Flexible control of representational dynamics in a disinhibition-based model of decision-making
- Neuroscience Institute, New York University Grossman School of Medicine, United States
- Center for Neural Science, New York University, United States
Figures
Figure 1
Standard circuit motifs and neural dynamics in existing decision-making models.
(A) Dynamic normalization model (DNM). Each pair of excitatory (R) and inhibitory (G) units corresponds to an option in the choice set, with R receiving value-dependent input V and providing output. …
Figure 2 with 1 supplement
Local disinhibition decision model (LDDM) and its biological plausibility.
(A) LDDM extends the dynamic normalization model (DNM) by incorporating a disinhibitory D unit to mediate the local disinhibition of the associated excitatory R unit; strength of R to D coupling is …
Figure 2—figure supplement 1
Testing and comparing different dynamic normalization model (DNM) modifications for integrating normalized value coding and winner-take-all (WTA) competition.
(A) The full model contains all possible modifications that allow the original DNM to generate WTA competition. Modifications: recurrent excitation on R units (controlled by ), local disinhibition …
Figure 3
Normalized value coding in the local disinhibition decision model (LDDM).
(A) In this example, the LDDM receives a set of two input values with varying V1 (framed in red) and V2 (framed in blue). (B) Example LDDM dynamics show relative value coding. R1 activity shows a …
Figure 4 with 2 supplements
Quantitative comparison of contextual value coding across the local disinhibition decision model (LDDM), dynamic normalization model (DNM), and recurrent network model (RNM) models.
(A) Comparison between the LDDM (left), the DNM (middle), and the RNM (right) in value coding. The LDDM and the DNM show normalized value coding. The neural activity of R1 (indicated by color) …
Figure 4—figure supplement 1
Parameter recovery in fitting the local disinhibition decision model (LDDM) to normalized value coding data.
(A) The refitted parameter as a function of the used to generate the activities during model equilibrium shows high consistency when the parameters are within a reasonable range. (B) The …
Figure 4—figure supplement 2
Recurrent network model (RNM) activity fit to normalized value coding data.
The predicted dynamics of neural firing rates without scaling, including the activities of all three pools across different input conditions. The predicted firing rates show an unrealistic low …
Figure 5 with 1 supplement
Recurrent network model (RNM)-like winner-take-all (WTA) selection dynamics in the local disinhibition decision model (LDDM).
(A) Example R1 (solid) and R2 (dashed) dynamics in a classic reaction-time motion discrimination task. The model predicts phasic stimulus onset dynamics during the pre-stimulus stage and WTA …
Figure 5—figure supplement 1
Phase plane analyses of the local disinhibition decision model (LDDM) across a wide range of recurrent excitation strengths () and local disinhibition strengths ().
(A) The five different territories in the space of and are distinguished by the patterns of equilibria and stabilities of the system. (B–F) Example nullclines of R1 (solid bold line) and R2 …
Figure 6 with 5 supplements
The local disinhibition decision model (LDDM) performs well in capturing empirical behavior and neurophysiological data during perceptual decision-making.
(A) Model predicted RT distributions fit to behavioral data. Predicted RT distribution (lines) match the histogram of empirical RT distribution (bars), with correct and error trials separately …
Figure 6—figure supplement 1
Log-likelihood surfaces for local disinhibition decision model (LDDM) fit to Roitman and Shadlen (2002) behavioral data over the regimes of the seven free parameters.
Each two parameters were paired to show the log-likelihood space, with other parameters set as the best-fitted values. The contour lines indicate the isolines of log-likelihood, with the colors …
Figure 6—figure supplement 2
Parameter recovery of local disinhibition decision model (LDDM) for the parameters from the best fit to Roitman and Shadlen, 2002 behavioral data.
We visualized the log-likelihood of the model when re-fitting to the generated data based on the set of parameters of best fit (shown in blue crosses). Each panel shows the log-likelihood values of …
Figure 6—figure supplement 3
Collinearity between self-excitation and baseline gain control .
The log-likelihood space showed high collinearity between and Other parameters were set as the best-fitting values shown in Figure 6.
Figure 6—figure supplement 4
Fit of the original recurrent network model (RNM) to Roitman and Shadlen, 2002 behavioral data with eight free parameters.
All legends are consistent with the corresponding panels in Figure 6. (A) Model predicts RT distributions (lines) are slightly more right-skewed than the empirical RT distribution (histograms). (B) …
Figure 6—figure supplement 5
Fit of the leaky competing accumulator (LCA) model to Roitman and Shadlen, 2002 behavioral data, with four free parameters (Usher and McClelland, 2001).
All legends were kept consistent with the corresponding panels in Figure 6. (A) Model predicts RT distributions (lines) were slightly more right-skewed than the empirical data histogram (bars). (B) …
Figure 7
Local disinhibition decision model (LDDM) replicates both the normalized coding and winner-take-all (WTA) competition observed sequentially in single neurons examined in a multi-alternative choice task.
(A) Parietal neuron activity during pre-motion representation is decreased in four-alternative (red) versus two-alternative (black) trials. (B) Neural activity during two-alternative choice …
Figure 8 with 2 supplements
Local disinhibition decision model (LDDM) disinhibition controls the flexible implementation of either line attractor or point attractor dynamics in persistent activity.
(A–C) LDDM under silent disinhibition preserves the input ratio information during persistent activity. (A) Example R1 (solid) and R2 (dashed) activities before and after withdrawal of stimuli under …
Figure 8—figure supplement 1
Analysis of local disinhibition decision model (LDDM) persistent activity under generalized gain control weights.
(A–C) Phase plane analysis shows that systems with different gain control weights have different patterns of equilibria and stabilities. (A) When the lateral gain control () is weaker than the …
Figure 8—figure supplement 2
Local disinhibition decision model (LDDM) persistent activity under different levels of local disinhibition.
(A–C) Phase plane analysis of persistent activity for the situations of inactive disinhibition (, A), moderate intensity of disinhibition (, B), and strong disinhibition (, C). When , the …
Figure 9
Gated disinhibition flexibly adapts the dynamics of the circuit to various types of tasks.
All of the tasks consist of a pre-stimulus stage with equal inputs to R1 (solid) and R2 (dashed) and a stimulus stage with input values determined by the stimuli (indicated by grayscale, the same …
Figure 10
The modeling predictions of inhibitory potentiation to decision-making neural dynamics and behaviors.
(A) The predicted neural dynamics of pyramidal neurons (R1, solid lines, and R2, dashed lines) activities in a fixed duration decision task from local disinhibition decision model (LDDM). The …
Author response image 1
The shape of predicted reaction-time distribution over a wide range of α and β values by LDDM.
Each grid indicates the predicted RT histogram normalized in the range of minimum and maximum RTs. The shape of RT distribution exhibits a pattern of increasing skewness when α increases and …
