Priority-based transformations of stimulus representation in visual working memory

How does the brain prioritize among the contents of working memory (WM) to appropriately guide behavior? Previous work, employing inverted encoding modeling (IEM) of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) datasets, has shown that unprioritized memory items (UMI) are actively represented in the brain, but in a “flipped”, or opposite, format compared to prioritized memory items (PMI). To acquire independent evidence for such a priority-based representational transformation, and to explore underlying mechanisms, we trained recurrent neural networks (RNNs) with a long short-term memory (LSTM) architecture to perform a 2-back WM task. Visualization of LSTM hidden layer activity using Principal Component Analysis (PCA) confirmed that stimulus representations undergo a representational transformation–consistent with a flip—while transitioning from the functional status of UMI to PMI. Demixed (d)PCA of the same data identified two representational trajectories, one each within a UMI subspace and a PMI subspace, both undergoing a reversal of stimulus coding axes. dPCA of data from an EEG dataset also provided evidence for priority-based transformations of the representational code, albeit with some differences. This type of transformation could allow for retention of unprioritized information in WM while preventing it from interfering with concurrent behavior. The results from this initial exploration suggest that the algorithmic details of how this transformation is carried out by RNNs, versus by the human brain, may differ.     

Exact neural mass model for synaptic-based working memory

A synaptic theory of Working Memory (WM) has been developed in the last decade as a possible alternative to the persistent spiking paradigm. In this context, we have developed a neural mass model able to reproduce exactly the dynamics of heterogeneous spiking neural networks encompassing realistic cellular mechanisms for short-term synaptic plasticity. This population model reproduces the macroscopic dynamics of the network in terms of the firing rate and the mean membrane potential. The latter quantity allows us to gain insight of the Local Field Potential and electroencephalographic signals measured during WM tasks to characterize the brain activity. More specifically synaptic facilitation and depression integrate each other to efficiently mimic WM operations via either synaptic reactivation or persistent activity. Memory access and loading are related to stimulus-locked transient oscillations followed by a steady-state activity in the β-γ band, thus resembling what is observed in the cortex during vibrotactile stimuli in humans and object recognition in monkeys. Memory juggling and competition emerge already by loading only two items. However more items can be stored in WM by considering neural architectures composed of multiple excitatory populations and a common inhibitory pool. Memory capacity depends strongly on the presentation rate of the items and it maximizes for an optimal frequency range. In particular we provide an analytic expression for the maximal memory capacity. Furthermore, the mean membrane potential turns out to be a suitable proxy to measure the memory load, analogously to event driven potentials in experiments on humans. Finally we show that the γ power increases with the number of loaded items, as reported in many experiments, while θ and β power reveal non monotonic behaviours. In particular, β and γ rhythms are crucially sustained by the inhibitory activity, while the θ rhythm is controlled by excitatory synapses.     

Distributed neural networks of tactile working memory

Microelectrode recordings of cortical activity in primates performing working memory tasks reveal some cortical neurons exhibiting sustained or graded persistent elevations in firing rate during the period in which sensory information is actively maintained in short-term memory. These neurons are called “memory cells”. Imaging and transcranial magnetic stimulation studies indicate that memory cells may arise from distributed cortical networks. Depending on the sensory modality of the memorandum in working memory tasks, neurons exhibiting memory-correlated patterns of firing have been detected in different association cortices including prefrontal cortex, and primary sensory cortices as well.Here we elaborate on neurophysiological experiments that lead to our understanding of the neuromechanisms of working memory, and mainly discuss findings on widely distributed cortical networks involved in tactile working memory.     

Visual working memory capacity does not modulate the feature-based information filtering in visual working memory

Background

The limited capacity of visual working memory (VWM) requires us to select the task relevant information and filter out the irrelevant information efficiently. Previous studies showed that the individual differences in VWM capacity dramatically influenced the way we filtered out the distracters displayed in distinct spatial-locations: low-capacity individuals were poorer at filtering them out than the high-capacity ones. However, when the target and distracting information pertain to the same object (i.e., multiple-featured object), whether the VWM capacity modulates the feature-based filtering remains unknown.

