions across much of the trial, with peak modulation occurring approximately 200–300 ms after peripheral target onset.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the three-target task, multiple regression showed that firing rate was significantly dependent on only value in 29 neurons, on both value and latency in 3 neurons, and on only latency in 1 neuron, suggesting that the observed modulation of LIP activity was primarily driven by the value context.

    “Of the recorded neurons, firing rate was significantly dependent on only value in 29 neurons, on both value and latency in 3 neurons, and on only latency in 1 neuron, suggesting that the observed modulation of LIP activity was primarily driven by the value context.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    The difference model predicted suppression even in RF target-absent conditions, but it could not adequately model both the RF target-driven and RF target-absent effects simultaneously.

    “In contrast, the difference model predicted suppression even in RF target-absent conditions; however, it could not adequately model both the RF target-driven and RF target-absent effects simultaneously.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the three-target task, for saccades to the small target, error rates varied by target configuration but were correlated with total value in only one monkey and uncorrelated across the combined data.

    “For saccades to the small target, error rates varied by target configuration (Monkey B, χ2(3) = 13.4, p = 0.004; Monkey W, χ2(3) = 17.2, p = 0.0006) but were correlated with total value in only one monkey (Monkey B: r = 0.83, p = 0.167; Monkey W: r = 0.95, p = 0.047) and uncorrelated across the combined data (r = 0.35, p = 0.401).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the three-target task, value modulation in the absence of RF stimulation was significant for a large fraction of LIP neurons and significant across the population.

    “This value modulation in the absence of RF stimulation was significant for a large fraction of LIP neurons (modulation n = 27/62, 43.6%; suppression 21/27, 77.8%) and significant across the population (mean correlation −0.087; p = 0.0019, t-test).”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the two-target task, the rate of incorrect saccadic choices did not vary consistently with Vout; both monkeys exhibited high success rates, with error rates independent of Vout condition in Monkey W and dependent but not significantly related to Vout in Monkey D.

    “Similarly, the rate of incorrect saccadic choices did not vary consistently with Vout; both monkeys exhibited high success rates (> 90%), with error rates independent of Vout condition in Monkey W (χ2(4) = 4.5, p = 0.34) and dependent but not significantly related to Vout in Monkey D (χ2(4) = 14.4, p = 0.006; r = −0.40, p = 0.51).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the three-target task, despite constant Vin levels, activity in trials with a target in the RF was suppressed as Vout increased, and in trials without a target in the RF, activity decreased as a function of alternative target values.

    “Despite constant Vin levels, activity in trials with a target in the RF (solid red symbols) was suppressed as Vout increased. Similarly, in trials without a target in the RF (solid blue symbols), activity decreased as a function of alternative target values.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the three-target task, both monkeys had low fixation break frequencies during cue presentation that were uncorrelated with total value in both monkeys individually and combined across animals.

    “Both monkeys had low fixation break frequencies during cue presentation (Monkey W: 2.8%; Monkey B: 4.0%) that were uncorrelated with total value in both monkeys individually (Monkey W: r = −0.36, p = 0.42; Monkey B: r = 0.28, p = 0.54) and combined across animals (r = 0.04, p = 0.90).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the 300 ms window preceding saccade initiation, average LIP activity was suppressed by Vout in RF target-absent conditions and unmodulated in RF target-present conditions.

    “consistent with the timepoint analysis, average LIP activity is suppressed by Vout in RF target-absent conditions (mean r = −.0.071, p = 3.7×10−4) and unmodulated in RF target-present (mean r = 0.024, p = 0.302) conditions.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the three-target task, increasing the value of the target outside the RF decreased LIP firing rates, and activity also decreased as target number increased from single to double to triple target conditions.

    “Consistent with the results described above, increasing the value of the target outside the RF in the two-target conditions decreased LIP firing rates (red, orange). Furthermore, activity was also a function of target number, decreasing from single target (black) to double target (red, orange) to triple target (yellow) conditions.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Findingown data

    In the two-target task, fixation break frequencies were low in both monkeys; in Monkey W they did not differ across Vout conditions, while in Monkey D they differed across conditions but were not significantly correlated with Vout magnitudes.

