Sequence Learning Task Research Articles

Dissociation between two aspects of procedural learning in Tourette syndrome: Enhanced statistical and impaired sequence learning

ABSTRACT Tourette syndrome (TS) is a childhood-onset neurodevelopmental disorder that primarily affects the cortico-basal ganglia-thalamo-cortical (CBGTC) circuitry and is characterized by motor and vocal tics. Previous studies have found enhancement in procedural memory, which depends on the CBGTC circuitry and plays an important role in the learning and processing of numerous motor, social, and cognitive skills and habits. Based on these studies, procedural hyperfunctioning in TS has been proposed. However, the neurocognitive mechanism underlying such hyperfunctioning is poorly understood. Here, we investigated how two aspects of procedural learning, namely 1) frequency-based statistical learning and 2) order-based sequence learning, are affected in TS. Twenty-one children with TS between the ages of ten and fifteen as well as 21 typically developing controls were tested on a probabilistic sequence learning task that enables the parallel assessment of these two aspects. We found that children with TS showed enhanced sensitivity to statistical information but impaired sequence learning compared to typically developing children. The deconstruction of procedural memory suggests that procedural hyperfunctioning in TS may be supported by enhanced sensitivity to statistical information. These results can provide a potential path for improving therapy methods and skill-oriented educational programs for TS.

Timing-specific effects of single-session M1 anodal tDCS on motor sequence retention in healthy older adults

Anodal transcranial direct current stimulation (tDCS) may assist in counteracting age-related decline in cognitive and motor functions. The current study investigated the potential impact of anodal tDCS, and the timing of its application, in mitigating age-related deficits in motor sequence learning.Forty-eight healthy older adults received, over the primary motor cortex (M1), tDCS – anodal and sham at least 1 week apart – before, during or after an explicit sequence-learning task with electrophysiological measures of corticospinal excitability (CSE) and short-interval intracortical inhibition (SICI) also obtained.Bayesian analyses revealed no generalised benefit of anodal tDCS to motor acquisition and immediate retention. Furthermore, there was not enough evidence to support timing-specific stimulation differences on performance during acquisition and immediate retention. However, performance at delayed retention – measured 24 ​h after acquisition – was worse in the anodal (13.1%) than sham (17.6%) tDCS session for the group receiving tDCS during sequence acquisition, but not before (anodal: 18.4%; sham: 16.7%) or after (anodal: 18.5%; sham: 16.3%) it. No corresponding task-specific stimulation-based changes in CSE and SICI were observed.Thus, single-session M1 anodal tDCS in healthy older adults not only proved ineffective in facilitating sequence acquisition and immediate retention but also, when administered during sequence learning, proved detrimental to delayed retention. Overall, these null and negative results may have implications for the use of tDCS in clinical and rehabilitative settings, especially in the elderly.

Daytime Exposure to Short Wavelength-Enriched Light Improves Cognitive Performance in Sleep-Restricted College-Aged Adults.

We tested the effect of daytime indoor light exposure with varying melanopic strength on cognitive performance in college-aged students who maintained an enforced nightly sleep opportunity of 7 h (i.e., nightly sleep duration no longer than 7 h) for 1 week immediately preceding the day of light exposure. Participants (n = 39; mean age ± SD = 24.5 ± 3.2 years; 21 F) were randomized to an 8 h daytime exposure to one of four white light conditions of equal photopic illuminance (~50 lux at eye level in the vertical plane) but different melanopic illuminance [24–45 melanopic-EDI lux (melEDI)] generated by varying correlated color temperatures [3000K (low-melEDI) or 5000K (high-melEDI)] and spectra [conventional or daylight-like]. Accuracy on a 2-min addition task was 5% better in the daylight-like high-melEDI condition (highest melEDI) compared to the conventional low-melEDI condition (lowest melEDI; p < 0.01). Performance speed on the motor sequence learning task was 3.2 times faster (p < 0.05) during the daylight-like high-melEDI condition compared to the conventional low-melEDI. Subjective sleepiness was 1.5 times lower in the conventional high-melEDI condition compared to the conventional low-melEDI condition, but levels were similar between conventional low- and daylight-like high-melEDI conditions. These results demonstrate that exposure to high-melanopic (short wavelength-enriched) white light improves processing speed, working memory, and procedural learning on a motor sequence task in modestly sleep restricted young adults, and have important implications for optimizing lighting conditions in schools, colleges, and other built environments.

