Subsequent Task Research Articles

Enhancing Human Activity Recognition in Smart Homes with Self-Supervised Learning and Self-Attention.

Deep learning models have gained prominence in human activity recognition using ambient sensors, particularly for telemonitoring older adults' daily activities in real-world scenarios. However, collecting large volumes of annotated sensor data presents a formidable challenge, given the time-consuming and costly nature of traditional manual annotation methods, especially for extensive projects. In response to this challenge, we propose a novel AttCLHAR model rooted in the self-supervised learning framework SimCLR and augmented with a self-attention mechanism. This model is designed for human activity recognition utilizing ambient sensor data, tailored explicitly for scenarios with limited or no annotations. AttCLHAR encompasses unsupervised pre-training and fine-tuning phases, sharing a common encoder module with two convolutional layers and a long short-term memory (LSTM) layer. The output is further connected to a self-attention layer, allowing the model to selectively focus on different input sequence segments. The incorporation of sharpness-aware minimization (SAM) aims to enhance model generalization by penalizing loss sharpness. The pre-training phase focuses on learning representative features from abundant unlabeled data, capturing both spatial and temporal dependencies in the sensor data. It facilitates the extraction of informative features for subsequent fine-tuning tasks. We extensively evaluated the AttCLHAR model using three CASAS smart home datasets (Aruba-1, Aruba-2, and Milan). We compared its performance against the SimCLR framework, SimCLR with SAM, and SimCLR with the self-attention layer. The experimental results demonstrate the superior performance of our approach, especially in semi-supervised and transfer learning scenarios. It outperforms existing models, marking a significant advancement in using self-supervised learning to extract valuable insights from unlabeled ambient sensor data in real-world environments.

The defect feature extraction of ultrasonic phased array detection based on adaptive region growth.

An ultrasonic phased array defect extraction method based on adaptive region growth is proposed, aiming at problems such as difficulty in defect identification and extraction caused by noise interference and complex structure of the detected object during ultrasonic phased array detection. First, bilateral filtering and grayscale processing techniques are employed for the purpose of noise reduction and initial data processing. Following this, the maximum sound pressure within the designated focusing region serves as the seed point. An adaptive region iteration method is subsequently employed to execute automatic threshold capture and region growth. In addition, mathematical morphology is applied to extract the processed defect features. In the final stage, two sets of B-scan images depicting hole defects of varying sizes are utilized for experimental validation of the proposed algorithm's effectiveness and applicability. The defect features extracted through this algorithm are then compared and analyzed alongside the histogram threshold method, Otsu method, K-means clustering algorithm, and a modified iterative method. The results reveal that the margin of error between the measured results and the actual defect sizes is less than 13%, representing a significant enhancement in the precision of defect feature extraction. Consequently, this method establishes a dependable foundation of data for subsequent tasks, such as defect localization and quantitative and qualitative analysis.

Breaking boredom: Interrupting the residual effect of state boredom on future productivity.

Boredom is an emotion that constantly fluctuates in employees of all ages and occupations. Here, we draw on functional theories of boredom and theories of emotion regulation to develop an episodic model of how boredom shapes employee attention and productivity over time. We argue that employees often suppress boredom at work to "power through" boring tasks and objectives, resulting in residual bouts of mind-wandering-and thus productivity deficits-during future performance episodes. However, following boredom on an initial task, the commencement of a subsequent task that employees perceive to be meaningful creates an attentional pull that breaks the link between boredom and future mind-wandering, preventing the effects of boredom from spilling over to inhibit future productivity. Study 1 draws on archival experience sampling data to test our hypotheses and examine whether boredom exhibits reciprocal relationships with mind-wandering and productivity over time. Study 2 uses an experimental design to determine whether boredom and task meaningfulness interact to exert a causal effect on future mind-wandering. Study 3 uses a time-separated single-day design to replicate Studies 1 and 2 and examine our contention that employees often suppress boredom at work which, rather than preventing the effects of boredom, puts them "on hold" until a later point in time. Our findings provide insight into how to mitigate the far-reaching effects of boredom at work; they also advance episodic accounts of emotions, attention, and performance in organizations. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Legal Judgment Prediction via graph boosting with constraints

