Working Memory and Decision-Making in a Frontoparietal Circuit Model

Working Memory and Decision-Making in a Frontoparietal Circuit Model Murray Jaramillo Wang

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.

Working Memory and Decision-Making in a Frontoparietal Circuit Model 2017 The Journal of Neuroscience 10.1523/jneurosci.0343-17.2017 Murray2017-working-memory-and-decision-making-in-a-fron Murray2017-working-memory-and-decision-making-in-a-fron.txt, the project's staged copy 35f956bdf7584ef3240d24ecfeb7862fd5145266 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 what two reciprocally connected cortical modules each contribute to working memory and decision-making, when their local recurrent strengths differ Quotation withheld because this record is not cleared for redistribution. why two areas whose activity looks similar during these tasks should be assumed to make comparable contributions at all Quotation withheld because this record is not cleared for redistribution. a tradeoff between optimising a circuit for distractor robustness and optimising it for slow evidence integration, and an architecture that ameliorates it by splitting the two across modules Quotation withheld because this record is not cleared for redistribution. mechanism Quotation withheld because this record is not cleared for redistribution. distributed working memory and decision-making in the frontoparietal circuit Quotation withheld because this record is not cleared for redistribution. simulation simulation Quotation withheld because this record is not cleared for redistribution. mechanisms Lateral inhibition and recurrent self-excitation, both set by J_S, jointly resist a distractor by counteracting activation of the distractor population and deactivation of the target population Quotation withheld because this record is not cleared for redistribution. mechanisms Structured PFC feedback (same-selectivity excitation plus cross-selectivity inhibition) onto a weak-attractor PPC pushes PPC back into the target memory state after the distractor ends Quotation withheld because this record is not cleared for redistribution. mechanisms Cross-inhibition from the active target population onto the distractor population produces the surround suppression that keeps distractor responses smaller than target responses Quotation withheld because this record is not cleared for redistribution. mechanisms Pathway-specific excitation–inhibition balance in the feedforward projection cancels the common-mode drive, so only the difference in activity between the two selection populations reaches the action module Quotation withheld because this record is not cleared for redistribution. mechanisms Slow synaptic gating variables mean the target population's suppressive effect on the distractor grows as the target-distractor delay lengthens, making later distractors weaker Quotation withheld because this record is not cleared for redistribution. mechanisms Stronger local recurrence in the PFC module widens its basin of attraction, so transient distractor inputs cannot dislodge the target-encoding state Quotation withheld because this record is not cleared for redistribution. the module receiving sensory input is a WEAK ATTRACTOR NETWORK so that it can TRANSIENTLY ENCODE DISTRACTORS and function as a saliency map, and so that INTEGRATION of perceptual evidence has A LONG TIMESCALE TO IMPROVE ACCURACY Quotation withheld because this record is not cleared for redistribution. and the downstream module is a STRONG ATTRACTOR NETWORK, which gives ROBUSTNESS AGAINST DISTRACTORS in memory and lets it RAMP QUICKLY TO THRESHOLD in choice - the same property serving two different functions Quotation withheld because this record is not cleared for redistribution. with the projections STRUCTURED - NET-EXCITATORY BETWEEN SAME-SELECTIVITY POPULATIONS AND NET-INHIBITORY BETWEEN DIFFERENT-SELECTIVITY POPULATIONS - which is what allows the feedback to SWITCH PPC BACK TO ENCODING THE TARGET after a distractor Quotation withheld because this record is not cleared for redistribution. and a PATHWAY-SPECIFIC E/I BALANCE in the