Investigating the Effects of Seizures on Procedural Memory Performance in Patients with Epilepsy.

Memory complaints are frequently reported by patients with epilepsy and are associated with seizure occurrence. Yet, the direct effects of seizures on memory retention are difficult to assess given their unpredictability. Furthermore, previous investigations have predominantly assessed declarative memory. This study evaluated within-subject effects of seizure occurrence on retention and consolidation of a procedural motor sequence learning task in patients with epilepsy undergoing continuous monitoring for five consecutive days. Of the total sample of patients considered for analyses (N = 53, Mage = 32.92 ± 13.80 y, range = 18–66 y; 43% male), 15 patients experienced seizures and were used for within-patient analyses. Within-patient contrasts showed general improvements over seizure-free (day + night) and seizure-affected retention periods. Yet, exploratory within-subject contrasts for patients diagnosed with temporal lobe epilepsy (n = 10) showed that only seizure-free retention periods resulted in significant improvements, as no performance changes were observed following seizure-affected retention. These results indicate general performance improvements and offline consolidation of procedural memory during the day and night. Furthermore, these results suggest the relevance of healthy temporal lobe functioning for successful consolidation of procedural information, as well as the importance of seizure control for effective retention and consolidation of procedural memory.

Alterations of procedural memory consolidation in children with developmental dyslexia.

[Correction Notice: An Erratum for this article was reported in Vol 35(2) of Neuropsychology (see record 2021-29488-002). In the article "Alterations of Procedural Memory Consolidation in Children With Developmental Dyslexia," by Martina Hedenius, Jarrad A. G. Lum, and Sven Bölte (Neuropsychology, 2021, Vol. 35, No. 2, pp. 185-196, https://doi.org/10.1037/neu0000708), three corrections have been made. The university affiliation for Martina Hedenius was listed incorrectly and should have read Uppsala University. The figures have been rearranged so that they appear in two-column format. In the Results section, the heading "Overnight Consolidation of Sequence Specific Knowledge" has been changed to "Consolidation of Sequence Specific Knowledge." All versions of this article have been corrected.] Objective: The objective of the present study was to determine if consolidation of procedural memory in developmental dyslexia (DD) may be supported by prolonging the initial learning session. Method: Children with DD (n = 31) and typically developing (TD) children (n = 34) were given a procedural (implicit sequence) learning task over two sessions, separated by a 24-hr interval. Session 1 was prolonged with 50% compared with previous studies of procedural memory consolidation in DD. Results: The amount of sequence learning was similar in the two groups at the end of Session 1 (p = .797, ηρ² = .001). At the follow-up session after 24 hr, performance in the DD group was poorer compared with the TD group (p = .003, ηρ² = .129). The amount of sequence knowledge in Session 2 predicted unique variance in reading fluency (p = .024, ηρ² = 0.083), independent of children's phoneme awareness and inattention symptoms. Conclusion: The present study shows that consolidation of procedural memory is impaired in DD, even when the initial learning session is prolonged. The findings encourage further research on the mechanisms supporting procedural memory consolidation in children with DD, and how these may be strengthened. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Alterations of procedural memory consolidation in children with developmental dyslexia. Correction to Hedenius et al. (2020).

Reports an error in "Alterations of procedural memory consolidation in children with developmental dyslexia" by Martina Hedenius, Jarrad A. G. Lum and Sven Bölte (Neuropsychology, Advanced Online Publication, Nov 19, 2020, np). In the article, three corrections have been made. The university affiliation for Martina Hedenius was listed incorrectly and should have read Uppsala University. The figures have been rearranged so that they appear in two-column format. In the Results section, the heading "Overnight Consolidation of Sequence Specific Knowledge" has been changed to "Consolidation of Sequence Specific Knowledge." All versions of this article have been corrected. (The following abstract of the original article appeared in record 2020-86309-001.) Objective: The objective of the present study was to determine if consolidation of procedural memory in developmental dyslexia (DD) may be supported by prolonging the initial learning session. Method: Children with DD (n = 31) and typically developing (TD) children (n = 34) were given a procedural (implicit sequence) learning task over two sessions, separated by a 24-hr interval. Session 1 was prolonged with 50% compared with previous studies of procedural memory consolidation in DD. Results: The amount of sequence learning was similar in the two groups at the end of Session 1 (p = .797, ηρ² = .001). At the follow-up session after 24 hr, performance in the DD group was poorer compared with the TD group (p = .003, ηρ² = .129). The amount of sequence knowledge in Session 2 predicted unique variance in reading fluency (p = .024, ηρ² = 0.083), independent of children's phoneme awareness and inattention symptoms. Conclusion: The present study shows that consolidation of procedural memory is impaired in DD, even when the initial learning session is prolonged. The findings encourage further research on the mechanisms supporting procedural memory consolidation in children with DD, and how these may be strengthened. (PsycInfo Database Record (c) 2021 APA, all rights reserved).