Legal Judgment Prediction (LJP) is a multi-task multi-label problem in the civil law system, involving the prediction of law articles, charges, and terms of penalty based on fact descriptions. However, most existing research approaches LJP as a single-label scenario, neglecting the correlations between multiple labels and failing to consider cross-task consistency constraints in a multi-label scenario. Moreover, although previous multi-task studies have proposed expert models and coarse-grained topology construction for inter-task relationships, the former neglects rich information exchange among different tasks, and the latter, if one task’s prediction is inaccurate, will affect subsequent tasks. This paper has designed legal label graphs and proposed a novel graph boosting with constraints framework, GJudge, for legal judgment prediction to address these limitations. The framework comprises a multi-perspective interactive encoder and a multi-graph attention consistency expert module. The encoder utilizes bidirectional LSTM, gated attention units, cross attention, and graph attention networks to integrate fact descriptions and label similarity relationships information from legal label graphs for multi-perspective interactive encoding. The expert module utilizes the multiple expert networks and the multi-graph attention network to differentiate between confusing labels and ensure consistent constraints across tasks, this is achieved through the fusion of label consistency constraints and confusion relationships information in the legal label graphs. Experimental results on two real-world datasets across different tasks show an improvement in F1 scores ranging from at least 0.93% to a maximum of 2.97%, illustrating the effectiveness of GJudge compared to the state-of-the-art model.

Research on image classification of power inspection using less sample learning technique

Abstract This study proposes a few-sample self-supervised power inspection image classification method based on Vision Transformer and masked autoencoders (MAE). Initially, an MAE is employed to pretrain an extensive collection of unlabeled power inspection images, leveraging the masking technique to encourage the model to acquire an understanding of the overarching features within these images. Then, the pretrained encoder part is combined with the visual transformer to further extract image features. For the subsequent classification task, the encoder’s pretrained weights are frozen and subsequently incorporated into a novel classification network. By fine-tuning this network with a limited amount of labeled data, experimental findings indicate a substantial enhancement in the classification accuracy of power inspection images, particularly in scenarios with limited sample sizes. By introducing visual transformers and MAE, this article provides an efficient and reliable solution for intelligent classification of power inspection images, which has important practical application value and broad market prospects.

Learning reservoir dynamics with temporal self-modulation

Reservoir computing (RC) can efficiently process time-series data by mapping the input signal into a high-dimensional space via randomly connected recurrent neural networks (RNNs), which are referred to as a reservoir. The high-dimensional representation of time-series data in the reservoir simplifies subsequent learning tasks. Although this simple architecture allows fast learning and facile physical implementation, the learning performance is inferior to that of other state-of-the-art RNN models. In this study, to improve the learning ability of RC, we propose self-modulated RC (SM-RC) that extends RC by adding a self-modulation mechanism. SM-RC can perform attention tasks where input information is retained or discarded depending on the input signal. We find that a chaotic state can emerge as a result of learning in SM-RC. Furthermore, we demonstrate that SM-RC outperforms RC in NARMA and Lorenz model tasks. Because the SM-RC architecture only requires two additional gates, it is physically implementable as RC, thereby providing a direction for realizing edge artificial intelligence.

The Reactivation of working memory representations affects attentional guidance.

