Target Task Research Articles

Evolutionary Optimization with a Simplified Helper Task for High-Dimensional Expensive Multiobjective Problems

In recent years, surrogate-assisted evolutionary algorithms (SAEAs) have been sufficiently studied for tackling computationally expensive multiobjective optimization problems (EMOPs), as they can quickly estimate the qualities of solutions by using surrogate models to substitute for expensive evaluations. However, most existing SAEAs only show promising performance for solving EMOPs with no more than 10 dimensions, and become less efficient for tackling EMOPs with higher dimensionality. Thus, this article proposes a new SAEA with a simplified helper task for tackling high-dimensional EMOPs. In each generation, one simplified task will be generated artificially by using random dimension reduction on the target task (i.e., the target EMOPs). Then, two surrogate models are trained for the helper task and the target task, respectively. Based on the trained surrogate models, evolutionary multitasking optimization is run to solve these two tasks so that the experiences of solving the helper task can be transferred to speed up the convergence of tackling the target task. Moreover, an effective model management strategy is designed to select new promising samples for training the surrogate models. When compared to five competitive SAEAs on four well-known benchmark suites, the experiments validate the advantages of the proposed algorithm on most test cases.

Research of Options for Constructing Information Management Systems Based on Network Models of Queuing Systems

The use of information management systems (IMSs) for the management of technical facilities is currently one of the directions for further improvement and increase in the effectiveness of the use of technical facilities in solving their target tasks. The existing modern IMSs are a set of hardware and software tools designed for collecting, processing and storing information and management. In the presence of a large amount of information and contradictory factors affecting the quality of management, making informed and timely decisions in the management process is impossible without the use of IMSs. The IMSs currently being developed are, for the most part, specialized systems and are designed to solve specific tasks. In this regard, the development and design of IMSs should be carried out taking into account the relationship with the target indicators and features of the management facilities, the results of a comprehensive analysis of information about the IMS elements in the process of functioning, and structural and algorithmic parameters that affect performance indicators. The use of mathematical models for the study of options for the construction of an IMS is the basis for the design and development of devices and subsystems of an IMS. The IMS models currently being developed make it possible to conduct research for single-stage management processes with the presence of similar service facilities in the system. At the same time, modern technical facilities and control systems are complex complexes with cyclically repeating control processes of various types of means. As a rule, such complexes have a set of parallel operating devices (control channels) that provide control of different types of objects at various stages of information processing. In this case, the structure of the IMS must be represented as a multiphase multichannel technical system in which the process of simultaneous management of several objects of various types takes place. In this regard, the purpose of the article is to develop a mathematical model of an IMS with two phases of management and the presence of an arbitrary number of serviced different types of management facilities. The basis of the model is a multiphase CFR network model with a limited waiting time for an application in the service queue. The study on the model allows choosing an option for building an IMS, in particular, choosing the optimal number of control channels for various types of objects according to the criterion of optimality and restrictions on the cost and time of management. An algorithm for selecting an option for building an IMS has been developed, and an example of calculating the number of control channels for managing three types of objects is given.

Deep joint subdomain alignment for unsupervised domain adaptation

Domain adaptation (DA) aims to transfer knowledge of the labeled source domain to the unlabeled target domain. Current DA methods learn domain-invariant features by aligning source and target feature spaces by distribution discrepancy minimization or adversarial training. However, existing DA methods fail to consider the semantic-level alignment, which can lead to the distortion of semantic feature structures for the target tasks. In this paper, we propose a novel domain adaptation method, called Deep Joint Subdomain Alignment (DJSA), which jointly conducts intra-subdomain and inter-subdomain alignment for fine-grained classification of target tasks. DJSA obtains the pseudo labels of unlabeled target samples by measuring the similarity between the target samples and source category centers, and then DJSA calculates an intra-subdomain loss to reduce the difference within the same class and increase the difference between different classes in both the source domain and the target domain, and DJSA utilizes a Local Maximum Mean Discrepancy (LMMD) metric to measure the inter-subdomain discrepancy between the source and target domain, and minimizes the discrepancy to align different domains. Furthermore, extensive experimental results demonstrate the effectiveness and superiority of our proposed method in unsupervised domain adaptation.