Methodology/Principal Findings

We explored this issue mainly based on one of our recent studies, in which we asked the participants to remember three colors of colored-shapes or colored-landolt-Cs while using two types of task irrelevant information. We found that the irrelevant high-discriminable information could not be filtered out during the extraction of VWM but the irrelevant fine-grained information could be. We added 8 extra participants to the original 16 participants and then split the overall 24 participants into low- and high-VWM capacity groups. We found that regardless of the VWM capacity, the irrelevant high-discriminable information was selected into VWM, whereas the irrelevant fine-grained information was filtered out. The latter finding was further corroborated in a second experiment in which the participants were required to remember one colored-landolt-C and a more strict control was exerted over the VWM capacity.

Conclusions/Significance

We conclude that VWM capacity did not modulate the feature-based filtering in VWM.     

Differential roles for parietal and occipital cortices in visual working memory

Visual working memory (VWM) is known as a highly capacity-limited cognitive system that can hold 3-4 items. Recent studies have demonstrated that activity in the intraparietal sulcus (IPS) and occipital cortices correlates with the number of representations held in VWM. However, differences among those regions are poorly understood, particularly when task-irrelevant items are to be ignored. The present fMRI-based study investigated whether memory load-sensitive regions such as the IPS and occipital cortices respond differently to task-relevant information. Using a change detection task in which participants are required to remember pre-specified targets, here we show that while the IPS exhibited comparable responses to both targets and distractors, the dorsal occipital cortex manifested significantly weaker responses to an array containing distractors than to an array containing only targets, despite that the number of objects presented was the same for the two arrays. These results suggest that parietal and occipital cortices engage differently in distractor processing and that the dorsal occipital, rather than parietal, activity appears to reflect output of stimulus filtering and selection based on behavioral relevance.     

From cognitive to neural models of working memory

Working memory refers to the temporary retention of information that was just experienced or just retrieved from long-term memory but no longer exists in the external environment. These internal representations are short-lived, but can be stored for longer periods of time through active maintenance or rehearsal strategies, and can be subjected to various operations that manipulate the information in such a way that makes it useful for goal-directed behaviour. Empirical studies of working memory using neuroscientific techniques, such as neuronal recordings in monkeys or functional neuroimaging in humans, have advanced our knowledge of the underlying neural mechanisms of working memory. This rich dataset can be reconciled with behavioural findings derived from investigating the cognitive mechanisms underlying working memory. In this paper, I review the progress that has been made towards this effort by illustrating how investigations of the neural mechanisms underlying working memory can be influenced by cognitive models and, in turn, how cognitive models can be shaped and modified by neuroscientific data. One conclusion that arises from this research is that working memory can be viewed as neither a unitary nor a dedicated system. A network of brain regions, including the prefrontal cortex (PFC), is critical for the active maintenance of internal representations that are necessary for goal-directed behaviour. Thus, working memory is not localized to a single brain region but probably is an emergent property of the functional interactions between the PFC and the rest of the brain.     

Differential activation requirements for virgin and memory T cells

Most studies of the activation requirements for T cells have used either T cell lines or populations of normal T cells that consist of a mixture of virgin and Ag-primed T cells. These two subpopulations of T cells can now be distinguished in humans by their reactivity with mAb. The anti-CD45R antibody HB10 identifies virgin T cells (T degrees) that are non-reactive to recall Ag and relatively poor at providing help for B cell differentiation. Conversely, memory T cells (T') that can react to recall Ag and enhance Ig production are non-reactive with anti-CD45R, but can be identified with the UCHL1 antibody. We have used these antibodies to separate the T degrees and T' populations and examine their activation requirements. On activation CD45R+ cells rapidly began to lose the CD45R Ag and express the UCHL1 Ag in increased amounts, whereas the UCHL1+ cells retained this phenotype. Both populations responded to PHA in the presence of monocytes, but when triggered by an antibody to CD3 only the T' cells were induced to express IL-2R, produce IL-2, and to proliferate. The T degrees population of cells remained relatively quiescent by all of these parameters. However, anti-CD3 stimulation conditioned the T degrees cells for IL-2 responsiveness, inasmuch as the addition of rIL-2 resulted in significant IL-2R expression and proliferation. When the CD4+ T degrees and CD4+ T' subpopulations were isolated and examined in the same assays similar results were obtained. The data indicate that fundamental differences exist in the triggering requirements for T degrees and T' cells.     