Recent studies have suggested that the neural activity that supported working memory (WM) storage is dynamic over time and this dynamic storage decides memory performance. Does the temporal dynamic of the WM representation also affect visual search, and how does it interact with distractor suppression over time? To address these issues, we tracked the time course of the reactivation of WM representations during visual search by analyzing the electroencephalogram (EEG) and event-related optical signals (EROS) in Experiments 1 and 2, respectively, and investigated the interaction between the representation reactivation and distractor suppression in Experiment 3. Participants had to maintain a color in WM under high- or low-precision requirement and perform a subsequent search task. The reactivation of WM representations was defined by the above-chance decoding accuracy. The EEG results showed that compared with the low-precision requirement, WM-matching distractors captured more attention and the WM representation were reactivated more frequently under high-precision requirement. The EROS results showed that compared with the low-precision requirement, the increased activity in occipital cortex in the WM-matching versus WM-mismatching conditions was observed at 224 ms during visual search under high-precision requirement. Regression analysis showed that the representation reactivation during visual search directly predicted the behavioral WM-based attentional capture effect, while the representation reactivation before visual search impacted the WM-based attentional capture effect through the mediation of distractor suppression during visual search. These results suggest that the reactivation of WM representations and distractor suppression collectively determine WM-based attentional capture.

STBI-GAN: An adversarial learning approach for data synthesis on traumatic brain segmentation

Automatic brain segmentation of magnetic resonance images (MRIs) from severe traumatic brain injury (sTBI) patients is critical for brain abnormality assessments and brain network analysis. Construction of sTBI brain segmentation model requires manually annotated MR scans of sTBI patients, which becomes a challenging problem as it is quite impractical to implement sufficient annotations for sTBI images with large deformations and lesion erosion. Data augmentation techniques can be applied to alleviate the issue of limited training samples. However, conventional data augmentation strategies such as spatial and intensity transformation are unable to synthesize the deformation and lesions in traumatic brains, which limits the performance of the subsequent segmentation task. To address these issues, we propose a novel medical image inpainting model named sTBI-GAN to synthesize labeled sTBI MR scans by adversarial inpainting. The main strength of our sTBI-GAN method is that it can generate sTBI images and corresponding labels simultaneously, which has not been achieved in previous inpainting methods for medical images. We first generate the inpainted image under the guidance of edge information following a coarse-to-fine manner, and then the synthesized MR image is used as the prior for label inpainting. Furthermore, we introduce a registration-based template augmentation pipeline to increase the diversity of the synthesized image pairs and enhance the capacity of data augmentation. Experimental results show that the proposed sTBI-GAN method can synthesize high-quality labeled sTBI images, which greatly improves the 2D and 3D traumatic brain segmentation performance compared with the alternatives. Code is available at .

Multi-scale dense spatially-adaptive residual distillation network for lightweight underwater image super-resolution

Underwater images are typically of poor quality, lacking texture and edge information, and are blurry and full of artifacts, which restricts the performance of subsequent tasks such as underwater object detection, and path planning for underwater unmanned submersible vehicles (UUVs). Additionally, the limitation of underwater equipment, most existing image enhancement and super-resolution methods cannot be implemented directly. Hence, developing a weightless technique for improving the resolution of submerged images while balancing performance and parameters is vital. In this paper, a multi-scale dense spatially-adaptive residual distillation network (MDSRDN) is proposed aiming at obtaining high-quality (HR) underwater images with odd parameters and fast running time. In particular, a multi-scale dense spatially-adaptive residual distillation module (MDSRD) is developed to facilitate the multi-scale global-to-local feature extraction like a multi-head transformer and enriching spatial attention maps. By introducing a spatial feature transformer layer (SFT layer) and residual spatial-adaptive feature attention (RSFA), an enhancing attention map for spatially-adaptive feature modulation is generated. Furthermore, to maintain the network lightweight enough, blue separable convolution (BS-Conv) and distillation module are applied. Extensive experimental results illustrate the superiority of MDSDRN in underwater image super-resolution reconstruction, which can achieve a great balance between parameters (only 0.32M), multi-adds (only 13G), and performance (26.38 dB on PSNR in USR-248) with the scale of ×4.

From Noise Addition to Denoising: A Self-Variation Capture Network for Point Cloud Optimization.