Alpha/beta power decreases during episodic memory formation predict the magnitude of alpha/beta power decreases during subsequent retrieval

Episodic memory retrieval is characterised by the vivid reinstatement of information about a personally-experienced event. Growing evidence suggests that this reinstatement is supported by reductions in the spectral power of alpha/beta activity. Given that the amount of information that can be recalled depends on the amount of information that was originally encoded, information-based accounts of alpha/beta activity would suggest that retrieval-related alpha/beta power decreases similarly depend upon decreases in alpha/beta power during encoding. To test this hypothesis, seventeen human participants completed a sequence-learning task while undergoing concurrent MEG recordings. Regression-based analyses were then used to estimate how alpha/beta power decreases during encoding predicted alpha/beta power decreases during retrieval on a trial-by-trial basis. When subjecting these parameter estimates to group-level analysis, we find evidence to suggest that retrieval-related alpha/beta (7–15Hz) power decreases fluctuate as a function of encoding-related alpha/beta power decreases. These results suggest that retrieval-related alpha/beta power decreases are contingent on the decrease in alpha/beta power that arose during encoding. Subsequent analysis uncovered no evidence to suggest that these alpha/beta power decreases reflect stimulus identity, indicating that the contingency between encoding- and retrieval-related alpha/beta power reflects the reinstatement of a neurophysiological operation, rather than neural representation, during episodic memory retrieval.

Shifting uncertainty intolerance: methylphenidate and attention-deficit hyperactivity disorder

Risk evaluation is a critical component of decision making. Risk tolerance is relevant in both daily decisions and pathological disorders such as attention-deficit hyperactivity disorder (ADHD), where impulsivity is a cardinal symptom. Methylphenidate, a commonly prescribed drug in ADHD, improves attention but has mixed reports on risk-based decision making. Using a double-blinded placebo protocol, we studied the risk attitudes of ADHD patients and age-matched healthy volunteers while performing the 2-step sequential learning task and examined the effect of methylphenidate on their choices. We then applied a novel computational analysis using the hierarchical drift–diffusion model to extract parameters such as threshold (‘a’—amount of evidence accumulated before making a decision), drift rate (‘v’—information processing speed) and response bias (‘z’ apriori bias towards a specific choice) focusing specifically on risky choice preference. Critically, we show that ADHD patients on placebo have an apriori bias towards risky choices compared to controls. Furthermore, methylphenidate enhanced preference towards risky choices (higher apriori bias) in both groups but had a significantly greater effect in the patient population independent of clinical scores. Thus, methylphenidate appears to shift tolerance towards risky uncertain choices possibly mediated by prefrontal dopaminergic and noradrenergic modulation. We emphasise the utility of computational models in detecting underlying processes. Our findings have implications for subtle yet differential effects of methylphenidate on ADHD compared to healthy population.

IMPULSE: A 65-nm Digital Compute-in-Memory Macro With Fused Weights and Membrane Potential for Spike-Based Sequential Learning Tasks

The inherent dynamics of the neuron membrane potential in spiking neural networks (SNNs) allows the processing of sequential learning tasks, avoiding the complexity of recurrent neural networks. The highly sparse spike-based computations in such spatiotemporal data can be leveraged for energy efficiency. However, the membrane potential incurs additional memory access bottlenecks in current SNN hardware. To that effect, we propose a 10T-SRAM compute-in-memory (CIM) macro, specifically designed for state-of-the-art SNN inference. It consists of a fused weight ( W <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> MEM</sub> ) and membrane potential ( V <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> MEM</sub> ) memory and inherently exploits sparsity in input spikes leading to ~97.4% reduction in energy-delay product (EDP) at 85% sparsity (typical of SNNs considered in this work) compared to the case of no sparsity. We propose staggered data mapping and reconfigurable peripherals for handling different bit precision requirements of W <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> MEM</sub> and V <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> MEM</sub> , while supporting multiple neuron functionalities. The proposed macro was fabricated in 65-nm CMOS technology, achieving energy efficiency of 0.99 TOPS/W at 0.85-V supply and 200-MHz frequency for signed 11-bit operations. We evaluate the SNN for sentiment classification from the IMDB dataset of movie reviews and achieve within ~1% accuracy difference and ~ 5× higher energy efficiency compared to a corresponding long short-term memory network.