Noise,not stimulus entropy,determines neural information rate

In the quest for deciphering the neural code, theoretical advances were made which allow for the determination of the information rate inherent in the spike trains of nerve cells. However, up to now, the dependence of the information rate on stimulus parameters has not been studied in any neuron in a systematic way. Here, I investigate the information carried by the spike trains of H1, a motion-sensitive visual interneuron of the blowfly (Calliphora vicina) using a moving grating as a stimulus. Stimulus parameters fall in two classes: those that have only a minor effect on the information rate like increasing the frequency bandwidth or the maximum amplitude of the stimulus velocity, and those which dramatically affect the neural information rate, like varying the spatial size or the contrast of the visual pattern being moved. It appears that, for a broad range of complex stimuli, the neuron covers the stimulus with its whole response repertoire regardless of the stimulus entropy, with the information rate being limited by the noise of the stimulus and the neural hardware.     

Visual working memory load-related changes in neural activity and functional connectivity

Background

Visual working memory (VWM) helps us store visual information to prepare for subsequent behavior. The neuronal mechanisms for sustaining coherent visual information and the mechanisms for limited VWM capacity have remained uncharacterized. Although numerous studies have utilized behavioral accuracy, neural activity, and connectivity to explore the mechanism of VWM retention, little is known about the load-related changes in functional connectivity for hemi-field VWM retention.

Methodology/Principal Findings

In this study, we recorded electroencephalography (EEG) from 14 normal young adults while they performed a bilateral visual field memory task. Subjects had more rapid and accurate responses to the left visual field (LVF) memory condition. The difference in mean amplitude between the ipsilateral and contralateral event-related potential (ERP) at parietal-occipital electrodes in retention interval period was obtained with six different memory loads. Functional connectivity between 128 scalp regions was measured by EEG phase synchronization in the theta- (4–8 Hz), alpha- (8–12 Hz), beta- (12–32 Hz), and gamma- (32–40 Hz) frequency bands. The resulting matrices were converted to graphs, and mean degree, clustering coefficient and shortest path length was computed as a function of memory load. The results showed that brain networks of theta-, alpha-, beta-, and gamma- frequency bands were load-dependent and visual-field dependent. The networks of theta- and alpha- bands phase synchrony were most predominant in retention period for right visual field (RVF) WM than for LVF WM. Furthermore, only for RVF memory condition, brain network density of theta-band during the retention interval were linked to the delay of behavior reaction time, and the topological property of alpha-band network was negative correlation with behavior accuracy.

Conclusions/Significance

We suggest that the differences in theta- and alpha- bands between LVF and RVF conditions in functional connectivity and topological properties during retention period may result in the decline of behavioral performance in RVF task.     

Differential effects of ballistic versus sensorimotor training on rate of force development and neural activation in humans

Balancing exercises on instable bases (sensorimotor training [SMT]) are often used in the rehabilitation process of an injured athlete to restore joint function. Recently it was shown that SMT was able to enhance rate of force development (RFD) in a maximal voluntary muscle contraction. The purpose of this study was to compare adaptations on strength capacity following ballistic strength training (BST) with those following an SMT during a training period of 1 microcycle (4 weeks). Maximum voluntary isometric strength (MVC), maximum RFD (RFDmax) and the corresponding neural activation of M. soleus (SOL), M. gastrocnemius (GAS), and M. tibialis anterior (TIB) were measured during plantar flexion in 33 healthy subjects. The subjects were randomly assigned to a SMT, BST, or control group. RFDmax increased significantly stronger following BST (48 +/- 16%; p < 0.01) compared to SMT (14 +/- 5%; p < 0.05), whereas MVC remained unchanged in both groups. Median frequencies of the electromyographic power spectrum during the first 200 ms of contraction for GAS increased following both BST (45 +/- 21%; p < 0.05) and SMT (45 +/- 22%; p < 0.05), but median frequencies for SOL increased only after SMT (13 +/- 4%; p < 0.05). Additionally, mean amplitude voltage increased following BST for SOL (38 +/- 12%; p < 0.01) and for GAS (73 +/- 23%; p < 0.01) during the first 100 ms, whereas it remained unchanged after SMT. It is concluded that BST and SMT may induce different neural adaptations that specifically affect recruitment and discharge rates of motor units at the beginning of voluntary contraction. Specific neural adaptations indicate that SMT might be used complementarily to BST, especially in sports that require contractile explosive properties in situations with high postural demands, e.g., during jumps in ball sports.     