Combining transfer learning and statistical measures to predict performance of composite materials with limited data

AbstractPredicting the performance of composite materials is crucial for their application in civil infrastructure, yet limited experimental data often hinder the development of accurate and generalizable models. This study introduces a deep neural network (DNN) approach that combines summarizing statistics (SS) and transfer learning (TL)—termed the SSTL‐DNN approach—to address data scarcity in modeling composite materials. The computational novelty lies in the SS method's ability to extract comprehensive information from limited datasets by converting complex constitutive laws into concise statistical representations, thereby enabling efficient and effective model training. Simultaneously, the TL method enhances computational efficiency by leveraging knowledge from related tasks with abundant data to improve learning in the target task with scarce data. This combination not only reduces dependency on large datasets but also significantly improves model generalization. The proposed SSTL‐DNN approach is validated through two case studies: fiber‐reinforced polymer confined concrete and engineered cementitious composites. In both case studies, the SSTL‐DNN model reduces the required dataset size by up to 75% and decreases the validation error by 39%, compared to traditional deep learning models. These results demonstrate that the SSTL‐DNN approach not only overcomes data scarcity but also provides accurate predictions and generalization to unseen data, offering a practical solution for modeling composite materials with limited data.

Advanced Transfer Learning Strategies for Improving Deep Learning Model Efficiency and Accuracy

Machine learning primarily focuses on supervised learning, which relies on labeled data, while unsupervised learning addresses data without labels. Reinforcement learning, on the other hand, develops agents that learn optimal actions based on past experiences. Transfer learning, distinctively, enhances learning in a target task by leveraging knowledge from related source tasks. In recent years, many machine learning models have remained single-task focused. This paper explores scenarios in which transfer learning can be applied effectively from a source domain to a target domain, employing feature selection, extraction, and construction techniques. These methods enable transformation into a new feature space by generating a refined subset of the original features.

Compensatory mechanisms amidst demyelinating disorders: insights into cognitive preservation

Abstract Demyelination disrupts the transmission of electrical signals in the brain and affects neurodevelopment in children with disorders such as multiple sclerosis and myelin oligodendrocyte glycoprotein-associated disorders. Although cognitive impairments are prevalent in these conditions, some children maintain cognitive function despite substantial structural injury. These findings raise an important question: in addition to the degenerative process, do compensatory neural mechanisms exist to mitigate the effects of myelin loss? We propose that a multi-dimensional approach integrating multiple neuroimaging modalities, including diffusion tensor imaging, magnetoencephalography and eye-tracking, is key to investigating this question. We examine the structural and functional connectivity of the default mode and executive control networks due to their significant roles in supporting higher-order cognitive processes. As cognitive proxies, we examine saccade reaction times and direction errors during an interleaved pro- (eye movement towards a target) and anti-saccade (eye movement away from a target) task. 28 typically developing children, 18 children with multiple sclerosis and 14 children with myelin oligodendrocyte glycoprotein-associated disorders between 5 and 18.9 years old were scanned at the Hospital for Sick Children. Tractography of diffusion MRI data examined structural connectivity. Intracellular and extracellular microstructural parameters were extracted using a white matter tract integrity model to provide specific inferences on myelin and axon structure. Magnetoencephalography scanning was conducted to examine functional connectivity. Within groups, participants had longer saccade reaction times and greater direction errors on the anti- versus pro-saccade task; there were no group differences on either task. Despite similar behavioural performance, children with demyelinating disorders had significant structural compromise and lower bilateral high gamma, higher left-hemisphere theta and higher right-hemisphere alpha synchrony relative to typically developing children. Children diagnosed with multiple sclerosis had greater structural compromise relative to children with myelin oligodendrocyte glycoprotein-associated disorders; there were no group differences in neural synchrony. For both patient groups, increased disease disability predicted greater structural compromise, which predicted longer saccade reaction times and greater direction errors on both tasks. Structural compromise also predicted increased functional connectivity, highlighting potential adaptive functional reorganisation in response to structural compromise. In turn, increased functional connectivity predicted faster saccade reaction times and fewer direction errors. These findings suggest that increased functional connectivity, indicated by increased alpha and theta synchrony, may be necessary to compensate for structural compromise and preserve cognitive abilities. Further understanding these compensatory neural mechanisms could pave the way for the development of targeted therapeutic interventions aimed at enhancing these mechanisms, ultimately improving cognitive outcomes for affected individuals.