Analog Signal Processing Using Stochastic Magnets

We present a low energy-barrier magnet based compact hardware unit for analog stochastic neurons (ASNs) and demonstrate its use as a building-block for neuromorphic hardware. Networks assembled from these units are particularly suited for temporal inferencing and pattern recognition. We demonstrate example applications of these ASNs including multi-layer perceptrons, convolutional neurons, and reservoir computers showing tasks such as temporal sequence learning, processing, and prediction tasks which prove that these units can be used to build efficient, scalable, and adaptive neural network based signal-processors. We also provide an illustrative comparison with digital CMOS based circuits that implement similar functionality with networks built using the presented units, demonstrating a possible two orders of magnitude reduction in component-count and concomitant increase in energy efficiency. Efficient non von-Neumann hardware implementation of such signal-processors can open up a pathway for integration of hardware based cognition in a wide variety of emerging systems such as IoT, industrial controls, bio- and photo-sensors, self-driving automotives, and unmanned aerial vehicles.

Retrieval of a well-established skill is resistant to distraction: Evidence from an implicit probabilistic sequence learning task.

The characteristics of acquiring new sequence information under dual-task situations have been extensively studied. A concurrent task has often been found to affect performance. In real life, however, we mostly perform a secondary task when the primary task is already well acquired. The effect of a secondary task on the ability to retrieve well-established sequence representations remains elusive. The present study investigates whether accessing well-acquired probabilistic sequence knowledge is affected by a concurrent task. Participants acquired non-adjacent regularities in an implicit probabilistic sequence learning task. After a 24-hour offline period, participants were tested on the same probabilistic sequence learning task under dual-task or single-task conditions. Here, we show that although the secondary task significantly prolonged the overall reaction times in the primary (sequence learning) task, access to the previously learned probabilistic representations remained intact. Our results highlight the importance of studying the dual-task effect not only in the learning phase but also during memory access to reveal the robustness of the acquired skill.

Predicting Financial Prices of Stock Market using Recurrent Convolutional Neural Networks

Financial time-series prediction has been long and the most challenging issues in financial market analysis.The deep neural networks is one of the excellent data mining approach has received great attention by researchers in several areas of time-series prediction since last 10 years. Convolutional neural network (CNN) and recurrent neural network (RNN) models have become the mainstream methods for financial predictions. In this paper, we proposed to combine architectures, which exploit the advantages of CNN and RNN simultaneously, for the prediction of trading signals. Our model is essentially presented to financial time series predicting signals through a CNN layer, and directly fed into a gated recurrent unit (GRU) layer to capture long-term signals dependencies. GRU model perform better in sequential learning tasks and solve the vanishing gradients and exploding issue in standard RNNs. We evaluate our model on three datasets for stock indexes of the Hang Seng Indexes (HSI), the Deutscher Aktienindex (DAX) and the S&amp;P 500 Index range 2008 to 2016, and associate the GRU-CNN based approaches with the existing deep learning models. Experimental results present that the proposed GRU-CNN model obtained the best prediction accuracy 56.2% on HIS dataset, 56.1% on DAX dataset and 56.3% on S&amp;P500 dataset respectively.

Divided attention does not affect the acquisition and consolidation of transitional probabilities

Statistical learning facilitates the efficient processing and prediction of environmental events and contributes to the acquisition of automatic behaviors. Whereas a minimal level of attention seems to be required for learning to occur, it is still unclear how acquisition and consolidation of statistical knowledge are affected when attention is divided during learning. To test the effect of divided attention on statistical learning and consolidation, ninety-six healthy young adults performed the Alternating Serial Reaction Time task in which they incidentally acquired second-order transitional probabilities. Half of the participants completed the task with a concurrent secondary intentional sequence learning task that was applied to the same stimulus stream. The other half of the participants performed the task without any attention manipulation. Performance was retested after a 12-h post-learning offline period. Half of each group slept during the delay, while the other half had normal daily activity, enabling us to test the effect of delay activity (sleep vs. wake) on the consolidation of statistical knowledge. Divided attention had no effect on statistical learning: The acquisition of second-order transitional probabilities was comparable with and without the secondary task. Consolidation was neither affected by divided attention: Statistical knowledge was similarly retained over the 12-h delay, irrespective of the delay activity. Our findings can contribute to a better understanding of the role of attentional processes in and the robustness of visuomotor statistical learning and consolidation.