Point clouds obtained from 3D scanners are often noisy and cannot be directly used for subsequent high-level tasks. In this article, we propose a novel point cloud optimization method capable of denoising and homogenizing point clouds. Our idea is based on the assumption that the noise is generally much smaller than the effective signal. We perform noise perturbation on the noisy point cloud to get a new noisy point cloud, called self-variation point cloud. The noisy point cloud and self-variation point cloud have different noise distribution, but the same point cloud distribution. We compute the potential commonality between two noisy point clouds to obtain a clean point cloud. To implement our idea, we propose a Self-Variation Capture Network (SVCNet). We perturb the point cloud features in the latent space to obtain self-variation feature vectors, and capture the commonality between two noisy feature vectors through the feature aggregation and averaging. In addition, an edge constraint module is introduced to suppress low-pass effects during denoising. Our denoising method does not take into account the noise characteristics, and can filter the drift noise located on the underlying surface, resulting in a uniform distribution of the generated point cloud. The experimental results show that our algorithm outperforms the current state-of-the-art algorithms, especially in generating more uniform point clouds. In addition, extended experiments demonstrate the potential of our algorithm for point clouds upsampling.

The effect of different interfaces during virtual game practice on motor performance of individuals with genetic ataxia: A cross-sectional study.

Reaching and coordination tasks are widely used in traditional physical rehabilitation programs for individuals with Ataxia. Virtual reality interventions could optimize the motor performance of these individuals; however, the type of virtual interface may influence performance during virtual practice. We aimed to estimate the extent of the effect of different interfaces (webcam and touchscreen) on the motor performance of individuals with various types of genetic ataxia, compared to a control group, during virtual computer game tasks. Repeated exposure quasi-experimental design, which included seventeen volunteers diagnosed with progressive ataxia between 21 and 64 years of age and sixteen age-matched controls. The virtual game tasks were based on the MoveHero software, performed using different interfaces (webcam or touchscreen). Subgroups of participants with genetic ataxia performed the virtual games using the interfaces in different orders (webcam interface followed by touchscreen interface, or vice-versa). The absolute error (AE), variable error (VE), number of hits, and anticipation were used to reflect the motor performance during the virtual task. Participants with ataxia presented more variable and absolute errors, a lower number of hits, and greater anticipation error than controls (p<0.05). For participants with ataxia, a greater AE was found only in the sequence touchscreen followed by webcam interface (i.e., the sequence webcam before touchscreen presented lower AE). The group of participants with genetic ataxia presented lower performance than the control group regardless of the interface (webcam or touchscreen). The most interesting observation was that although practicing with the webcam interface offers features that make the task more complex than the touchscreen interface, resulting in lower performance, this interface facilitated performance in a subsequent touchscreen task only in individuals with ataxia, suggesting that a virtual interface engenders greater transfer to other tasks. Registered at Registro Brasileiro de Ensaios Clínicos (ReBEC) database number identifier: RBR-3q685r5.

Does culture moderate the encoding and recognition of negative cues? Evidence from an eye-tracking study.

Cross-cultural research has elucidated many important differences between people from Western European and East Asian cultural backgrounds regarding how each group encodes and consolidates the contents of complex visual stimuli. While Western European groups typically demonstrate a perceptual bias towards centralised information, East Asian groups favour a perceptual bias towards background information. However, this research has largely focused on the perception of neutral cues and thus questions remain regarding cultural group differences in both the perception and recognition of negative, emotionally significant cues. The present study therefore compared Western European (n = 42) and East Asian (n = 40) participants on a free-viewing task and a subsequent memory task utilising negative and neutral social cues. Attentional deployment to the centralised versus background components of negative and neutral social cues was indexed via eye-tracking, and memory was assessed with a cued-recognition task two days later. While both groups demonstrated an attentional bias towards the centralised components of the neutral cues, only the Western European group demonstrated this bias in the case of the negative cues. There were no significant differences observed between Western European and East Asian groups in terms of memory accuracy, although the Western European group was unexpectedly less sensitive to the centralised components of the negative cues. These findings suggest that culture modulates low-level attentional deployment to negative information, however not higher-level recognition after a temporal interval. This paper is, to our knowledge, the first to concurrently consider the effect of culture on both attentional outcomes and memory for both negative and neutral cues.