Dissociated mechanisms of extracting perceptual information into visual working memory

Background

The processing mechanisms of visual working memory (VWM) have been extensively explored in the recent decade. However, how the perceptual information is extracted into VWM remains largely unclear. The current study investigated this issue by testing whether the perceptual information was extracted into VWM via an integrated-object manner so that all the irrelevant information would be extracted (object hypothesis), or via a feature-based manner so that only the target-relevant information would be extracted (feature hypothesis), or via an analogous processing manner as that in visual perception (analogy hypothesis).

Methodology/Principal Findings

High-discriminable information which is processed at the parallel stage of visual perception and fine-grained information which is processed via focal attention were selected as the representatives of perceptual information. The analogy hypothesis predicted that whereas high-discriminable information is extracted into VWM automatically, fine-grained information will be extracted only if it is task-relevant. By manipulating the information type of the irrelevant dimension in a change-detection task, we found that the performance was affected and the ERP component N270 was enhanced if a change between the probe and the memorized stimulus consisted of irrelevant high-discriminable information, but not if it consisted of irrelevant fine-grained information.

Conclusions/Significance

We conclude that dissociated extraction mechanisms exist in VWM for information resolved via dissociated processes in visual perception (at least for the information tested in the current study), supporting the analogy hypothesis.     

Impact of load-related neural processes on feature binding in visuospatial working memory

Background

The capacity of visual working memory (WM) is substantially limited and only a fraction of what we see is maintained as a temporary trace. The process of binding visual features has been proposed as an adaptive means of minimising information demands on WM. However the neural mechanisms underlying this process, and its modulation by task and load effects, are not well understood.

Objective

To investigate the neural correlates of feature binding and its modulation by WM load during the sequential phases of encoding, maintenance and retrieval.

Methods and Findings

18 young healthy participants performed a visuospatial WM task with independent factors of load and feature conjunction (object identity and position) in an event-related functional MRI study. During stimulus encoding, load-invariant conjunction-related activity was observed in left prefrontal cortex and left hippocampus. During maintenance, greater activity for task demands of feature conjunction versus single features, and for increased load was observed in left-sided regions of the superior occipital cortex, precuneus and superior frontal cortex. Where these effects were expressed in overlapping cortical regions, their combined effect was additive. During retrieval, however, an interaction of load and feature conjunction was observed. This modulation of feature conjunction activity under increased load was expressed through greater deactivation in medial structures identified as part of the default mode network.

Conclusions and Significance

The relationship between memory load and feature binding qualitatively differed through each phase of the WM task. Of particular interest was the interaction of these factors observed within regions of the default mode network during retrieval which we interpret as suggesting that at low loads, binding processes may be ‘automatic’ but at higher loads it becomes a resource-intensive process leading to disengagement of activity in this network. These findings provide new insights into how feature binding operates within the capacity-limited WM system.     

Modulation of inferotemporal cortex activation during verbal working memory maintenance

Regions of the left inferotemporal cortex are involved in visual word recognition and semantics. We utilized functional magnetic resonance imaging to localize an inferotemporal language area and to demonstrate that this area is involved in the active maintenance of visually presented words in working memory. Maintenance activity in this inferotemporal area showed an effect of memory load for words, but not pseudowords. In the absence of visual input, the selective modulation of this language-related inferotemporal area for the maintenance of words is accompanied by an increased functional connectivity with left prefrontal cortex. These results demonstrate an involvement of inferotemporal cortex in verbal working memory and provide neurophysiological support for the notion that nonphonological language representations can be recruited in the service of verbal working memory. More generally, they suggest that verbal working memory should be conceptualized as the frontally guided, sustained activation of pre-existing cortical language representations.     