    “Fixation break frequencies were low in both monkeys (Monkey D: 9.2%; Monkey W: 1.6%); in Monkey W, these frequencies did not differ across Vout conditions (Monkey W: χ2(4) = 5.6, p = 0.23), while in Monkey D, these frequencies differed across conditions (χ2(4) = 20.6, p = 0.0004) but were not significantly correlated with Vout magnitudes (r = −0.87, p = 0.06).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The fractional and simple normalization models failed to account for suppression in the absence of a RF target because both incorporate only the quantity Vin in the numerator, predicting constant levels of activity in all RF-target absent conditions.

    “The fractional and simple normalization models failed to account for suppression in the absence of a RF target; because both models incorporate only the quantity Vin in the numerator, they predicted constant levels of activity in all RF-target absent conditions (simple normalization: R = 0; fractional: R = a).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Across the population, suppression was independent of the relative spatial location of the extra-RF target, supporting a modulation driven primarily by target value rather than spatial location.

    “Across the population, suppression was independent of the relative spatial location of the extra-RF target, supporting a modulation driven primarily by target value rather than spatial location.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    Targets at these locations did not activate the recorded neuron when presented alone, consistent with a contralateral visual bias in LIP.

    “Targets at these locations did not activate the recorded neuron when presented alone, consistent with a contralateral visual bias in LIP (Ben Hamed et al., 2001; Patel et al., 2010).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Ben Hamed et al., 2001

      Not checked in the free run.

    • cites Patel et al., 2010

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    The fractional model failed to account for background suppression in the absence of a RF target, and its predicted linear responses corresponded poorly to the observed nonlinear activity.

    “As noted previously, the fractional model failed to account for background suppression in the absence of a RF target, evident as a horizontal line in predicted activity (Figure 6, top); furthermore, the predicted linear responses correspond poorly to the observed nonlinear activity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The value of targets situated far from the RF can modulate both stimulus-driven activity and activity in the absence of direct feed-forward RF drive.

    “These results show that the value of targets situated far from the RF can modulate both stimulus-driven activity and activity in the absence of direct feed-forward RF drive.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The value of σ derived from the present study is ~1000 μl, a quantity large enough at normal volumes of reward to significantly shift value coding away from a strict fractional representation and towards a richer function of value context.

    “The value of σ derived from the present study is ~1000 µl, a quantity large enough at normal volumes of reward to significantly shift value coding away from a strict fractional representation and towards a richer function of value context.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    We find a monotonic relationship between the value of the target in the RF and LIP activity, as well as a parametric suppression related to the value of extra-RF targets.

    “First, we find a monotonic relationship between the value of the target in the RF and LIP activity, as well as a parametric suppression related to the value of extra-RF targets.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    In the three-target task, extra-RF targets were far from the RF center (average 35 deg) and located primarily in the opposite visual hemifield.

    “In the three-target task, extra-RF targets were far from the RF center (average 35 deg) and located primarily in the opposite visual hemifield.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    When examined across all data conditions, the full divisive normalization model explained the full dataset (R2 = 0.961) far better than the fractional (R2 = 0.730), difference (R2 = 0.601), or simple normalization (R2 = 0.219) models.

    “When examined across all data conditions, the full divisive normalization model explained the full dataset (R2 = 0.961) far better than the fractional (R2 = 0.730), difference (R2 = 0.601), or simple normalization (R2 = 0.219) models (Figure 6, right column).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The difference model predicted some RF target-absent suppression, but performed particularly poorly in characterizing the range of responses to different Vin conditions.

    “The difference model predicted some RF target-absent suppression, but performed particularly poorly in characterizing the range of responses to different Vin conditions.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisereasoning

    Linear models accurately characterized basic visual neuron responses such as orientation and direction selectivity, but they could not explain more complicated, nonlinear phenomena such as contrast saturation and nonspecific suppression.

