Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are ...Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are built on the assumption of a static learning environment,but in practical situations,the data generated by the process is dynamic.This evolution of the data is termed concept drift.This research paper presents an approach for predictingmechanical failure in real-time using incremental learning based on the statistically calculated parameters of mechanical equipment.The method proposed here is applicable to allmechanical devices that are susceptible to failure or operational degradation.The proposed method in this paper is equipped with the capacity to detect the drift in data generation and adaptation.The proposed approach evaluates the machine learning and deep learning models for their efficacy in handling the errors related to industrial machines due to their dynamic nature.It is observed that,in the settings without concept drift in the data,methods like SVM and Random Forest performed better compared to deep neural networks.However,this resulted in poor sensitivity for the smallest drift in the machine data reported as a drift.In this perspective,DNN generated the stable drift detection method;it reported an accuracy of 84%and an AUC of 0.87 while detecting only a single drift point,indicating the stability to performbetter in detecting and adapting to new data in the drifting environments under industrial measurement settings.展开更多
In this paper, we explore a novel ensemble method for spectral clustering. In contrast to the traditional clustering ensemble methods that combine all the obtained clustering results, we propose the adaptive spectral ...In this paper, we explore a novel ensemble method for spectral clustering. In contrast to the traditional clustering ensemble methods that combine all the obtained clustering results, we propose the adaptive spectral clustering ensemble method to achieve a better clustering solution. This method can adaptively assess the number of the component members, which is not owned by many other algorithms. The component clusterings of the ensemble system are generated by spectral clustering （SC） which bears some good characteristics to engender the diverse committees. The selection process works by evaluating the generated component spectral clustering through resampling technique and population-based incremental learning algorithm （PBIL）. Experimental results on UCI datasets demonstrate that the proposed algorithm can achieve better results compared with traditional clustering ensemble methods, especially when the number of component clusterings is large.展开更多
In order to improve the efficiency of learning the triangular membership functions( TMFs) for mining fuzzy association rule( FAR) in dynamic database,a single-pass fuzzy c means( SPFCM)algorithm is combined with the r...In order to improve the efficiency of learning the triangular membership functions( TMFs) for mining fuzzy association rule( FAR) in dynamic database,a single-pass fuzzy c means( SPFCM)algorithm is combined with the real-coded CHC genetic model to incrementally learn the TMFs. The cluster centers resulting from SPFCM are regarded as the midpoint of TMFs. The population of CHC is generated randomly according to the cluster center and constraint conditions among TMFs. Then a new population for incremental learning is composed of the excellent chromosomes stored in the first genetic process and the chromosomes generated based on the cluster center adjusted by SPFCM. The experiments on real datasets show that the number of generations converging to the solution of the proposed approach is less than that of the existing batch learning approach. The quality of TMFs generated by the approach is comparable to that of the batch learning approach. Compared with the existing incremental learning strategy,the proposed approach is superior in terms of the quality of TMFs and time cost.展开更多
In this study, we propose an incremental learning approach based on a machine-machine interaction via relative attribute feedbacks that exploit comparative relationships among top level image categories. One machine a...In this study, we propose an incremental learning approach based on a machine-machine interaction via relative attribute feedbacks that exploit comparative relationships among top level image categories. One machine acts as ＇Student （S）＇ with initially limited information and it endeavors to capture the task domain gradually by questioning its mentor on a pool of unlabeled data. The other machine is ＇Teacher （T）＇ with the implicit knowledge for helping S on learning the class models. T initiates relative attributes as a communication channel by randomly sorting the classes on attribute space in an unsupervised manner. S starts modeling the categories in this intermediate level by using only a limited number of labeled data. Thereafter, it first selects an entropy-based sample from the pool of unlabeled data and triggers the conversation by propagating the selected image with its belief class in a query. Since T already knows the ground truth labels, it not only decides whether the belief is true or false, but it also provides an attribute-based feedback to S in each case without revealing the true label of the query sample if the belief is false. So the number of training data is increased virtually by dropping the falsely predicted sample back into the unlabeled pool. Next, S updates the attribute space which, in fact, has an impact on T＇s future responses, and then the category models are updated concurrently for the next run. We experience the weakly supervised algorithm on the real world datasets of faces and natural scenes in comparison with direct attribute prediction and semi-supervised learning approaches, and a noteworthy performance increase is achieved.展开更多