feedforward projection, which regulates a cascade of activations and PROVIDES A MECHANISM FOR DISCRETE FLOW WITHOUT REQUIRING DIRECT GATING of neuronal responses Quotation withheld because this record is not cleared for redistribution. Distractor simulations fed stimulus input to the parietal module only, which made the prefrontal responses more attenuated than the empirical data. what this limits: the distractor-resistance result specifically: WE CONSIDERED INPUT TO THE PPC MODULE ONLY, WHICH RESULTED IN HIGHLY ATTENUATED RESPONSES IN THE PFC - the modelling choice produces the attenuation being reported Quotation withheld because this record is not cleared for redistribution. Simultaneous input to both modules, and how it would interact with the hierarchical recurrence, was not modeled. what this limits: the generality of that result: THE EFFECTS OF SIMULTANEOUS EXTERNAL INPUTS TO BOTH MODULES ... REMAINS A TOPIC FOR FUTURE COMPUTATIONAL STUDIES Quotation withheld because this record is not cleared for redistribution. The model uses homogeneous populations and cannot yet account for heterogeneous single-cell dynamics under stable population-level coding. what this limits: any comparison with single-cell data: the model uses homogeneous populations, and HOW TO INCLUDE SUFFICIENT HETEROGENEITY ... TO ACCOUNT FOR THE DYNAMIC RESPONSES OF INDIVIDUAL CELLS is an open question Quotation withheld because this record is not cleared for redistribution. modeling distractors used INPUT TO THE PPC MODULE ONLY, WHICH RESULTED IN HIGHLY ATTENUATED RESPONSES IN THE PFC - TO A GREATER EXTENT THAN THE FINDINGS it is compared against, so what was investigated was EFFECTIVELY A SUBCIRCUIT what this limits: responses to simultaneous external inputs to both modules, which remains a topic for future computational studies Quotation withheld because this record is not cleared for redistribution. recent INACTIVATION STUDIES FOUND THAT EVEN THOUGH PPC SHOWS DM-RELATED SIGNALS, IT DOES NOT PLAY A CAUSAL ROLE in perceptual decision-making - the tasks differ from the one modeled here what this limits: random-dot motion and auditory accumulation tasks, which are different from the visuospatial attention task modeled here Quotation withheld because this record is not cleared for redistribution. IT IS UNCLEAR WHETHER THESE RESULTS CAN BE ACCOUNTED FOR via compensatory mechanisms and/or distributed processing within a broader region, and the issue SHOULD BE ADDRESSED BY FUTURE EXPERIMENTS that consider task design and species-specific differences what this limits: the LIP conclusion: IT IS UNCLEAR WHETHER THESE RESULTS CAN BE ACCOUNTED FOR VIA COMPENSATORY MECHANISMS AND/OR DISTRIBUTED PROCESSING WITHIN A BROADER REGION IN PPC THAT INCLUDES LIP Quotation withheld because this record is not cleared for redistribution. the model is TWO POPULATIONS rather than a quasi-continuous network, so effects that depend on THE SIMILARITY AND DISTANCE BETWEEN DISTRACTORS AND TARGETS cannot be explored, and generalisation beyond two would be needed for set-size effects what this limits: any effect that depends on graded tuning similarity: the model is a TWO-POPULATION DISCRETE NETWORK rather than a quasi-continuous one, so such effects cannot appear in it Quotation withheld because this record is not cleared for redistribution. and HOW TO INCLUDE SUFFICIENT HETEROGENEITY to account for THE DYNAMIC RESPONSES OF INDIVIDUAL CELLS BUT STABLE REPRESENTATIONS AT THE POPULATION LEVEL is named as an open question what this limits: the same scope as limitations[2] - how to include sufficient heterogeneity to account for dynamic single-cell responses under a stable population representation is unresolved Quotation withheld because this record is not cleared for redistribution. attractor networks, in which strong recurrent excitation and lateral inhibition produce both a stable memory state and winner-take-all competition Quotation withheld because this record is not cleared for redistribution. pathway-specific excitation-inhibition balance, used here to produce discrete flow between stages without a gate Quotation withheld because this