FedATA: Adaptive attention aggregation for federated self-supervised medical image segmentation

Pre-trained on large-scale datasets has profoundly promoted the development of deep learning models in medical image analysis. For medical image segmentation, collecting a large number of labeled volumetric medical images from multiple institutions is an enormous challenge due to privacy concerns. Self-supervised learning with mask image modeling (MIM) can learn general representation without annotations. Integrating MIM into FL enables collaborative learning of an efficient pre-trained model from unlabeled data, followed by fine-tuning with limited annotations. However, setting pixels as reconstruction targets in traditional MIM fails to facilitate robust representation learning due to the medical image's complexity and distinct characteristics. On the other hand, the generalization of the aggregated model in FL is also impaired under the heterogeneous data distributions among institutions. To address these issues, we proposed a novel self-supervised federated learning, which combines masked self-distillation with adaptive attention federated learning. Such incorporation enjoys two vital benefits. First, masked self-distillation sets high-quality latent representations of masked tokens as the target, improving the descriptive capability of the learned presentation rather than reconstructing low-level pixels. Second, adaptive attention aggregation with Personalized federate learning effectively captures specific-related representation from the aggregated model, thus facilitating local fine-tuning performance for target tasks. We conducted comprehensive experiments on two medical segmentation tasks using a large-scale dataset consisting of volumetric medical images from multiple institutions, demonstrating superior performance compared to existing federated self-supervised learning approaches.

Technologies for Increasing the Control Efficiency of Small Spacecraft with Solar Panels by Taking into Account Temperature Shock

The problem of the effective control of a small spacecraft is very relevant for solving a number of target tasks. Such tasks include, for example, remote sensing of the Earth or the implementation of gravity-sensitive processes. Therefore, it is necessary to develop new technologies for controlling small spacecraft. These technologies must take into account a number of disturbing factors that have not been taken into account previously. Temperature shock is one such factor for small spacecraft with solar panels. Therefore, the goal of the work is to create a new technology for controlling a small spacecraft based on a mathematical model of the stressed/deformed state of a solar panel during a temperature shock. The main methods for solving the problem are mathematical methods for solving initial/boundary value problems, in particular, the initial/boundary value problem of the third kind. As a result, an approximate solution for the deformation of a solar panel during a temperature shock was obtained. This solution is more general than those obtained previously. In particular, it satisfies the symmetrical condition of the solar panel. This could not be achieved by the previous solutions. We also observe an improvement (as compared to the previous solutions) in the fulfillment of the boundary conditions for the whole duration of the temperature shock. Based on this, a new technology for controlling a small spacecraft was created and its effectiveness was demonstrated. Application of the developed technology will improve the performance of the target tasks such as remote sensing of the Earth or the implementation of gravity-sensitive processes.