Deep Convolution Neural Networks(DCNNs)can capture discriminative features from large datasets.However,how to incrementally learn new samples without forgetting old ones and recognize novel classes that arise in the d...Deep Convolution Neural Networks(DCNNs)can capture discriminative features from large datasets.However,how to incrementally learn new samples without forgetting old ones and recognize novel classes that arise in the dynamically changing world,e.g.,classifying newly discovered fish species,remains an open problem.We address an even more challenging and realistic setting of this problem where new class samples are insufficient,i.e.,Few-Shot Class-Incremental Learning(FSCIL).Current FSCIL methods augment the training data to alleviate the overfitting of novel classes.By contrast,we propose Filter Bank Networks(FBNs)that augment the learnable filters to capture fine-detailed features for adapting to future new classes.In the forward pass,FBNs augment each convolutional filter to a virtual filter bank containing the canonical one,i.e.,itself,and multiple transformed versions.During back-propagation,FBNs explicitly stimulate fine-detailed features to emerge and collectively align all gradients of each filter bank to learn the canonical one.FBNs capture pattern variants that do not yet exist in the pretraining session,thus making it easy to incorporate new classes in the incremental learning phase.Moreover,FBNs introduce model-level prior knowledge to efficiently utilize the limited few-shot data.Extensive experiments on MNIST,CIFAR100,CUB200,andMini-ImageNet datasets show that FBNs consistently outperformthe baseline by a significantmargin,reporting new state-of-the-art FSCIL results.In addition,we contribute a challenging FSCIL benchmark,Fishshot1K,which contains 8261 underwater images covering 1000 ocean fish species.The code is included in the supplementary materials.展开更多
This article presents hourly load forecasting by using an incremental learning model called Online Sequential Extreme Learning Machine(OSELM),which can learn and adapt automatically according to new arrival input.Howe...This article presents hourly load forecasting by using an incremental learning model called Online Sequential Extreme Learning Machine(OSELM),which can learn and adapt automatically according to new arrival input.However,the use of OS-ELM requires a sufficient amount of initial training sample data,which makes OS-ELM inoperable if sufficiently accurate sample data cannot be obtained.To solve this problem,a synthesis of the initial training sample is proposed.The synthesis of the initial sample is achieved by taking the first data received at the start of working and adding random noises to that data to create new and sufficient samples.Then the synthesis samples are used to initial train the OS-ELM.This proposed method is compared with Fully Online Extreme Learning Machine(FOS-ELM),which is an incremental learning model that also does not require the initial training samples.Both the proposed method and FOS-ELM are used for hourly load forecasting from the Hourly Energy Consumption dataset.Experiments have shown that the proposed method with a wide range of noise levels,can forecast hourly load more accurately than the FOS-ELM.展开更多
At this current time,data stream classification plays a key role in big data analytics due to its enormous growth.Most of the existing classification methods used ensemble learning,which is trustworthy but these metho...At this current time,data stream classification plays a key role in big data analytics due to its enormous growth.Most of the existing classification methods used ensemble learning,which is trustworthy but these methods are not effective to face the issues of learning from imbalanced big data,it also supposes that all data are pre-classified.Another weakness of current methods is that it takes a long evaluation time when the target data stream contains a high number of features.The main objective of this research is to develop a new method for incremental learning based on the proposed ant lion fuzzy-generative adversarial network model.The proposed model is implemented in spark architecture.For each data stream,the class output is computed at slave nodes by training a generative adversarial network with the back propagation error based on fuzzy bound computation.This method overcomes the limitations of existing methods as it can classify data streams that are slightly or completely unlabeled data and providing high scalability and efficiency.The results show that the proposed model outperforms stateof-the-art performance in terms of accuracy(0.861)precision(0.9328)and minimal MSE(0.0416).展开更多