record is not cleared for redistribution. one parameter set held fixed across every working-memory and decision-making paradigm tested eliminates: a fit obtained by retuning per phenomenon - WITH A SINGLE SET OF NETWORK PARAMETERS the model captures salient empirical observations across both functions · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. the feedforward inhibition onto the action cells decreased systematically to break the balance eliminates: the balance being incidental - a STRONGLY IMBALANCED CIRCUIT PRODUCED A DISTINCT TYPE OF ERROR, with action cells ramping BEFORE SELECTION WAS ACCOMPLISHED · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. the same two modules run under both task families rather than one eliminates: an architecture tuned to one function - the tradeoff only appears because both are asked of the same circuit · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. ONE CIRCUIT CANNOT BE OPTIMAL FOR BOTH FUNCTIONS: A SINGLE LOCAL CIRCUIT FACES A TRADEOFF BETWEEN OPTIMIZATION FOR WM VERSUS DM FUNCTION - and the distributed model ameliorates it, capturing the observations WITH A SINGLE SET OF NETWORK PARAMETERS Quotation withheld because this record is not cleared for redistribution. with the tradeoff named on its two axes: optimizing the recurrent structure FOR ROBUSTNESS AGAINST DISTRACTORS IN WM VERSUS SLOW INTEGRATION OF EVIDENCE IN DM - ameliorated BY EXTENSION TO TWO MODULES WITH A HIERARCHICAL INCREASE IN RECURRENT STRENGTHS Quotation withheld because this record is not cleared for redistribution. THE MODULES ARE labeled PPC AND PFC THROUGHOUT, AND THEN MAY NOT BE AREAS AT ALL: BOTH MODULES COULD REPRESENT NEURAL POPULATIONS THAT ARE DISTRIBUTED ACROSS MULTIPLE AREAS - visual cells in LIP, FEF and SC, movement cells in FEF and SC. The anatomical names are a reading of the model, not a claim of it Quotation withheld because this record is not cleared for redistribution. THE CAUSAL EVIDENCE AGAINST THE modeled AREA IS CITED, AND ANSWERED BY TASK: RECENT INACTIVATION STUDIES FOUND THAT EVEN THOUGH PPC SHOWS DM-RELATED SIGNALS, IT DOES NOT PLAY A CAUSAL ROLE IN PERCEPTUAL DM - and the reply is that THE RANDOM-DOT MOTION AND AUDITORY ACCUMULATION TASKS WHERE SUCH DEFICITS WERE OBSERVED ARE DIFFERENT FROM THE VISUOSPATIAL ATTENTION TASK WE MODELED, with rat PPC inactivation disrupting VISUAL BUT NOT AUDITORY evidence Quotation withheld because this record is not cleared for redistribution. and the question is left to be settled by design rather than by argument: it SHOULD BE ADDRESSED BY FUTURE EXPERIMENTS THAT CAREFULLY CONSIDER BEHAVIORAL PERFORMANCE, TASK DESIGN, SPECIES-SPECIFIC DIFFERENCES, AND THE PRESENCE OF EVIDENCE ACCUMULATION IN MULTIPLE SIMULTANEOUSLY RECORDED AREAS Quotation withheld because this record is not cleared for redistribution. A MODEL RESULT THAT DOES NOT MATCH THE DATA, WITH THE MISMATCH USED RATHER THAN HIDDEN: input to one module only RESULTED IN HIGHLY ATTENUATED RESPONSES IN THE PFC, TO A GREATER EXTENT THAN THE FINDINGS it is compared with - THUS, IN OUR MODEL WE EFFECTIVELY INVESTIGATED A SUBCIRCUIT Quotation withheld because this record is not cleared for redistribution. THE PREMISE BEING ARGUED AGAINST IS AN INFERENCE FROM RESEMBLANCE: THE GENERAL SIMILARITY OF NEURAL ACTIVITY OF PPC AND PFC DURING WM AND DM HAS SUPPORTED THE VIEW THAT THEY MAKE COMPARABLE CONTRIBUTIONS TO THESE FUNCTIONS Quotation withheld because this record is not cleared for redistribution. and the alternative readings of the same signal are both live: persistent activity MAY BE A LOCALLY GENERATED PHENOMENON OR, ALTERNATIVELY, THE RESULT OF DISTRIBUTED INTERAREAL INTERACTIONS - and IT IS UNCLEAR TO WHAT DEGREE the specialization may be due to INTRINSIC DIFFERENCES IN LOCAL MICROCIRCUITRY Quotation withheld because this record is not cleared for redistribution. with the unit of description itself in question: recordings have revealed A DIVERSITY OF FUNCTIONALLY DEFINED CELL TYPES, raising QUESTIONS OF THE DIVISION OF LABOR AMONG BRAIN AREAS, OR FUNCTIONAL CELL TYPES - areas or cell types, and the paper does not choose Quotation withheld