A lightweight model design approach for few-shot malicious traffic classification

Classifying malicious traffic, which can trace the lineage of attackers’ malicious families, is fundamental to safeguarding cybersecurity. However, the deep learning approaches currently employed require substantial volumes of data, conflicting with the challenges in acquiring and accurately labeling malicious traffic data. Additionally, edge network devices vulnerable to cyber-attacks often cannot meet the computational demands required to deploy deep learning models. The rapid mutation of malicious activities further underscores the need for models with strong generalization capabilities to adapt to evolving threats. This paper introduces an innovative few-shot malicious traffic classification method that is precise, lightweight, and exhibits enhanced generalization. By refining traditional transfer learning, the source model is segmented into public and private feature extractors for stepwise transfer, enhancing parameter alignment with specific target tasks. Neuron importance is then sorted based on the task of each feature extractor, enabling precise pruning to create an optimal lightweight model. An adversarial network guiding principle is adopted for retraining the public feature extractor parameters, thus strengthening the model’s generalization power. This method achieves an accuracy of over 97% on few-shot datasets with no more than 15 samples per class, has fewer than 50 K model parameters, and exhibits superior generalization compared to baseline methods.

Effects of Constraining Postural Sway During Upper-Limb Precision Aiming Task Practice in Individuals with Stroke

Individuals post-stroke commonly demonstrate alterations in motor behavior with regard to both task performance and the motor strategies used in pursuit of task goals. We evaluated whether constraining postural sway (motor strategy) during practice would affect upper-limb precision aiming performance (task performance) and postural control adaptations. Adults with stroke stood on a force plate while immersed in a virtual scene displaying an anterior target. Participants aimed to position a virtual laser pointer (via handheld device) in the target. Participants then completed practice trials involving aiming at a lateral target. For this practice session, participants were randomized to either (a) a “constraint” group wherein they received physical constraint to limit postural sway, or (b) a “no-constraint” group. Task performance and postural control were assessed before and after practice, and transfer to another upper-limb task was evaluated. After practice, both groups improved paretic upper-limb performance. For the target task, the no-constraint group showed task-sensitive changes in postural control. The constraint group showed no changes in postural control. At transfer, the constraint group increased postural sway. Constraining postural sway after stroke should be carefully considered with the recognition that postural sway arises from exploratory movements involved in the discovery of adaptable motor solutions.

A multi-task graph deep learning model to predict drugs combination of synergy and sensitivity scores

BackgroundDrug combination treatments have proven to be a realistic technique for treating challenging diseases such as cancer by enhancing efficacy and mitigating side effects. To achieve the therapeutic goals of these combinations, it is essential to employ multi-targeted drug combinations, which maximize effectiveness and synergistic effects.ResultsThis paper proposes ‘MultiComb’, a multi-task deep learning (MTDL) model designed to simultaneously predict the synergy and sensitivity of drug combinations. The model utilizes a graph convolution network to represent the Simplified Molecular-Input Line-Entry (SMILES) of two drugs, generating their respective features. Also, three fully connected subnetworks extract features of the cancer cell line. These drug and cell line features are then concatenated and processed through an attention mechanism, which outputs two optimized feature representations for the target tasks. The cross-stitch model learns the relationship between these tasks. At last, each learned task feature is fed into fully connected subnetworks to predict the synergy and sensitivity scores.The proposed model is validated using the O’Neil benchmark dataset, which includes 38 unique drugs combined to form 17,901 drug combination pairs and tested across 37 unique cancer cells. The model’s performance is tested using some metrics like mean square error (MSE\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$MSE$$\\end{document}), mean absolute error (MAE\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$MAE$$\\end{document}), coefficient of determination (R2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${R}^{2}$$\\end{document}), Spearman, and Pearson scores. The mean synergy scores of the proposed model are 232.37, 9.59, 0.57, 0.76, and 0.73 for the previous metrics, respectively. Also, the values for mean sensitivity scores are 15.59, 2.74, 0.90, 0.95, and 0.95, respectively.ConclusionThis paper proposes an MTDL model to predict synergy and sensitivity scores for drug combinations targeting specific cancer cell lines. The MTDL model demonstrates superior performance compared to existing approaches, providing better results.