Multi-Level Object-Aware Guidance Network for Biomedical Image Segmentation

Most state-of-the-art models for biomedical image segmentation are developed based on U-shape architecture, which has two renowned, yet mutually affected, shortcomings: 1) difficulties in capturing global long-range dependencies, and 2) semantic information dilution in the decoding process. In this paper, we propose a novel network with a new object-aware module (OAM) to effectively establish global dependencies at multiple levels within the network and compensate high-level semantic information dilution when fusing the extracted multi-level features; we call the network MOG-Net. Specifically, the OAM is designed to figure out the relations between each pixel and targeting object region and recalibrate class-level semantic information according to the relations. Compared with non-local models, which construct pixel-wise global dependencies, our OAM is more efficient and target-specific, enabling us to achieve satisfactory results with less extra computational overhead. In addition, we embed a pyramid context encoder module (PCEM) in the proposed OAM to alleviate semantic information dilution; this scheme is able to bridge the spatial-semantic gap when fusing features extracted from different levels. We extensively evaluate the proposed MOG-Net on four diverse biomedical image segmentation tasks with different imaging modalities, achieving segmentation performance with 88.19%, 90.95% and 66.03% in Dice on three one-class datasets, as well as 88.83% and 87.11% in Dice for two classes on a multi-class dataset, respectively. Experimental results demonstrate the effectiveness of the proposed method, consistently outperforming state-of-the-art methods in most evaluation metrics. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —Semantic segmentation of biomedical images is a critical prerequisite for subsequent diagnosis, treatment, and quantitative tasks in clinical practice. This article proposes a novel biomedical image segmentation network, namely MOG-Net, with a new object-aware module (OAM) to model global context dependencies from a category perspective and a pyramid context encoder module (PCEM) to enhance feature representation capabilities of spatial and channel dimensions. We experimentally demonstrate the effectiveness and generalization capability of proposed MOG-Net on diverse biomedical image segmentation tasks with different imaging modalities. We believe that our proposed method can serve as a practical clinical tool and has the potential to be applied to existing computer-aided medical systems and clinical measurement.

Willed Attentional Selection of Visual Features: An EEG Study.

Visual selective attention studies generally tend to apply cuing paradigms to instructively direct observers' attention to certain locations, features or objects. However, in real situations, attention in humans often flows spontaneously without any specific instructions. Recently, a concept named "willed attention" was raised in visuospatial attention, in which participants are free to make volitional attention decisions. Several ERP components during willed attention were found, along with a perspective that ongoing alpha activity may bias the subsequent attentional choice. However, it remains unclear whether similar neural mechanisms exist in feature- or object-based willed attention. Here, we included choice cues and instruct cues in a feature-based selective attention paradigm, allowing participants to freely choose or to be instructed to attend a color for the subsequent target detection task. Pre-cue ongoing alpha oscillations, cue-evoked potentials and target-related steady-state visual evoked potentials (SSVEPs) were simultaneously measured as markers of attentional processing. As expected, SSVEP responses were similarly modulated by attention between choice and instruct cue trials. Similar to the case of spatial attention, a willed-attention component (Willed Attention Component, WAC) was isolated during the cue-related choice period by comparing choice and instruct cues. However, pre-cue ongoing alpha oscillations did not predict the color choice (yellow vs blue), as indicated by the chance level decoding accuracy (50%). Overall, our results revealed both similarities and differences between spatial and feature-based willed attention, and thus extended the understanding toward the neural mechanisms of volitional attention.