because this record is not cleared for redistribution. THE SAME PROPERTY IS GOOD FOR TWO DIFFERENT REASONS, WHICH IS WHY THE ARCHITECTURE WORKS: the strong attractor gives ROBUSTNESS AGAINST DISTRACTORS in memory and lets the module RAMP QUICKLY TO THRESHOLD in choice - and it PREDICTS A DIFFERENCE IN THE REPRESENTATION OF ACCUMULATED VALUE, MORE GRADED IN MODULE 1 AND MORE CATEGORICAL IN MODULE 2 Quotation withheld because this record is not cleared for redistribution. and the weak attractor is chosen for what it fails to hold: it can TRANSIENTLY ENCODE DISTRACTORS, its memory state CAN BE EFFECTIVELY CONTROLLED BY WEAK FEEDBACK, and it supports INTEGRATION WITH A LONG TIMESCALE TO IMPROVE ACCURACY Quotation withheld because this record is not cleared for redistribution. with the long-range projections STRUCTURED rather than uniform - NET-EXCITATORY BETWEEN SAME-SELECTIVITY POPULATIONS AND NET-INHIBITORY BETWEEN DIFFERENT-SELECTIVITY ones, which is what lets feedback SWITCH PPC BACK TO ENCODING THE TARGET Quotation withheld because this record is not cleared for redistribution. A PROPERTY DECLARED NOT TO BE LOCAL: the effective strength of the lateral inhibition in PPC IS NOT A PURELY LOCAL PROPERTY AND CAN BE FLEXIBLY CONTROLLED BY TOP-DOWN PREFRONTAL ENGAGEMENT - so a measurement of that circuit is a measurement of its inputs too Quotation withheld because this record is not cleared for redistribution. and the model is made to say what would falsify it: OUR MODEL PREDICTS THAT PFC INACTIVATION SHOULD DISRUPT THIS MODULATION of surround suppression in PPC, and that weak feedback puts the system in a REMEMBER-LAST regime while engagement switches it to REMEMBER-FIRST Quotation withheld because this record is not cleared for redistribution. with a second prediction already matched: representations of accumulated evidence are MORE GRADED IN PPC AND MORE CATEGORICAL IN PFC, AS SUPPORTED BY SINGLE-NEURON RECORDINGS Quotation withheld because this record is not cleared for redistribution. BREAKING THE MECHANISM PRODUCES A DISTINCT ERROR RATHER THAN A DEGRADED ONE: a STRONGLY IMBALANCED CIRCUIT PRODUCED A DISTINCT TYPE OF ERROR - ACTION CELLS STARTED RAMPING BEFORE THE POPULATIONS OF SELECTION CELLS HAD DIVERGED, that is BEFORE SELECTION WAS ACCOMPLISHED - and the modulation gave A SMOOTH TRADEOFF BETWEEN SPEED AND ACCURACY, A PLAUSIBLE MECHANISM AMONG OTHERS Quotation withheld because this record is not cleared for redistribution. and the balance is what removes the need for a separate gate: PATHWAY-SPECIFIC BALANCE THEREBY PROVIDES A MECHANISM FOR DISCRETE FLOW WITHOUT REQUIRING DIRECT GATING OF NEURONAL RESPONSES Quotation withheld because this record is not cleared for redistribution. THE FRAMEWORK IS FLEXIBLE ENOUGH TO DO BOTH JOBS, WHICH IS THE PROBLEM IT CREATES: attractor networks CONSTITUTE A FLEXIBLE TYPE OF COGNITIVE CIRCUIT CAPABLE OF PERFORMING BOTH WM AND DM - so the same architecture explains both, and the specialisation has to come from somewhere else Quotation withheld because this record is not cleared for redistribution. and the claim is made for design principles rather than for the brain: the study SUGGESTS IMPORTANT DESIGN PRINCIPLES FOR CONSTRUCTING MULTIREGIONAL NEURAL CIRCUIT MODELS, and the model MAY THEREFORE INSTANTIATE FEATURES OF A CANONICAL COGNITIVE CIRCUIT USEFUL FOR STUDYING distributed computation Quotation withheld because this record is not cleared for redistribution. with the division of labor stated as the finding: PPC TRANSIENTLY ENCODES DISTRACTORS WHILE PFC FILTERS DISTRACTORS AND SUPPORTS WM ROBUSTNESS, and the PPC module generates GRADED representations of accumulated evidence while PFC generates MORE CATEGORICAL responses Quotation withheld because this record is not cleared for redistribution. and the limits are listed as extensions: a QUASI-CONTINUOUS network for distractor-target similarity, MORE THAN TWO POPULATIONS for set-size effects, and A SPIKING CIRCUIT MODEL for synchronisation between the areas Quotation withheld because this record is not cleared for redistribution. with the hardest one named separately: how to include