    “While such linear models accurately characterized basic visual neuron responses such as orientation and direction selectivity, they could not explain more complicated, nonlinear phenomena such as contrast saturation and nonspecific suppression.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Premisecitation

    Some researchers have suggested that the difference in option values may better describe LIP activity.

    “More recently, motivated by computational models of two-alternative choice behavior, some researchers have suggested that the difference in option values may better describe LIP activity (Corrado et al., 2005).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Corrado et al., 2005

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    LIP encodes a context-dependent value representation, where the value configuration of the entire choice set controls both stimulus-driven and target-absent activity early in the valuation process.

    “Together, these results suggest that LIP encodes a context-dependent value representation, where the value configuration of the entire choice set controls both stimulus-driven and target-absent activity early in the valuation process.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Premisenone given

    Both fractional and difference models are consistent with LIP value modulation, but neither theoretical proposal is directly based on the structure and function of cortical circuits.

    “However, while both the fractional and difference models are consistent with LIP value modulation, neither theoretical proposal is directly based on the structure and function of cortical circuits.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Normalization in decision circuits effectively implements an adaptive gain control for value coding and provides a possible mechanistic basis for behavioral context-dependent violations of rationality.

    “Such normalization in decision circuits effectively implements an adaptive gain control for value coding and provides a possible mechanistic basis for behavioral context-dependent violations of rationality.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Early models of striate cortex described visual neuron function as a linear operation producing responses that are weighted sums of relevant inputs.

    “Early models of striate cortex described visual neuron function as a linear operation, producing responses that are weighted sums of relevant inputs (Hubel and Wiesel, 1962, 1968; Movshon et al., 1978).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Hubel and Wiesel, 1962, 1968

      Not checked in the free run.

    • cites Movshon et al., 1978

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Initial studies of value coding in LIP focused on a fractional representation of value due to its association with classic matching law behavior.

    “Initial studies of value coding in LIP (Platt and Glimcher, 1999; Sugrue et al., 2004) focused on a fractional representation of value due to its association with classic matching law behavior (Herrnstein, 1961).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Platt and Glimcher, 1999

      Not checked in the free run.

    • cites Sugrue et al., 2004

      Not checked in the free run.

    • cites Herrnstein, 1961

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    The relative representation of saccadic value in LIP incorporates target values outside the response field and is observed in both stimulus-driven activity and baseline firing rates.

    “Analogous to extra-classical receptive field effects in visual cortex, this relative representation incorporates target values outside the response field and is observed in both stimulus-driven activity and baseline firing rates.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • May combine two claims.

    Interpretationown data

    Activity positively related to Vout was absent in the two-target task, suggesting that such activity may be a salience effect related to the instructional offset of the peripheral targets which only occurred in the three-target task.

    “Notably, activity positively related to Vout was absent in the two-target task (Figure 1C), suggesting that such activity may be a salience effect related to the instructional offset of the peripheral targets which only occurred in the three-target task.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    The general results presented in the paper are not sensitive to the exact definition of the 1000 ms target array epoch.

    “This epoch represents the entire period of target array presentation prior to the signal identifying the saccade target; the general results presented here are not sensitive to the exact definition of this epoch.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    LIP activity was both an increasing function of Vin and a decreasing function of Vout, suggesting that LIP value representation is relative rather than absolute.

    “Thus, activity was both an increasing function of Vin and a decreasing function of Vout, suggesting that LIP value representation is relative rather than absolute; we explore the specific algorithm underlying this relative representation below.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The value of a given target is equated with the associated reward magnitude.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Normalization-style models yield efficient coding of natural signal statistics in both visual and auditory modalities and may play a role in attentional processing, suggesting that divisive normalization may serve as a canonical computational algorithm in cortical circuits.

    “Furthermore, normalization-style models yield efficient coding of natural signal statistics in both visual and auditory modalities (Schwartz and Simoncelli, 2001; Valerio and Navarro, 2003) and may play a role in attentional processing (Reynolds et al., 1999; Reynolds and Heeger, 2009), suggesting that divisive normalization may serve as a canonical computational algorithm in cortical circuits.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Schwartz and Simoncelli, 2001

      Not checked in the free run.