Background With the development of information technology,there is a significant increase in the number of network traffic logs mixed with various types of cyberattacks.Traditional intrusion detection systems(IDSs)are...Background With the development of information technology,there is a significant increase in the number of network traffic logs mixed with various types of cyberattacks.Traditional intrusion detection systems(IDSs)are limited in detecting new inconstant patterns and identifying malicious traffic traces in real time.Therefore,there is an urgent need to implement more effective intrusion detection technologies to protect computer security.Methods In this study,we designed a hybrid IDS by combining our incremental learning model(KANSOINN)and active learning to learn new log patterns and detect various network anomalies in real time.Conclusions Experimental results on the NSLKDD dataset showed that KAN-SOINN can be continuously improved and effectively detect malicious logs.Meanwhile,comparative experiments proved that using a hybrid query strategy in active learning can improve the model learning efficiency.展开更多
Intrusion detection is a hot field in the direction of network security.Classical intrusion detection systems are usually based on supervised machine learning models.These offline-trained models usually have better pe...Intrusion detection is a hot field in the direction of network security.Classical intrusion detection systems are usually based on supervised machine learning models.These offline-trained models usually have better performance in the initial stages of system construction.However,due to the diversity and rapid development of intrusion techniques,the trained models are often difficult to detect new attacks.In addition,very little noisy data in the training process often has a considerable impact on the performance of the intrusion detection system.This paper proposes an intrusion detection system based on active incremental learning with the adaptive capability to solve these problems.IDS consists of two modules,namely the improved incremental stacking ensemble learning detection method called Multi-Stacking model and the active learning query module.The stacking model can cope well with concept drift due to the diversity and generalization selection of its base classifiers,but the accuracy does not meet the requirements.The Multi-Stacking model improves the accuracy of the model by adding a voting layer on the basis of the original stacking.The active learning query module improves the detection of known attacks through the committee algorithm,and the improved KNN algorithm can better help detect unknown attacks.We have tested the latest industrial IoT dataset with satisfactory results.展开更多
Recently, deep convolutional neural networks (DCNNs) have achieved remarkable results in image classification tasks. Despite convolutional networks’ great successes, their training process relies on a large amount of...Recently, deep convolutional neural networks (DCNNs) have achieved remarkable results in image classification tasks. Despite convolutional networks’ great successes, their training process relies on a large amount of data prepared in advance, which is often challenging in real-world applications, such as streaming data and concept drift. For this reason, incremental learning (continual learning) has attracted increasing attention from scholars. However, incremental learning is associated with the challenge of catastrophic forgetting: the performance on previous tasks drastically degrades after learning a new task. In this paper, we propose a new strategy to alleviate catastrophic forgetting when neural networks are trained in continual domains. Specifically, two components are applied: data translation based on transfer learning and knowledge distillation. The former translates a portion of new data to reconstruct the partial data distribution of the old domain. The latter uses an old model as a teacher to guide a new model. The experimental results on three datasets have shown that our work can effectively alleviate catastrophic forgetting by a combination of the two methods aforementioned.展开更多
In scenarios where a large amount of data needs to be learned,incremental learning can make full use of old knowledge,signif-icantly reduce the computational cost of the overall learning process,and maintain high perf...In scenarios where a large amount of data needs to be learned,incremental learning can make full use of old knowledge,signif-icantly reduce the computational cost of the overall learning process,and maintain high performance.In this paper,taking the MaxCut problem as our example,we introduce the idea of incremental learning into quantum computing,and propose a Quantum Proactive Incremental Learning algorithm(QPIL).Instead of a one-off training of quantum circuit,QPIL contains a multi-phase training on gradually-increased subgraphs of all vertices,proactively reducing large-scale problems to smaller ones to solve in steps,providing an efficient solution for MaxCut.Specifically,some vertices and corresponding edges are randomly selected for training to obtain optimized parameters of the quantum circuit at first.Then,in each incremental phase,the remaining vertices and corresponding edges are gradually added and the parameters obtained from the previous phase are reused in the parameter initialization of the current phase.We perform experiments on 120 different small-scale graphs,and it shows that QPIL performs superior to prevalent quantum and classical baselines in terms of approximation ratio(AR),time cost,anti-forgetting,and solv-ing stability.In particular,QPIL’s AR surpasses 20%of mainstream quantum baselines,and the time cost is less than 1/5 of them.The idea of QPIL is expected to inspire efficient and high-quality solutions in large-scale MaxCut and other combinatorial optimization problems.展开更多