Gradient-guided channel masking for cross-domain few-shot learning

Cross-Domain Few-Shot Learning (CD-FSL) addresses the Few-Shot Learning with a domain gap between source and target domains, which facilitates the transfer of knowledge from a source domain to a target domain with limited labeled samples. Current approaches often incorporate an auxiliary target dataset containing a few labeled samples to enhance model generalization on specific target domains. However, we observe that many models retain a substantial number of channels that learn source-specific knowledge and extract features that perform adequately on the source domain but generalize poorly to the target domain. This often results in compromised performance due to the influence of source-specific knowledge. To address this challenge, we introduce a novel framework, Gradient-Guided Channel Masking (GGCM), designed for CD-FSL to mitigate model channels from acquiring too much source-specific knowledge. GGCM quantifies each channel’s contribution to solving target tasks using gradients of target loss and identifies those with smaller gradients as source-specific. These channels are then masked during the forward propagation of source features to mitigate the learning of source-specific knowledge. Conversely, GGCM mutes non-source-specific channels during the forward propagation of target features, forcing the model to depend on the source-specific channels and thereby enhancing their generalizability. Moreover, we propose a consistency loss that aligns the predictions made by source-specific channels with those made by the entire model. This approach further enhances the generalizability of these channels by enabling them to learn from the generalizable knowledge contained in other non-source-specific channels. Validated across multiple CD-FSL benchmark datasets, our framework demonstrates state-of-the-art performance and effectively suppresses the learning of source-specific knowledge.

Efficient Bayesian Policy Reuse With a Scalable Observation Model in Deep Reinforcement Learning.

Bayesian policy reuse (BPR) is a general policy transfer framework for selecting a source policy from an offline library by inferring the task belief based on some observation signals and a trained observation model. In this article, we propose an improved BPR method to achieve more efficient policy transfer in deep reinforcement learning (DRL). First, most BPR algorithms use the episodic return as the observation signal that contains limited information and cannot be obtained until the end of an episode. Instead, we employ the state transition sample, which is informative and instantaneous, as the observation signal for faster and more accurate task inference. Second, BPR algorithms usually require numerous samples to estimate the probability distribution of the tabular-based observation model, which may be expensive and even infeasible to learn and maintain, especially when using the state transition sample as the signal. Hence, we propose a scalable observation model based on fitting state transition functions of source tasks from only a small number of samples, which can generalize to any signals observed in the target task. Moreover, we extend the offline-mode BPR to the continual learning setting by expanding the scalable observation model in a plug-and-play fashion, which can avoid negative transfer when faced with new unknown tasks. Experimental results show that our method can consistently facilitate faster and more efficient policy transfer.

Abnormal recognition-assisted and onset-offset aware network for pathological wearable ECG delineation

Electrocardiogram (ECG) delineation is essential to the identification of abnormal cardiac status, especially when ECG signals are remotely monitored with wearable devices. The complexity and diversity of cardiac conditions generate numerous pathological ECG patterns, not only requiring the recognition of normal ECG but also addressing an extensive range of abnormal ECG patterns, posing a challenging task. Therefore, we propose an abnormal recognition-assisted network to integrate supplementary information on diverse ECG patterns. Simultaneously, we design an onset-offset aware loss to enhance precise waveform localization. Specifically, we establish a two-branch framework where ECG delineation serves as the target task, producing the final segmentation results. Additionally, the abnormal recognition-assisted network serves as an auxiliary task, extracting multi-label pathological information from ECGs. This joint learning approach establishes crucial correlations between ECG delineation and associated ECG abnormalities. The correlations enable the model to demonstrate sufficient generalization in the presence of diverse abnormal ECG patterns. Besides, onset-offset aware loss focuses intensively on wave onsets and offsets by applying biased weights to various waveform positions. This approach ensures a focus on precise localization, facilitating seamless integration into cross-entropy loss function. A large-scale wearable 12-lead dataset containing 4,913 signals is collected, offering an extensive range of ECG data for model training. Results demonstrate that our method achieves outstanding performance on two test datasets, attaining sensitivity of 94.97% and 94.27% and an error tolerance lower than 20 ms. Furthermore, our method is effective for various aberrant ECG signals, including ST-segment changes, atrial premature beats, and right and left bundle branch blocks.