A neural model of the dynamic activation of memory

We study an Attractor Neural Network that stores natural concepts, organized in semantic classes. The concepts are represented by distributed patterns over a space of attributes, and are related by both semantic and episodic associations. While semantic relations are expressed through an hierarchical coding over the attribute space, episodic links are realized via specific synaptic projections. Due to dynamic thresholds expressing neuronal fatigue, the network's behavior is characterized by convergence toward the concept patterns on a short time scale, and by transitions between the various patterns on a longer time scale. In its baseline, undamaged state, the network manifests semantic, episodic, and random transitions, and demonstrates the phenomen of priming. Modeling possible pathological changes, we have found that increasing the noise level or the rate of neuronal fatigue decreases the frequency of semantic transitions. When neurons characterized by large synaptic connectivity are deleted, semantic transitions decay before the episodic ones, in accordance with the findings in patients with Alzheimer's disease.     

Biased belief priors versus biased belief updating: Differential correlates of depression and anxiety

Individuals prone to anxiety and depression often report beliefs and make judgements about themselves that are more negative than those reported by others. We use computational modeling of a richly naturalistic task to disentangle the role of negative priors versus negatively biased belief updating and to investigate their association with different dimensions of Internalizing psychopathology. Undergraduate participants first provided profiles for a hypothetical tech internship. They then viewed pairs of other profiles and selected the individual they would prefer to work alongside out of each pair. In a subsequent phase of the experiment, participants made judgments about their relative popularity as hypothetical internship partners both before any feedback and after each of 20 items of feedback revealing whether or not they had been selected as the preferred teammate from a given pairing. Scores on latent factors of general negative affect, anxiety-specific affect and depression-specific affect were estimated using participants’ self-report scores on standardized measures of anxiety and depression together with factor loadings from a bifactor analysis conducted previously. Higher scores on the depression-specific factor were linked to more negative prior beliefs but were not associated with differences in belief updating. In contrast, higher scores on the anxiety-specific factor were associated with a negative bias in belief updating but no difference in prior beliefs. These findings indicate that, to at least some extent, distinct processes may impact the formation of belief priors and in-the-moment belief updating and that these processes may be differentially disrupted in depression and anxiety. Future directions for enquiry include examination of the possibility that prior beliefs biases in depression might reflect generalization from prior experiences or global schema whereas belief updating biases in anxiety might be more situationally specific.     

How the speed of working memory updating influences the on-line thematic processing of simple sentences in Mandarin Chinese

This ERP study used electrophysiological technique to examine how individual differences in the speed of working memory updating influence the use of syntactic and semantic information during on-line sentence argument interpretation, and the time course of that working memory updating effect. The basic structure of the experimental sentences was “Noun + Verb + adverb + ‘le’ + a two-character word”, with the Noun being the sentence initial argument. This initial argument is animate or inanimate and the following verb disambiguates it as an agent or patient. The results at the initial argument revealed that, the quick-updating group elicited a larger positivity over the frontal cortex (within 500–800 ms post-noun onset) as compared with the slow-updating group. At the following disambiguating verb, the slow-updating group only showed a word order effect, indicating that the patient-first condition elicited a larger P600 (within 500–1,000 ms post-verb onset) than the agent-first one; for the quick-updating group, at the early stage of processing, the patient-first sentences elicited a larger N400 (within 300–500 ms post-verb onset) than the agent-first ones only when the initial argument was inanimate; however, at the late stage, the patient-first sentences elicited an enhanced P600 (within 800–1,000 ms post-verb onset) only when the initial argument was animate. These results suggested that the speed of working memory updating not only influences the maintenance of sentence argument when the contents of working memory change but also influences the efficiency of integrating that argument with the verb at a late time point. When integrating the argument with the disambiguating verb, individuals with quick-updating ability can combine multiple sources of information (both noun animacy and word order), and conduct rapid and fine-grained two-stage processing; individuals with slow-updating ability, however, only rely on one dominant source of information types (word order), and conducted slow and course-grained processing.     