The field of biometric identification has seen significant advancements over the years,with research focusing on enhancing the accuracy and security of these systems.One of the key developments is the integration of d...The field of biometric identification has seen significant advancements over the years,with research focusing on enhancing the accuracy and security of these systems.One of the key developments is the integration of deep learning techniques in biometric systems.However,despite these advancements,certain challenges persist.One of the most significant challenges is scalability over growing complexity.Traditional methods either require maintaining and securing a growing database,introducing serious security challenges,or relying on retraining the entiremodelwhen new data is introduced-a process that can be computationally expensive and complex.This challenge underscores the need for more efficient methods to scale securely.To this end,we introduce a novel approach that addresses these challenges by integrating multimodal biometrics,cancelable biometrics,and incremental learning techniques.This work is among the first attempts to seamlessly incorporate deep cancelable biometrics with dynamic architectural updates,applied incrementally to the deep learning model as new users are enrolled,achieving high performance with minimal catastrophic forgetting.By leveraging a One-Dimensional Convolutional Neural Network(1D-CNN)architecture combined with a hybrid incremental learning approach,our system achieves high recognition accuracy,averaging 98.98% over incrementing datasets,while ensuring user privacy through cancelable templates generated via a pre-trained CNN model and random projection.The approach demonstrates remarkable adaptability,utilizing the least intrusive biometric traits like facial features and fingerprints,ensuring not only robust performance but also long-term serviceability.展开更多
Learning domain-invariant feature representations is critical to alleviate the distribution differences between training and testing domains.The existing mainstream domain generalization approaches primarily pursue to...Learning domain-invariant feature representations is critical to alleviate the distribution differences between training and testing domains.The existing mainstream domain generalization approaches primarily pursue to align the across-domain distributions to extract the transferable feature representations.However,these representations may be insufficient and unstable.Moreover,these networks may also undergo catastrophic forgetting because the previous learned knowledge is replaced by the new learned knowledge.To cope with these issues,we propose a novel causality-based contrastive incremental learning model for domain generalization,which mainly includes three components:(1)intra-domain causal factorization,(2)inter-domain Mahalanobis similarity metric,and(3)contrastive knowledge distillation.The model extracts intra and inter domain-invariant knowledge to improve model generalization.Specifically,we first introduce a causal factorization to extract intra-domain invariant knowledge.Then,we design a Mahalanobis similarity metric to extract common inter-domain invariant knowledge.Finally,we propose a contrastive knowledge distillation with exponential moving average to distill model parameters in a smooth way to preserve the previous learned knowledge and mitigate model forgetting.Extensive experiments on several domain generalization benchmarks prove that our model achieves the state-of-the-art results,which sufficiently show the effectiveness of our model.展开更多
Current machine learning models for predicting geological conditions during earth pressure balance(EPB)shield tunneling predominantly rely on accurate geological conditions as model label inputs.This study introduces ...Current machine learning models for predicting geological conditions during earth pressure balance(EPB)shield tunneling predominantly rely on accurate geological conditions as model label inputs.This study introduces an innovative approach for the real-time prediction of geological conditions in EPB shield tunneling by utilizing an unsupervised incremental learning model that integrates deep temporal clustering(DTC)with elastic weight consolidation(EWC).The model was trained and tested using data from an EPB shield tunneling project in Nanjing,China.Results demonstrate that the DTC model outperforms nine comparison models by clustering the entire dataset into four distinct groups representing various geological conditions without requiring labeled data.Additionally,integrating EWC into the DTC model significantly enhances its continuous learning capabilities,enabling automatic parameter updates with incoming data and facilitating the real-time recognition of geological conditions.Feature importance was evaluated using the feature elimination method and the Shapley additive explanations(SHAP)method,underscoring the critical roles of earth chamber pressure and cutterhead rotation speed in predicting geological conditions.The proposed EWC-DTC model demonstrates practical utility for EPB shield tunneling in complex environments.展开更多