A single, clinically relevant dose of the GABAB agonist baclofen impairs visuomotor learning.

The GABAB agonist baclofen is taken daily as a treatment for spasticity by millions of stroke, brain injury and multiple sclerosis patients, many of whom are also undergoing motor rehabilitation. However, decreases in GABA are suggested to be a key feature of human motor learning, which raises questions about whether drugs increasing GABAergic activity may impair motor learning and rehabilitation. In this double-blind, placebo-controlled study, we investigated whether a single 10 mg dose of the GABAB agonist baclofen impaired motor sequence learning and visuomotor learning in 20 young healthy participants of both sexes. Participants trained on visuomotor and sequence learning tasks using their right hand. Transcranial magnetic stimulation (TMS) measures of corticospinal excitability, GABAA (short-interval intracortical inhibition, 2.5 ms) and GABAB (long-interval intracortical inhibition, 150 ms) receptor activation were recorded from left M1. Behaviourally, baclofen caused a significant reduction of visuomotor aftereffect (F 1,137.8 = 6.133, P = 0.014) and retention (F 1,130.7 = 4.138, P = 0.044), with no significant changes to sequence learning. There were no overall changes to TMS measured GABAergic inhibition with this low dose of baclofen. This result confirms the causal importance of GABAB inhibition in mediating visuomotor learning and suggests that chronic baclofen use could negatively impact aspects of motor rehabilitation.

Medial frontal negativities predict performance improvements during motor sequence but not motor adaptation learning.

Alterations in our environment require us to learn or alter motor skills to remain efficient. Also, damage or injury may require the relearning of motor skills. Two types have been identified: movement adaptation and motor sequence learning. Doyonet al. (2003, Distinct contribution of the cortico‐striatal and cortico‐cerebellar systems to motor skill learning. Neuropsychologia, 41(3), 252‐262) proposed a model to explain the neural mechanisms related to adaptation (cortico‐cerebellar) and motor sequence learning (cortico‐striatum) tasks. We hypothesized that medial frontal negativities (MFNs), event‐related electrocortical responses including the error‐related negativity (ERN) and correct‐response‐related negativity (CRN), would be trait biomarkers for skill in motor sequence learning due to their relationship with striatal neural generators in a network involving the anterior cingulate and possibly the supplementary motor area. We examined 36 participants' improvement in a motor adaptation and a motor sequence learning task and measured MFNs elicited in a separate Spatial Stroop (conflict) task. We found both ERN and CRN strongly predicted performance improvement in the sequential motor task but not in the adaptation task, supporting this aspect of the Doyon model. Interestingly, the CRN accounted for additional unique variance over the variance shared with the ERN suggesting an expansion of the model.

Stress Modulates the Balance between Hippocampal and Motor Networks during Motor Memory Processing.

The functional interaction between hippocampo- and striato-cortical regions during motor sequence learning is essential to trigger optimal memory consolidation. Based on previous evidence from other memory domains that stress alters the balance between these systems, we investigated whether exposure to stress prior to motor learning modulates motor memory processes. Seventy-two healthy young individuals were exposed to a stressful or nonstressful control intervention prior to training on a motor sequence learning task in a magnetic resonance imaging (MRI) scanner. Consolidation was assessed with an MRI retest after a sleep episode. Behavioral results indicate that stress prior to learning did not influence motor performance. At the neural level, stress induced both a larger recruitment of sensorimotor regions and a greater disengagement of hippocampo-cortical networks during training. Brain-behavior regression analyses showed that while this stress-induced shift from (hippocampo-)fronto-parietal to motor networks was beneficial for initial performance, it was detrimental for consolidation. Our results provide the first experimental evidence that stress modulates the neural networks recruited during motor memory processing and therefore effectively unify concepts and mechanisms from diverse memory fields. Critically, our findings suggest that intersubject variability in brain responses to stress determines the impact of stress on motor learning and subsequent consolidation.