Analisis Kewenangan Intelijen Kejaksaan Dalam Mengungkap Dugaan Tindak Pidana Korupsi Di Wilayah Hukum Kejaksaan Negeri Karanganyar

The Prosecutor's Intelligence has the sex of intelligence responsible for running the entire investigation chain, including the planning, collection, processing, and use of data, to maximize the results of the investigation. When a corruption crime is suspected or has occurred, data and information is collected and processed by the Prosecutor's Intelligence as part of the prosecution's efforts to uncover corruption crimes. At the investigation stage this task is carried out by the intelligence that exists at each stage of the prosecution position. The stage of investigation is carried out by the special criminal section which is given authority in the investigation of criminal offences, while in the stage of prosecution it is conducted by the Prosecutor General as stipulated in the regulations of the law – invitation. (1) What is the role of the Prosecutor's Intelligence in uncovering alleged corruption crimes? (2) Are the obstacles encountered by the Prosecutor's Intelligence in uncovering alleged criminal acts of corruption?The research approach used in this study is the Social Legal Approach conducted by studying the influence of law on social behavior. The location of the research was carried out at the Karanganyar State Prosecutor's Office located at Jl. Lawu No.361, Badran Asri, Cangakan, Kec. Data Collection Techniques (1) Interviews (2) Library is using Primary Data (1) Act No. 11 of 2021 on Amendments to Law No. 16 of 2004 on the Prosecutor's Office of the Republic of Indonesia; (2) Procurator's Regulations No. 6 of 2019 on Amending the Procedure Regulations of the Attorney-General No. PER-006/A/JA/07/2017 on the Organization and Working Rules of the Prosecution of the Republik of Indonesia (3) Act No.3 of 17 of 2011 on State Intelligence (4) Procuror-Governor's Ordinance No. Per-037 / A/JA/09/2011 of 23 September 2011 on the Operational Standards of Procedures (SOPs) of the Intelligence of the Ministry of Indonesia. Secondary data. Data analysis techniques with data analysis techniques used are qualitative analysis.The results of the research and the conclusion that (1) the Prosecutor's Intelligence played an important role in revealing the occurrence of corruption crimes. Before the case is brought into investigation, a network of intelligence activities is required to investigate whether or not there are indications of corruption offences that are detrimental to the state/region's finances. Successful disclosure of corruption crimes is in the hands of prosecutor's intelligence agents. The functions of the prosecution intelligence as the eye and ear of leadership, carrying out supporting in the implementation of all fields, as a source of information, data and support. In the course of the investigation, the intelligence agency performs activities such as Target Analysis, Task Analysis and Operation Target Determination to gather data and gather evidence that will serve as evidence that corruption crime has been committed that damages the state/region finances. Once the data and information are obtained, the intelligence agency of the Justice Intelligence Operation will conduct a more detailed investigation into the corruption crime. The information obtained is processed, analyzed and poured into reports of intelligence operations. If there are indications of corruption, the Intelligence Department will submit the report of the outcome of the judicial intelligence operation to the Special Criminal Prosecution Department for prosecution with the consent of the Chief Prosecutor of Karanganyar State. (2) Obstacles encountered by the Intelligence Branch of the State Prosecutor of Karanganyar in revealing the presence of corruption criminal acts, inter alia: (a) the time limit provided by the prosecutor's regulations; (b) in the Regulations of the Prosecution of the Republic of Indonesia there is no authority of Intelligence Agencies to undertake enforced efforts, so intelligence agents in the search for data must carry out disguise, slide or even steal the required data, this takes time related to the situation and circumstances; (c) Intelligence agents cannot guarantee that the outcomes of the Judicial Intelligence Operations referred to the Special Criminal Area go to the Corruption Penal Court, since the Intelligent Branch is only authorized to conduct investigations and bring the case to the special criminal area to carry out the investigation and prosecution; (d) the incidents or discovered have occurred over a very long time so that it is difficult to trace the existence of tools and evidence of the candidate suspect at the time of the suspension; (e) dealing with longer time due to lack of cooperation with the authorized authorities in the field

Dynamic Routing and Knowledge Re-Learning for Data-Free Black-Box Attack.