SUFFICIENT HETEROGENEITY to account for THE DYNAMIC RESPONSES OF INDIVIDUAL CELLS BUT STABLE REPRESENTATIONS AT THE POPULATION LEVEL Quotation withheld because this record is not cleared for redistribution. claim_type: mechanism a second reading reaching the same value. Recorded: mechanism — 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: a parsimonious two-module attractor model with hierarchical differences in recurrent strength and structured long-range projections, tested with a single parameter set a second reading DIFFERING from the recorded value. Recorded: simulation — 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: stimulus-selective persistent activity and ramping evidence accumulation, which appear in both areas and have been read as comparable contributions a second reading DIFFERING from the recorded value. Recorded: distributed working memory and decision-making in the frontoparietal circuit — 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 → Wang (2002) and Wong and Wang (2006) local attractor-network models of working memory and decision-making Quotation withheld because this record is not cleared for redistribution. extends → Hanks et al. (2015) Quotation withheld because this record is not cleared for redistribution. extends → Murray et al. (2014b) Quotation withheld because this record is not cleared for redistribution. critiques → the view that two areas make comparable contributions because their activity looks similar during the same tasks differs in: the local recurrent strength, which differs hierarchically between the modules Quotation withheld because this record is not cleared for redistribution. qualifies → inactivation results showing that parietal cortex is not causally required for perceptual decisions differs in: the task - random-dot motion and auditory accumulation against the visuospatial attention task modeled here Quotation withheld because this record is not cleared for redistribution. supports → single-neuron recordings showing more graded evidence representations in parietal and more categorical ones in prefrontal cortex Quotation withheld because this record is not cleared for redistribution. extends → local attractor-network models of working memory and decision-making, by asking what changes when two such circuits are coupled Quotation withheld because this record is not cleared for redistribution. predicts → that inactivating prefrontal cortex should disrupt the top-down modulation of surround suppression in parietal cortex Quotation withheld because this record is not cleared for redistribution. supports → combined stimulation-imaging work in humans showing a role for prefrontal feedback in distractor filtering Quotation withheld because this record is not cleared for redistribution. circuit —causes→ tradeoff circuit tradeoff strength of the relation: 1 · note: A SINGLE LOCAL CIRCUIT FACES A TRADEOFF BETWEEN OPTIMIZATION FOR WM VERSUS DM FUNCTION - and the distributed model captures the observations WITH A SINGLE SET OF NETWORK PARAMETERS Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span says a single circuit faces a tradeoff; causes is loose. Quotation withheld because this record is not cleared for redistribution. cell type —part-of→ circuit cell type circuit strength of the relation: 1 · note: A DIVERSITY OF FUNCTIONALLY DEFINED CELL TYPES, raising QUESTIONS OF THE DIVISION OF LABOR AMONG BRAIN AREAS, OR FUNCTIONAL CELL TYPES - the unit of description is left open Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span says frontoparietal circuits contain diverse cell types; part-of is loose. Quotation withheld because this record is not cleared for redistribution. attractor —operationalised-as→ recurrent attractor recurrent strength of the relation: 1 · note: attractor networks CONSTITUTE A FLEXIBLE TYPE OF COGNITIVE CIRCUIT CAPABLE OF PERFORMING BOTH WM AND DM - which is why specialisation must come from elsewhere Audit 2026-09-28: DOUBTFUL - read against its span and kept, but attractor networks work through strong recurrent interactions; operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. feedback —causes→ inhibition feedback inhibition strength of the relation: 1 · note: A PROPERTY DECLARED NOT TO BE LOCAL - the effective strength of the lateral inhibition in PPC IS NOT A PURELY LOCAL PROPERTY AND CAN BE FLEXIBLY CONTROLLED BY TOP-DOWN PREFRONTAL ENGAGEMENT Audit 2026-09-28: DOUBTFUL - read against its span and kept, but the span suggests lateral inhibition can be controlled by top-down feedback; causes is loose. Quotation withheld because this record is not cleared for redistribution. curve fitting (exponential) curve fitting (exponential) Quotation withheld because this record is not cleared for redistribution. autocorrelation autocorrelation Quotation withheld because this record is not cleared for redistribution. standard deviation standard deviation Quotation withheld because this record is not cleared for redistribution. curve fitting (sigmoid) curve fitting (sigmoid) Quotation withheld because this record is not cleared for redistribution. mean mean Quotation withheld because this record is not cleared for redistribution. quantile binning quantile binning Quotation withheld because this record is not cleared for redistribution. correlation correlation Quotation withheld because this record is not cleared for redistribution. psychometric function fitting psychometric function fitting Quotation withheld because this record is not cleared for redistribution. Recurrent structure must pass a threshold for winner-take-all selection, but increasing it further limits the gradual accumulation of sensory evidence. Quotation withheld because this record is not cleared for redistribution. a TRADEOFF EXISTS WHEN OPTIMIZING THE RECURRENT STRUCTURE FOR ROBUSTNESS AGAINST DISTRACTORS IN WM VERSUS SLOW INTEGRATION OF EVIDENCE IN DM, and the model AMELIORATES THIS TRADEOFF BY EXTENSION TO TWO MODULES with a hierarchical increase in recurrent strengths Quotation withheld because this record is not cleared for redistribution. Both modules hold persistent activity during working memory, but the PPC module encodes intervening distractors transiently while the PFC module filters them out and keeps the memory robust. Quotation withheld because this record is not cleared for redistribution. During perceptual decision-making the PPC module represents accumulated evidence in a graded fashion while the PFC module produces more categorical, choice-related responses. Quotation withheld because this record is not cleared for redistribution. Observing persistent activity in one area of a distributed circuit does not necessarily imply that the area is independently capable of multistability. Quotation withheld because this record is not cleared for redistribution. Feedback projections from the PFC module are what switch the PPC module back to encoding the target after a distractor is withdrawn. Quotation withheld because this record is not cleared for redistribution. WITHOUT FEEDBACK FROM PFC, THE DISTRACTOR STIMULUS SWITCHES THE NETWORK TO ENCODING THE DISTRACTOR, WHICH IT CONTINUES TO ENCODE THROUGH THE SUBSEQUENT DELAY Quotation withheld because this record is not cleared for redistribution. For the given parameters, the PFC module's intrinsic fluctuation timescale is 3.44 times slower than the PPC module's. Quotation withheld because this record is not cleared for redistribution. relative to PPC, the surround suppression of the distractor population in PFC is STRONGER, toward a lower baseline activity, and MORE RAPID, with a lower time constant Quotation withheld because this record is not cleared for redistribution. Weakening feedforward inhibition in the selection-to-action pathway makes responses faster but less accurate, i.e. it implements a speed–accuracy tradeoff. Quotation withheld because this record is not cleared for redistribution. The tradeoff is ameliorated when local recurrent structure increases hierarchically from the PPC/selection module to the PFC/action module. Quotation withheld because this record is not cleared for redistribution. Breaking pathway-specific excitation–inhibition balance creates a new error type in which action cells commit to a quasi-random choice