Learning to select actions shapes recurrent dynamics in the corticostriatal system

Learning to select appropriate actions based on their values is fundamental to adaptive behavior. This form of learning is supported by fronto-striatal systems. The dorsal-lateral prefrontal cortex (dlPFC) and the dorsal striatum (dSTR), which are strongly interconnected, are key nodes in this circuitry. Substantial experimental evidence, including neurophysiological recordings, have shown that neurons in these structures represent key aspects of learning. The computational mechanisms that shape the neurophysiological responses, however, are not clear. To examine this, we developed a recurrent neural network (RNN) model of the dlPFC-dSTR circuit and trained it on an oculomotor sequence learning task. We compared the activity generated by the model to activity recorded from monkey dlPFC and dSTR in the same task. This network consisted of a striatal component which encoded action values, and a prefrontal component which selected appropriate actions. After training, this system was able to autonomously represent and update action values and select actions, thus being able to closely approximate the representational structure in corticostriatal recordings. We found that learning to select the correct actions drove action-sequence representations further apart in activity space, both in the model and in the neural data. The model revealed that learning proceeds by increasing the distance between sequence-specific representations. This makes it more likely that the model will select the appropriate action sequence as learning develops. Our model thus supports the hypothesis that learning in networks drives the neural representations of actions further apart, increasing the probability that the network generates correct actions as learning proceeds. Altogether, this study advances our understanding of how neural circuit dynamics are involved in neural computation, revealing how dynamics in the corticostriatal system support task learning.

Accurate deep neural network model to detect cardiac arrhythmia on more than 10,000 individual subject ECG records

Background and objectiveCardiac arrhythmia, which is an abnormal heart rhythm, is a common clinical problem in cardiology. Detection of arrhythmia on an extended duration electrocardiogram (ECG) is done based on initial algorithmic software screening, with final visual validation by cardiologists. It is a time consuming and subjective process. Therefore, fully automated computer-assisted detection systems with a high degree of accuracy have an essential role in this task. In this study, we proposed an effective deep neural network (DNN) model to detect different rhythm classes from a new ECG database. MethodsOur DNN model was designed for high performance on all ECG leads. The proposed model, which included both representation learning and sequence learning tasks, showed promising results on all 12-lead inputs. Convolutional layers and sub-sampling layers were used in the representation learning phase. The sequence learning part involved a long short-term memory (LSTM) unit after representation of learning layers. ResultsWe performed two different class scenarios, including reduced rhythms (seven rhythm types) and merged rhythms (four rhythm types) according to the records from the database. Our trained DNN model achieved 92.24% and 96.13% accuracies for the reduced and merged rhythm classes, respectively. ConclusionRecently, deep learning algorithms have been found to be useful because of their high performance. The main challenge is the scarcity of appropriate training and testing resources because model performance is dependent on the quality and quantity of case samples. In this study, we used a new public arrhythmia database comprising more than 10,000 records. We constructed an efficient DNN model for automated detection of arrhythmia using these records.

Integration of waking experience through dreams considered in light of individual differences in implicit learning ability.

The present study explored whether individual differences in implicit learning were related to the incorporation of waking events into dreams. Participants (N=60) took part in a sequence learning task, a measure of implicit learning ability. They were then asked to keep a record of their waking experiences (personally significant events [PSEs]/major concerns), as well as their nightly dreams for a week. Of these, the responses of 51 participants were suitable for further analysis in which participants themselves and three independent judges rated the correlation between waking events and dreams of the same day. Implicit learning ability was found to significantly correlate with the incorporation of PSEs into dreams. The present results may lend support to the Horton and Malinowski autobiographical memory (AM) model, which accounts for the activation of memories in dreams as a reflection of sleep-dependent memory consolidation processes that focusses in particular on the hyperassociative nature of AM during sleep.

CSLM: Convertible Short-Term and Long-Term Memory in Differential Neural Computers

External memory-based neural networks, such as differentiable neural computers (DNCs), have recently gained importance and popularity to solve complex sequential learning tasks that pose challenges to conventional neural networks. However, a trained DNC usually has a low-memory utilization efficiency. This article introduces a variation of DNC architecture with a convertible short-term and long-term memory, named CSLM-DNC. Unlike the memory architecture of the original DNC, the new scheme of short-term and long-term memories offers different importance of memory locations for read and write, and they can be converted over time. This is mainly motivated by the human brain where short-term memory stores large amounts of noisy and unimportant information and decays rapidly, while long-term memory stores important information and lasts for a long time. The conversion of these two types of memory is allowed and is able to be learned according to their reading and writing frequency. We quantitatively and qualitatively evaluate the proposed CSLM-DNC architecture on the tasks of question answering, copy and repeat copy, showing that it can significantly improve memory efficiency and learning performance.