The impact of electronic despatch advice on the service time of a means of transport in a distribution center

Purpose: The aim of the article is to verify the impact of the use of electronic notification of deliveries on the service time of a means of transport in a distribution center. Methodology: The study used electronic notification in the form of EDI DESADV messages. The research was carried out in one of the distribution centers of a retail company operating in Poland. The study used the method of recording the time of logistics operations. The study was conducted in two stages (each lasting a week), in which the time of logistics operations was recorded in conditions without and with the DESADV message implemented. Results: The results of the conducted research show that the use of electronic despatch advice allows you to reduce the time of handling the means of transport in the distribution center. Conclusion: Reducing the time of handling the means of transport in the distribution center brings numerous additional benefits, such as faster release of the means of transport and enabling it to carry out subsequent transport tasks, handling a larger number of unloadings with the same number of resources (e.g. warehouse ramps), accelerating the internal process and enabling faster reloading to other means of transport performing the distribution process.

A Fusion Dimension Reduction Method for the Features of Surface Electromyographic Signals.

Surface electromyographic signals (sEMG) usually have high-dimensional properties, and direct processing of these data consumes significant computational resources. Dimensionality reduction processing can reduce the dimension of the data and improve the real-time performance and response speed. This is especially important for application scenarios such as prosthetic control and rehabilitation training where rapid feedback is required. This paper proposes a feature fusion dimension reduction method for sEMG signals. This method is constructed based on the unique correlation between the features of sEMG. To test the performance of the new dimension reduction method, the sEMG signals from five leg movements were collected from eight subjects and the classification of the feature matrix before and after dimension reduction was tested by six classifiers. The results show that the feature matrix after fusion dimension reduction has excellent classification performance in the subsequent classification tasks. It produces up to 98.3% accuracy. And the highest comprehensive evaluation index can reach 0.9958. This paper also compares the new method with three commonly used dimensionality reduction methods. The results show that the performance of the new method is not only optimal but also extremely stable. Because its classification performance will not be lower than other dimensionality reduction methods due to the change of classifiers. This confirms that the new method has a higher utility value in sEMG signals processing compared to other dimension reduction methods.

Children's task persistence in first grade: The role of parent-child and teacher-child relationships

The present study investigated the extent to which the quality of both parent-child and teacher-child relationships uniquely and interactively relates to a child's task persistence in first grade. Furthermore, the study also examined the role of a child's task persistence in the quality of their relationships with their parents and teachers after controlling for the child's gender, temperament, academic performance, and parental education level. The study followed Lithuanian children from the beginning (T1, n = 337) to the end (T2, n = 341) of first grade, along with their parents (T1, n = 347; T2, n = 323) and first-grade teachers (T1, n = 24; T2, n = 25). At the start of first grade, the children completed reading, spelling, and math (i.e., addition and subtraction) performance tests. The parents and teachers filled out questionnaires at the beginning and end of these students’ first-grade academic year. The parents reported on the quality of their relationships with their children and on their children's temperament, while teachers reported on their relationship quality with each student and their task persistence. First, the results indicated that a conflictual relationship between the students and their teachers was related negatively to subsequent task persistence. Second, the moderation results indicated that if the teacher-child relationship was close, then a close parent-child relationship longitudinally was related positively to the children's task persistence. Third, the results indicated that a child's higher level of task persistence was associated longitudinally with a closer relationship with their first-grade teachers.