before the selection cells have discriminated the target. Quotation withheld because this record is not cleared for redistribution. When the feedforward pathway is balanced, every simulated error comes from the selection cells mis-selecting the distractor rather than the target. Quotation withheld because this record is not cleared for redistribution. A SINGLE LOCAL CIRCUIT FACES A TRADEOFF BETWEEN OPTIMIZATION FOR WM VERSUS DM FUNCTION Quotation withheld because this record is not cleared for redistribution. the tradeoff is between ROBUSTNESS AGAINST DISTRACTORS IN WM VERSUS SLOW INTEGRATION OF EVIDENCE IN DM, ameliorated BY EXTENSION TO TWO MODULES WITH A HIERARCHICAL INCREASE IN RECURRENT STRENGTHS Quotation withheld because this record is not cleared for redistribution. BOTH MODULES COULD REPRESENT NEURAL POPULATIONS THAT ARE DISTRIBUTED ACROSS MULTIPLE AREAS Quotation withheld because this record is not cleared for redistribution. RECENT INACTIVATION STUDIES FOUND THAT EVEN THOUGH PPC SHOWS DM-RELATED SIGNALS, IT DOES NOT PLAY A CAUSAL ROLE IN PERCEPTUAL DM, and IN RATS, INACTIVATION OF PPC DISRUPTS THE ACCUMULATION OF VISUAL BUT NOT AUDITORY EVIDENCE Quotation withheld because this record is not cleared for redistribution. input to the PPC module only RESULTED IN HIGHLY ATTENUATED RESPONSES IN THE PFC, TO A GREATER EXTENT THAN THE FINDINGS compared against - THUS, IN OUR MODEL WE EFFECTIVELY INVESTIGATED A SUBCIRCUIT Quotation withheld because this record is not cleared for redistribution. THE GENERAL SIMILARITY OF NEURAL ACTIVITY OF PPC AND PFC DURING WM AND DM HAS SUPPORTED THE VIEW THAT THEY MAKE COMPARABLE CONTRIBUTIONS TO THESE FUNCTIONS Quotation withheld because this record is not cleared for redistribution. WM-related persistent activity MAY BE A LOCALLY GENERATED PHENOMENON OR, ALTERNATIVELY, THE RESULT OF DISTRIBUTED INTERAREAL INTERACTIONS Quotation withheld because this record is not cleared for redistribution. single-neuron recordings have revealed A DIVERSITY OF FUNCTIONALLY DEFINED CELL TYPES, raising QUESTIONS OF THE DIVISION OF LABOR AMONG BRAIN AREAS, OR FUNCTIONAL CELL TYPES Quotation withheld because this record is not cleared for redistribution. representations of accumulated evidence are MORE GRADED IN PPC AND MORE CATEGORICAL IN PFC, AS SUPPORTED BY SINGLE-NEURON RECORDINGS Quotation withheld because this record is not cleared for redistribution. the authors suggest that the effective strength of lateral inhibition in PPC IS NOT A PURELY LOCAL PROPERTY AND CAN BE FLEXIBLY CONTROLLED BY TOP-DOWN PREFRONTAL ENGAGEMENT Quotation withheld because this record is not cleared for redistribution. OUR MODEL PREDICTS THAT PFC INACTIVATION SHOULD DISRUPT THIS MODULATION OF SURROUND SUPPRESSION IN PPC Quotation withheld because this record is not cleared for redistribution. PATHWAY-SPECIFIC BALANCE PROVIDES A MECHANISM FOR DISCRETE FLOW WITHOUT REQUIRING DIRECT GATING OF NEURONAL RESPONSES Quotation withheld because this record is not cleared for redistribution. A STRONGLY IMBALANCED CIRCUIT PRODUCED A DISTINCT TYPE OF ERROR - action cells STARTED RAMPING BEFORE THE POPULATIONS OF SELECTION CELLS HAD DIVERGED IN ACTIVITY Quotation withheld because this record is not cleared for redistribution. attractor networks CONSTITUTE A FLEXIBLE TYPE OF COGNITIVE CIRCUIT CAPABLE OF PERFORMING BOTH WM AND DM Quotation withheld because this record is not cleared for redistribution. PPC TRANSIENTLY ENCODES DISTRACTORS WHILE PFC FILTERS DISTRACTORS AND SUPPORTS WM ROBUSTNESS Quotation withheld because this record is not cleared for redistribution. the model MAY INSTANTIATE FEATURES OF A CANONICAL COGNITIVE CIRCUIT USEFUL FOR STUDYING DISTRIBUTED COMPUTATION IN THE BRAIN Quotation withheld because this record is not cleared for redistribution. how to include SUFFICIENT HETEROGENEITY to account for THE DYNAMIC RESPONSES OF INDIVIDUAL CELLS BUT STABLE REPRESENTATIONS AT THE POPULATION LEVEL is an open question Quotation withheld because this record is not cleared for redistribution. structure of the network structure of the network Quotation withheld because this record is not cleared