Digital twin technology used to realize the interactive mapping between digital model and physical entity in virtual space plays a crucial role in promoting the transformation of battery management to digitalization a...Digital twin technology used to realize the interactive mapping between digital model and physical entity in virtual space plays a crucial role in promoting the transformation of battery management to digitalization and intelligence.The key to achieving a digital twin is developing a virtual model that can accurately reflect the physical object.However,the intricate time-varying and nonlinear reaction characteristics within the battery often pose challenges in simulating complex operational conditions and maintaining high accuracy throughout the full life cycle.Incremental learning algorithms are suitable for online streaming data processing and can adapt to concept drift of the data stream by receiving new data online without retraining the entire model from scratch.This paper employs a simplified electrochemical model of the battery and integrates an Aggregated Mondrian Forest with incremental learning capabilities to construct a hybrid mechanism-data battery model.The developed hybrid model can acquire battery data and train in real time during battery operation to realize co-evolution with the physical battery to ensure the voltage prediction accuracy during the full life cycle of the battery.展开更多
The visions of Industry 4.0 and 5.0 have reinforced the industrial environment.They have also made artificial intelligence incorporated as a major facilitator.Diagnosing machine faults has become a solid foundation fo...The visions of Industry 4.0 and 5.0 have reinforced the industrial environment.They have also made artificial intelligence incorporated as a major facilitator.Diagnosing machine faults has become a solid foundation for automatically recognizing machine failure,and thus timely maintenance can ensure safe operations.Transfer learning is a promising solution that can enhance the machine fault diagnosis model by borrowing pre-trained knowledge from the source model and applying it to the target model,which typically involves two datasets.In response to the availability of multiple datasets,this paper proposes using selective and adaptive incremental transfer learning(SA-ITL),which fuses three algorithms,namely,the hybrid selective algorithm,the transferability enhancement algorithm,and the incremental transfer learning algorithm.It is a selective algorithm that enables selecting and ordering appropriate datasets for transfer learning and selecting useful knowledge to avoid negative transfer.The algorithm also adaptively adjusts the portion of training data to balance the learning rate and training time.The proposed algorithm is evaluated and analyzed using ten benchmark datasets.Compared with other algorithms from existing works,SA-ITL improves the accuracy of all datasets.Ablation studies present the accuracy enhancements of the SA-ITL,including the hybrid selective algorithm(1.22%-3.82%),transferability enhancement algorithm(1.91%-4.15%),and incremental transfer learning algorithm(0.605%-2.68%).These also show the benefits of enhancing the target model with heterogeneous image datasets that widen the range of domain selection between source and target domains.展开更多
A new incremental support vector machine （SVM） algorithm is proposed which is based on multiple kernel learning. Through introducing multiple kernel learning into the SVM incremental learning, large scale data set l...A new incremental support vector machine （SVM） algorithm is proposed which is based on multiple kernel learning. Through introducing multiple kernel learning into the SVM incremental learning, large scale data set learning problem can be solved effectively. Furthermore, different punishments are adopted in allusion to the training subset and the acquired support vectors, which may help to improve the performance of SVM. Simulation results indicate that the proposed algorithm can not only solve the model selection problem in SVM incremental learning, but also improve the classification or prediction precision.展开更多
Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize faci...Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize facies identification with high efficiency and accuracy;however,it depends on the usage of a large amount of well-labeled data.To solve this issue,we propose herein an incremental semi-supervised method for intelligent facies identification.Our method considers the continuity of the lateral variation of strata and uses cosine similarity to quantify the similarity of the seismic data feature domain.The maximum-diff erence sample in the neighborhood of the currently used training data is then found to reasonably expand the training sets.This process continuously increases the amount of training data and learns its distribution.We integrate old knowledge while absorbing new ones to realize incremental semi-supervised learning and achieve the purpose of evolving the network models.In this work,accuracy and confusion matrix are employed to jointly control the predicted results of the model from both overall and partial aspects.The obtained values are then applied to a three-dimensional(3D)real dataset and used to quantitatively evaluate the results.Using unlabeled data,our proposed method acquires more accurate and stable testing results compared to conventional supervised learning algorithms that only use well-labeled data.A considerable improvement for small-sample categories is also observed.Using less than 1%of the training data,the proposed method can achieve an average accuracy of over 95%on the 3D dataset.In contrast,the conventional supervised learning algorithm achieved only approximately 85%.展开更多