    • cites Valerio and Navarro, 2003

      Not checked in the free run.

    • cites Reynolds and Heeger, 2009

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    In sensory cortices, stimuli outside the classical receptive field can significantly modulate neuronal activity, and many extra-classical effects are well-described by divisive normalization models.

    “In sensory cortices, stimuli outside the classical receptive field can nonetheless significantly modulate neuronal activity. Many of these extra-classical effects are well-described by models where response is specified by the sensory properties of the stimulus inside the receptive field, divided by the weighted sum of the sensory properties of stimuli both outside and inside the receptive field (”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Heeger, 1992

      Not checked in the free run.

    • cites Carandini et al., 1997

      Not checked in the free run.

    • cites Cavanaugh et al., 2002

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Given the relatively small number of neurons recorded in multiple Vin conditions, analyses across all value conditions were restricted to the population responses.

    “Given the relatively small number of neurons recorded in multiple Vin conditions, we restricted analyses across all value conditions to the population responses.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Heeger proposed that visual cortical function involves a normalization stage where a given cell's response is divided by a quantity representing the pooled activity of a large number of other similar neurons.

    “To address these issues, Heeger proposed that visual cortical function involves a normalization stage, where a given cell’s response is divided by a quantity representing the pooled activity of a large number of other similar neurons (Heeger, 1992).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Heeger, 1992

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The divisive normalization model explains a number of nonlinear phenomena in striate and extrastriate cortex as well as responses in the ventral visual stream.

    “This divisive normalization model explains a number of nonlinear phenomena in striate and extrastriate cortex (Carandini et al., 1997; Britten and Heuer, 1999; Cavanaugh et al., 2002; Heuer and Britten, 2002) as well as responses in the ventral visual stream (Zoccolan et al., 2005).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Carandini et al., 1997

      Not checked in the free run.

    • cites Britten and Heuer, 1999

      Not checked in the free run.

    • cites Cavanaugh et al., 2002

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The initial ten trials in a block were discarded to avoid learning effects.

    “For both tasks, the initial ten trials in a block were discarded to avoid learning effects.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Together, these results indicate that full divisive normalization outperformed the alternative models in terms of both response characterization and predictive power.

    “Together, these results indicate that full divisive normalization outperformed the alternative models in terms of both response characterization and predictive power.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Due to nonlinearity introduced by the semisaturation term (σ), the divisive normalization representation can produce responses on a continuum between absolute and relative value coding; this form of adaptive gain control depends on the relative balance between total value and σ.

    “Due to nonlinearity introduced by the semisaturation term (σ), the divisive normalization representation can produce responses on a continuum between absolute and relative value coding; this form of adaptive gain control depends on the relative balance between total value and σ.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    At intermediate total values (V ~ σ), the denominator in the normalization model depends on both total value and σ, and activity grows increasingly sensitive to choice context, similar to a fractional model.

    “However, at intermediate total values (V ~ σ) the denominator in the normalization model depends on both total value and σ, and activity grows increasingly sensitive to choice context, similar to a fractional model (angled shading and angled contour lines).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Our results provide the first example of normalization outside a strictly sensory area, and the most anterior extension of divisive normalization coding in the dorsal stream.

    “Our results provide the first example of normalization outside a strictly sensory area, and the most anterior extension of divisive normalization coding in the dorsal stream.”
    Quote in the paperSubstantial claim
45 of 45

  • The paper relies on this without support.

    The neural basis of context-dependent valuation is unknown.

    “However, the neural basis of such context-dependent valuation is unknown.”

  • The paper relies on this without support.

    The specific form of value representation in LIP and how it arises out of cortical circuit interactions remains largely unexplored.

    “Current evidence suggests that parietal neurons are modulated by the value of their associated saccade, but the specific form of value representation and how it arises out of cortical circuit interactions remains largely unexplored.”