Leave It to Large Language Models! Correction and Planning with Memory Integration

As humans, we can naturally break down a task into individual steps in our daily lives and we are able to provide feedback or dynamically adjust the plan when encountering obstacles. Similarly, our aim is to facilitate agents in comprehending and carrying out natural language instructions in a more efficient and cost-effective manner. For example, in Vision–Language Navigation (VLN) tasks, the agent needs to understand instructions such as “go to the table by the fridge”. This understanding allows the agent to navigate to the table and infer that the destination is likely to be in the kitchen. The traditional VLN approach mainly involves training models using a large number of labeled datasets for task planning in unseen environments. However, manual labeling incurs a high cost for this approach. Considering that large language models (LLMs) already possess extensive commonsense knowledge during pre-training, some researchers have started using LLMs as decision modules in embodied tasks, although this approach shows the LLMs’ reasoning ability to plan a logical sequence of subtasks based on global information. However, executing subtasks often encounters issues, such as obstacles that hinder progress and alterations in the state of the target object. Even one mistake can cause the subsequent tasks to fail, which makes it challenging to complete the instructions through a single plan. Therefore, we propose a new approach—C (Correction) and P (Planning) with M (Memory) I (Integration)—that centered on an LLM for embodied tasks. In more detail, the auxiliary modules of the CPMI facilitate dynamic planning by the LLM-centric planner. These modules provide the agent with memory and generalized experience mechanisms to fully utilize the LLM capabilities, allowing it to improve its performance during execution. Finally, the experimental results on public datasets demonstrate that we achieve the best performance in the few-shot scenario, improving the efficiency of the successive task while increasing the success rate.

The relationship of aerobic fitness with verbal and spatial working memory: An ERP study.

Working memory (WM) plays an important role in daily life and is known to correlated with aerobic fitness. However, whether the relationship between aerobic fitness and WM is dependent on the stimulus modality or is associated with one or multiple subprocesses involved in WM remains unknown. Accordingly, this study utilized event-related potentials (ERPs) to comprehensively examine the encoding, preparation, and retrieval processes during verbal and spatial WM performance. Eighty-eight young adults aged 18-30years were recruited to participate in two laboratory visits on separate days. On day 1, aerobic fitness was assessed by maximum oxygen consumption (V˙O2max) during a treadmill-based graded exercise test. On day 2, participants completed verbal and spatial WM tasks while P2, contingent negative voltage (CNV), and P3 components of ERP were recorded during the encoding, preparatory, and retrieval stages of WM, respectively. Results of hierarchical regression analysis showed that V˙O2max was positively correlated with response accuracy during the high-demanding condition of spatial WM after controlling for age, sex, and self-reported physical activity. Additionally, a higher level of V˙O2max was associated with larger terminal CNV amplitude at the Cz electrode during the high-demanding condition of spatial WM. These findings suggest that aerobic fitness may have selective beneficial associations with the motor preparatory process and subsequent task performance requiring a greater amount of spatial information but not the encoding and retrieval stages nor the verbal modality of WM.

Imbalance Mitigation for Continual Learning via Knowledge Decoupling and Dual Enhanced Contrastive Learning.

Continual learning (CL) aims at studying how to learn new knowledge continuously from data streams without catastrophically forgetting the previous knowledge. One of the key problems is catastrophic forgetting, that is, the performance of the model on previous tasks declines significantly after learning the subsequent task. Several studies addressed it by replaying samples stored in the buffer when training new tasks. However, the data imbalance between old and new task samples results in two serious problems: information suppression and weak feature discriminability. The former refers to the information in the sufficient new task samples suppressing that in the old task samples, which is harmful to maintaining the knowledge since the biased output worsens the consistency of the same sample's output at different moments. The latter refers to the feature representation being biased to the new task, which lacks discrimination to distinguish both old and new tasks. To this end, we build an imbalance mitigation for CL (IMCL) framework that incorporates a decoupled knowledge distillation (DKD) approach and a dual enhanced contrastive learning (DECL) approach to tackle both problems. Specifically, the DKD approach alleviates the suppression of the new task on the old tasks by decoupling the model output probability during the replay stage, which better maintains the knowledge of old tasks. The DECL approach enhances both low-and high-level features and fuses the enhanced features to construct contrastive loss to effectively distinguish different tasks. Extensive experiments on three popular datasets show that our method achieves promising performance under task incremental learning (Task-IL), class incremental learning (Class-IL), and domain incremental learning (Domain-IL) settings.