Neural correlate of filtering of irrelevant information from visual working memory

In a dynamic environment stimulus task relevancy could be altered through time and it is not always possible to dissociate relevant and irrelevant objects from the very first moment they come to our sight. In such conditions, subjects need to retain maximum possible information in their WM until it is clear which items should be eliminated from WM to free attention and memory resources. Here, we examined the neural basis of irrelevant information filtering from WM by recording human ERP during a visual change detection task in which the stimulus irrelevancy was revealed in a later stage of the task forcing the subjects to keep all of the information in WM until test object set was presented. Assessing subjects' behaviour we found that subjects' RT was highly correlated with the number of irrelevant objects and not the relevant one, pointing to the notion that filtering, and not selection, process was used to handle the distracting effect of irrelevant objects. In addition we found that frontal N150 and parietal N200 peak latencies increased systematically as the amount of irrelevancy load increased. Interestingly, the peak latency of parietal N200, and not frontal N150, better correlated with subjects' RT. The difference between frontal N150 and parietal N200 peak latencies varied with the amount of irrelevancy load suggesting that functional connectivity between modules underlying fronto-parietal potentials vary concomitant with the irrelevancy load. These findings suggest the existence of two neural modules, responsible for irrelevant objects elimination, whose activity latency and functional connectivity depend on the number of irrelevant object.     

Unsupervised learning for robust working memory

Working memory is a core component of critical cognitive functions such as planning and decision-making. Persistent activity that lasts long after the stimulus offset has been considered a neural substrate for working memory. Attractor dynamics based on network interactions can successfully reproduce such persistent activity. However, it requires a fine-tuning of network connectivity, in particular, to form continuous attractors which were suggested for encoding continuous signals in working memory. Here, we investigate whether a specific form of synaptic plasticity rules can mitigate such tuning problems in two representative working memory models, namely, rate-coded and location-coded persistent activity. We consider two prominent types of plasticity rules, differential plasticity correcting the rapid activity changes and homeostatic plasticity regularizing the long-term average of activity, both of which have been proposed to fine-tune the weights in an unsupervised manner. Consistent with the findings of previous works, differential plasticity alone was enough to recover a graded-level persistent activity after perturbations in the connectivity. For the location-coded memory, differential plasticity could also recover persistent activity. However, its pattern can be irregular for different stimulus locations under slow learning speed or large perturbation in the connectivity. On the other hand, homeostatic plasticity shows a robust recovery of smooth spatial patterns under particular types of synaptic perturbations, such as perturbations in incoming synapses onto the entire or local populations. However, homeostatic plasticity was not effective against perturbations in outgoing synapses from local populations. Instead, combining it with differential plasticity recovers location-coded persistent activity for a broader range of perturbations, suggesting compensation between two plasticity rules.     

A working memory model for serial order that stores information in the intrinsic excitability properties of neurons

Models for temporary information storage in neuronal populations are dominated by mechanisms directly dependent on synaptic plasticity. There are nevertheless other mechanisms available that are well suited for creating short-term memories. Here we present a model for working memory which relies on the modulation of the intrinsic excitability properties of neurons, instead of synaptic plasticity, to retain novel information for periods of seconds to minutes. We show that it is possible to effectively use this mechanism to store the serial order in a sequence of patterns of activity. For this we introduce a functional class of neurons, named gate interneurons, which can store information in their membrane dynamics and can literally act as gates routing the flow of activations in the principal neurons population. The presented model exhibits properties which are in close agreement with experimental results in working memory. Namely, the recall process plays an important role in stabilizing and prolonging the memory trace. This means that the stored information is correctly maintained as long as it is being used. Moreover, the working memory model is adequate for storing completely new information, in time windows compatible with the notion of “one-shot” learning (hundreds of milliseconds).     

Differential neural responses evoked by orthonasal versus retronasal odorant perception in humans

Odors perceived through the mouth (retronasally) as flavor are referred to the oral cavity, whereas odors perceived through the nose (orthonasally) are referred to the external world. We delivered vaporized odorants via the orthonasal and retronasal routes and measured brain response with fMRI. Comparison of retronasal versus orthonasal delivery produced preferential activity in the mouth area at the base of the central sulcus, possibly reflecting olfactory referral to the mouth, associated with retronasal olfaction. Routes of delivery produced differential activation in the insula/operculum, thalamus, hippocampus, amygdala, and caudolateral orbitofrontal cortex in orthonasal > retronasal and in the perigenual cingulate and medial orbitofrontal cortex in retronasal > orthonasal in response to chocolate, but not lavender, butanol, or farnesol, so that an interaction of route and odorant may be inferred. These findings demonstrate differential neural recruitment depending upon the route of odorant administration and suggest that its effect is influenced by whether an odorant represents a food.