for redistribution. tone of the network tone of the network Quotation withheld because this record is not cleared for redistribution. module module Quotation withheld because this record is not cleared for redistribution. robust stimulus range robust stimulus range Quotation withheld because this record is not cleared for redistribution. discrimination threshold discrimination threshold Quotation withheld because this record is not cleared for redistribution. integration time constant integration time constant Quotation withheld because this record is not cleared for redistribution. error trial error trial Quotation withheld because this record is not cleared for redistribution. target-distractor onset asynchrony (TDOA) target-distractor onset asynchrony (TDOA) Quotation withheld because this record is not cleared for redistribution. attractor attractor Quotation withheld because this record is not cleared for redistribution. module module Quotation withheld because this record is not cleared for redistribution. recurrent structure recurrent structure Quotation withheld because this record is not cleared for redistribution. target-distractor onset asynchrony (TDOA) target-distractor onset asynchrony (TDOA) Quotation withheld because this record is not cleared for redistribution. strength of feedforward inhibition strength of feedforward inhibition Quotation withheld because this record is not cleared for redistribution. PFC → PPC feedback structure PFC → PPC feedback structure Quotation withheld because this record is not cleared for redistribution. strength of feedforward excitation strength of feedforward excitation Quotation withheld because this record is not cleared for redistribution. feedforward PPC → PFC structure feedforward PPC → PFC structure Quotation withheld because this record is not cleared for redistribution. reaction time reaction time Quotation withheld because this record is not cleared for redistribution. error rate error rate Quotation withheld because this record is not cleared for redistribution. discrimination threshold and robust stimulus range discrimination threshold and robust stimulus range Quotation withheld because this record is not cleared for redistribution. recurrent recurrent Quotation withheld because this record is not cleared for redistribution. inhibition inhibition Quotation withheld because this record is not cleared for redistribution. 3.44 times — how much slower the modeled PFC module's spontaneous fluctuation timescale is than the PPC module's Quotation withheld because this record is not cleared for redistribution. 127 ms, 438 ms — fluctuation timescales extracted from the autocorrelation of spontaneous firing in the PPC and PFC modules respectively Quotation withheld because this record is not cleared for redistribution. 0.13 and 0.19 — Quotation withheld Quotation withheld because this record is not cleared for redistribution. 81.6% — the accuracy level used to define the discrimination threshold in the simulated two-alternative forced choice task Quotation withheld because this record is not cleared for redistribution. ∼560 ms — median decision time of the moderate-structure network for a zero-contrast stimulus (slow integration) Quotation withheld because this record is not cleared for redistribution. ∼430 ms — median decision time of the higher-structure network for the same zero-contrast stimulus (curtailed integration) Quotation withheld because this record is not cleared for redistribution. 40, 12, and 7 Hz — firing-rate thresholds used to read out reaction time, target discrimination time and action-cell onset time Quotation withheld because this record is not cleared for redistribution. J S (2→2) = 0.4182 nA — the local recurrent structure assigned to Module 2 (PFC), higher than Module 1's 0.35 nA Quotation withheld because this record is not cleared for redistribution. a single set of network parameters — the constraint the model holds itself to across every paradigm tested, which is what makes the fit informative rather than fitted 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/10.1523/jneurosci.0343-17.2017.