Deep learning models have emerged as strong and efficient tools that can be applied to a broad spectrum of complex learning problems and many real-world applications. However, more and more works show that deep models are vulnerable to adversarial examples. Compared to vanilla attack settings, this paper advocates a more practical setting of data-free black-box attack, for which the attackers can completely not access the structures and parameters of the target model, as well as the intermediate features and any training data associated with the model. To tackle this task, previous methods generate transferable adversarial examples from a transparent substitute model to the target model. However, we found that these works have the limitations of taking static substitute model structure for different targets, only using hard synthesized examples once, and still relying on data statistics of the target model. This may potentially harm the performance of attacking the target model. To this end, we propose a novel Dynamic Routing and Knowledge Re-Learning framework (DraKe) to effectively learn a dynamic substitute model from the target model. Specifically, given synthesized training samples, a dynamic substitute structure learning strategy is proposed to adaptively generate optimal substitute model structure via a policy network according to different target models and tasks. To facilitate the substitute training, we present a graph-based structure information learning to capture the structural knowledge learned from the target model. For the inherent limitation that online data generation can only be learned once, a dynamic knowledge re-learning strategy is proposed to adjust the weights of optimization objectives and re-learn hard samples. Extensive experiments on four public image classification datasets and one face recognition benchmark are conducted to evaluate the efficacy of our Drake. We can obtain significant improvement compared with state-of-the-art competitors. More importantly, our DraKe consistently achieves attack superiority for different target models (e.g., residual networks, and vision transformers), showing great potential for complex real-world applications.

M-adapter: Multi-level image-to-video adaptation for video action recognition

With the growing size of visual foundation models, training video models from scratch has become costly and challenging. Recent attempts focus on transferring frozen pre-trained Image Models (PIMs) to video fields by tuning inserted learnable parameters such as adapters and prompts. However, these methods require saving PIM activations for gradient calculations, leading to limited savings of GPU memory. In this paper, we propose a novel parallel branch that adapts the multi-level outputs of the frozen PIM for action recognition. It avoids passing gradients through the PIMs, thus naturally owning much lower GPU memory footprints. The proposed adaptation branch consists of hierarchically combined multi-level output adapters (M-adapters), comprising a fusion module and a temporal module. This design digests the existing discrepancies between the pre-training task and the target task with lower training costs. We show that when using larger models or on scenarios with higher demands for temporal modelling, the proposed method performs better than those with the full-parameter tuning manner. Finally, despite only tuning fewer parameters, our method achieves superior or comparable performance against current state-of-the-art methods.

Decoding motor imagery loaded on steady-state somatosensory evoked potential based on complex task-related component analysis

