Computational Models, Neural Mechanisms And Clinical Applications Of Visual Working Memory Precision

Visual working memory refers to the maintenance of the visual information to serve for the ongoing task,which is highly related to the majority of cognitive functions including attention,language,reasoning,intelligence,etc.Despite its great importance,the mechanism of visual working memory remains unclear,while two dominatnt theories are favored in the field: the discrete-slot theory and the continuous-resource theory.The center of argument between these two theories focuses on whether the storage of visual working memory content is discrete or continuous.Under the frame of continuous-resource theory,the variable precision(VP)model acts as a winning model in behavioral data fitting and has been widely accepted.The fundamental assumption of this model is that the memory precision changes trial-to-trial and item-to-item while the mean precision decreases with the increase of memory load.Although the VP model is statistically in favored than other computational models,few neural evidences have been found to support it.Further,how this model can be used in application is also unchartered.The current thesis provides neural supports and clinical applications to this model.The continuous recall tasks were used in the current thesis.In these tasks,subjects were firstly asked to remember visual features(orientation in part I and color in part II)of several items and hold them for a period of delay.Then in the probe array,they were required to recall the feature of one of the items by clicking on a colored wheel(in color experiments)or rotate the visual stimulus to a certain degree(in orientation experiments)using a computer mouse.In the three experiments of part I,the purpose was to explore the neural mechanism of the VP model,based on the assumption of the VP model which states that memory precision changes across trial.In the first experiment,univariate analysis of the functional magnetic resonance imaging(fMRI)was used to search for brain regions that have activities related to memory precision.Results revealed that activities in bilateral lateral occipital complex(LOC)were negatively correlated with trial-to-trial recall errors.Specifically,higher brain activities in LOC led to smaller recall errors,which equals to higher precision.Moreover,LOC activities also predicted the behavioral precision in computational model,which was obtained in another experiment finished outside the scanner.These results indicated the roles of the visual cortex in maintaining visual working memory precision.Further,the functional connectivity was stronger between LOC and inferior frontal junction(IFJ)when memory load was high,indicating the necessity of communication between the prefrontal cortex and the sensory cortex in highdemanding tasks.By using the transcranial magnetic stimulation(TMS)in experiment 2,we further proved the causal roles of LOC,early visual cortex(EVC)and IFJ in visual working memory.Results found that,TMS over all three brain areas improved visual working memory precision and this effect was the strongest over LOC.All these findings revealed that brain activities of LOC could track the changes of visual working memory precision and additionally provided neural evidences for the VP model.Yet further questions of how the variable precision is represented on the neural level and how it is related to memory load remain to be answered.To address these questions,in experiment 3,the inverted encoding model(one of multivariate analysis methods with fMRI data)was used to estimate memory representations in the visual cortex under the change of memory loads.Results showed that representations of the memory target in EVC were stronger when memory load was low,while representations in V3 a was stronger at a higher load and lasted longer than that in EVC.These results indicated a hierarchical structure in variable precision representation of visual working memory,and also indicated that higher-level cortex may manipulate the activity pattern of lower-level cortex.Experiments in chapter two adapted the VP model to the investigation of schizophrenia patients.Results proved that the VP model fitted better than other models in both schizophrenia and healthy control cases.First,visual working memory deficits in schizophrenia were found in both experiment 4 and 5.Further comparison of model parameters,surprisingly,showed compatible working memory capacity(experiment 4)and selective attention(experiment 5)between schizophrenia and healthy subjects,which challenged previous studies.Instead,worsened working memory performance in schizophrenia was derived from the larger variability in resource allocation,indicating their inferior ability in reasonably allocating resources item-toitem and trial-to-trial.Furthermore,this resource allocation variability was found to correlate to patients' symptom severity,which revealed the value of VP model in clinical application.To conclude,the current studies in this thesis have found neural correlates of trial-level working memory precision and have proved its causal relationship with behavioral performance.In addition,results have shown stronger memory representations in the sensory cortex at lower working memory load.All these results provided neural evidences for the VP model and the continuous-resource theory.Interestingly,the VP model further revealed that the variability of resource allocation was the main reason for the working memory impairment in schizophrenia,providing future directions for studies based on this model,which may help the diagnosis and treatment in schizophrenia.