  • The paper relies on this without support.

    LIP neurons have a circumscribed response field and many have a similarly tuned response to upcoming saccade amplitude and direction.

    “LIP neurons, like those in visual cortex, respond to stimulus presentation in a circumscribed region of visual space termed the response field (RF); many of these neurons also possess a similarly tuned response to upcoming saccade amplitude and direction.”

  • The authors state this as an assumption.

    The value of a given target is equated with the associated reward magnitude.

    “The value of a given target was equated with the associated reward magnitude.”
4 of 4

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Similar results are obtained when target desirability is manipulated by relative magnitude, probability, or delay of expected reward.

    “Similar results are obtained in tasks that manipulate the desirability of a target using different methods, for example by controlling the relative magnitude, probability or delay of its expected reward (Bernacchia et al., 2011; Louie et al., 2011; Sugrue et al., 2004; Yang and Shadlen, 2007).”

    From Attention, learning, and the value of information (2012) · cites this paper as Louie et al., 2011

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Consistentχ²(4) = 5.6reported p = .23 · recomputed p = .231

    “χ2(4) = 5.6, p = .23”

  • Consistentχ²(4) = 20.6reported p = .0004 · recomputed p < .001

    “χ2(4) = 20.6, p = .0004”

  • Consistentχ²(4) = 4.5reported p = .34 · recomputed p = .343

    “χ2(4) = 4.5, p = .34”

  • Consistentχ²(4) = 14.4reported p = .006 · recomputed p = .006

    “χ2(4) = 14.4, p = .006”

  • Consistentχ²(3) = 13.4reported p = .004 · recomputed p = .004

    “χ2(3) = 13.4, p = .004”

  • Consistentχ²(3) = 17.2reported p = .0006 · recomputed p < .001

    “χ2(3) = 17.2, p = .0006”

6 of 6

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

  • LOW-RF neuron percentage does not match its count

    Minor

    The reported 26 of 41 neurons is 63.4%, not 61.9%.

    Reported 61.9 · recomputed 63.414634

1 of 1

Cites

Flagged first, then by citations

3 of the 60 references couldn’t be matched to a known paper.

Research works printed but not linked: 1

  1. Stephens DW Krebs JR Foraging theory 1986 Princeton, NJ Princeton University Press

Cited by

Sorted by citations

Related

Authors
Louie, K., Grattan, L. E., & Glimcher, P. W.
Affiliations
New York University (US)
ORCID
Venue
Journal of Neuroscience
Publisher
Society for Neuroscience
Published
2011-07-20
Volume
31
Issue
29
Pages
10627-10639
ISSN
  • 0270-6474 (ISSN-L)
  • 1529-2401
DOI
10.1523/jneurosci.1237-11.2011
PMC
PMC3285508
OpenAlex
W2069815761
Semantic Scholar
2c68a342adf5dd25439ef7f3f553afd1e9a0d1d4
Cited by
308
References
60
Citations by year
2026: 7 · 2025: 13 · 2024: 20 · 2023: 22 · 2022: 18 · 2021: 29 · 2020: 19 · 2019: 25 · 2018: 23 · 2017: 22 · 2016: 16 · 2015: 24 · 2014: 23 · 2013: 24 · 2012: 19
FWCI
11.36
Type
journal article
Language
en
Open access
Locations
  • Journal of Neuroscience (publishedVersion)
  • Europe PMC (PubMed Central) (publishedVersion)
  • PubMed Central (publishedVersion)
  • PubMed (publishedVersion)
Funders
National Eye Institute (R01EY010536)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
Keywords
divisive normalization, gain control, context-dependent choice, comparative analysis, decision-making, monkey visual cortex
MeSH terms
Action Potentials, Analysis of Variance, Animals, Choice Behavior, Macaca mulatta, Male, Models, Neurological, Neurons, Parietal Lobe, Photic Stimulation, Predictive Value of Tests, Reference Values, Reward, Time Factors, Chi-Square Distribution

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review