Background and objectiveMotor Imagery (MI) recognition is one of the most critical decoding problems in brain- computer interface field. Combined with the steady-state somatosensory evoked potential (MI-SSSEP), this new paradigm can achieve higher recognition accuracy than the traditional MI paradigm. Typical algorithms do not fully consider the characteristics of MI-SSSEP signals. Developing an algorithm that fully captures the paradigm's characteristics to reduce false triggering rate is the new step in improving performance. MethodsThe idea to use complex signal task-related component analysis (cTRCA) algorithm for spatial filtering processing has been proposed in this paper according to the features of SSSEP signal. In this research, it's proved from the analysis of simulation signals that task-related component analysis (TRCA) as typical method is affected when the response between stimuli has reduced correlation and the proposed algorithm can effectively overcome this problem. The experimental data under the MI-SSSEP paradigm have been used to identify right-handed target tasks and three unique interference tasks are used to test the false triggering rate. cTRCA demonstrates superior performance as confirmed by the Wilcoxon signed-rank test. ResultsThe recognition algorithm of cTRCA combined with mutual information-based best individual feature (MIBIF) and minimum distance to mean (MDM) can obtain AUC value up to 0.89, which is much higher than traditional algorithm common spatial pattern (CSP) combined with support vector machine (SVM) (the average AUC value is 0.77, p < 0.05). Compared to CSP+SVM, this algorithm model reduced the false triggering rate from 38.69 % to 20.74 % (p < 0.001). ConclusionsThe research prove that TRCA is influenced by MI-SSSEP signals. The results further prove that the motor imagery task in the new paradigm MI-SSSEP causes the phase change in evoked potential. and the cTRCA algorithm based on such phase change is more suitable for this hybrid paradigm and more conducive to decoding the motor imagery task and reducing false triggering rate.

Nondestructive fatigue life prediction for additively manufactured metal parts through a multimodal transfer learning framework

Understanding the fatigue behavior and accurately predicting the fatigue life of laser powder bed fusion (L-PBF) parts remain a pressing challenge due to complex failure mechanisms, time-consuming tests, and limited fatigue data. This study proposes a physics-informed data-driven framework, a multimodal transfer learning (MMTL) framework, to understand process-defect-fatigue relationships in L-PBF by integrating various modalities of fatigue performance, including process parameters, XCT-inspected defects, and fatigue test conditions. It aims to leverage a pre-trained model with abundant process and defect data in the source task to predict fatigue life nondestructively with limited fatigue test data in the target task. MMTL employs a hierarchical graph convolutional network (HGCN) to classify defects in the source task by representing process parameters and defect features in graphs, thereby enhancing its interpretability. The feature embedding learned from HGCN is then transferred to fatigue life modeling in neural network layers, enabling fatigue life prediction for L-PBF parts with limited data. MMTL validation through a numerical simulation and real-case study demonstrates its effectiveness, achieving an F1-score of 0.9593 in defect classification and a mean absolute percentage log error of 0.0425 in fatigue life prediction. MMTL can be extended to other applications with multiple modalities and limited data.

Identification of Priority Areas for Carbon Compensation in Chongqing Based on the Difference in Land Use Carbon Budget

This study aimed to explore the spatiotemporal patterns and balance characteristics of land use carbon budget, measure the value of carbon compensation, and delineate carbon compensation type zoning to provide scientific reference for further strengthening the connection between the construction of an ecological compensation system and the "dual carbon" target task. Based on the land cover data of Chongqing from 2000 to 2020, this study analyzed the spatiotemporal dynamics and balance relationship characteristics of the land use carbon budget. By using the revised carbon compensation model to measure the horizontal compensation standards, the normalized revealed comparative advantage （NRCA） index and K-means clustering analysis method were used to divide the carbon compensation area. The research results demonstrated that： ① the total land use carbon sequestration in Chongqing grew slowly from 2000 to 2020, whereas carbon emissions continued to increase significantly, and the net carbon emissions showed a distribution pattern of "high in the center and low in the two wings." ② The average coefficient of variation in Chongqing was 0.602, and the carbon emission economy contributive coefficient and carbon ecological support coefficient were concentrated between 0.64-1.14 and 0.00-32.86, respectively. The difference in the contribution of carbon emissions and economic benefits between districts and counties was relatively small, but there was a mismatch between carbon supply and demand. ③ A significant spatial difference existed in the value of carbon compensation, with a total of 1.098 billion yuan in carbon payment and 634 million yuan in carbon compensation, respectively. Moreover, it was ultimately determined that there were eight key payment areas, seven general payment areas, three key compensation areas, and 20 general compensation areas. In conclusion, the research results can provide a reference for implementing differentiated development strategies in different types of carbon compensation regions, improve the collaborative governance capacity of the regional ecological environment, and promote the achievement of carbon neutrality goals.