The maintenance of an aero-engine usually includes three levels,and the maintenance cost and period greatly differ depending on the different maintenance levels.To plan a reasonable maintenance budget program, airline...The maintenance of an aero-engine usually includes three levels,and the maintenance cost and period greatly differ depending on the different maintenance levels.To plan a reasonable maintenance budget program, airlines would like to predict the maintenance level of aero-engine before repairing in terms of performance parameters,which can provide more economic benefits.The maintenance level decision rules are mined using the historical maintenance data of a civil aero-engine based on the rough set theory,and a variety of possible models of updating rules produced by newly increased maintenance cases added to the historical maintenance case database are investigated by the means of incremental machine learning.The continuously updated rules can provide reasonable guidance suggestions for engineers and decision support for planning a maintenance budget program before repairing. The results of an example show that the decision rules become more typical and robust,and they are more accurate to predict the maintenance level of an aero-engine module as the maintenance data increase,which illustrates the feasibility of the represented method.展开更多
In view of the incremental learning problem of manifold learning algorithm, an adaptive neighborhood incremental principal component analysis(PCA) and locality preserving projection(LPP) manifold learning algorithm is...In view of the incremental learning problem of manifold learning algorithm, an adaptive neighborhood incremental principal component analysis(PCA) and locality preserving projection(LPP) manifold learning algorithm is presented, and the incremental learning principle of algorithm is introduced. For incremental sample data, the adjacency and covariance matrices are incrementally updated by the existing samples; then the dimensionality reduction results of the incremental samples are estimated by the dimensionality reduction results of the existing samples; finally, the dimensionality reduction results of the incremental and existing samples are updated by subspace iteration method. The adaptive neighborhood incremental PCA-LPP manifold learning algorithm is applied to processing of gearbox fault signals. The dimensionality reduction results by incremental learning have very small error, compared with those by batch learning. Spatial aggregation of the incremental samples is basically stable, and fault identification rate is increased.展开更多
文摘Ever since the research in machine learning gained traction in recent years,it has been employed to address challenges in a wide variety of domains,including mechanical devices.Most of the machine learning models are built on the assumption of a static learning environment,but in practical situations,the data generated by the process is dynamic.This evolution of the data is termed concept drift.This research paper presents an approach for predictingmechanical failure in real-time using incremental learning based on the statistically calculated parameters of mechanical equipment.The method proposed here is applicable to allmechanical devices that are susceptible to failure or operational degradation.The proposed method in this paper is equipped with the capacity to detect the drift in data generation and adaptation.The proposed approach evaluates the machine learning and deep learning models for their efficacy in handling the errors related to industrial machines due to their dynamic nature.It is observed that,in the settings without concept drift in the data,methods like SVM and Random Forest performed better compared to deep neural networks.However,this resulted in poor sensitivity for the smallest drift in the machine data reported as a drift.In this perspective,DNN generated the stable drift detection method;it reported an accuracy of 84%and an AUC of 0.87 while detecting only a single drift point,indicating the stability to performbetter in detecting and adapting to new data in the drifting environments under industrial measurement settings.
基金Supported by the National Natural Science Foundation of China (60661003)the Research Project Department of Education of Jiangxi Province (GJJ10566)
文摘In this paper, we explore a novel ensemble method for spectral clustering. In contrast to the traditional clustering ensemble methods that combine all the obtained clustering results, we propose the adaptive spectral clustering ensemble method to achieve a better clustering solution. This method can adaptively assess the number of the component members, which is not owned by many other algorithms. The component clusterings of the ensemble system are generated by spectral clustering （SC） which bears some good characteristics to engender the diverse committees. The selection process works by evaluating the generated component spectral clustering through resampling technique and population-based incremental learning algorithm （PBIL）. Experimental results on UCI datasets demonstrate that the proposed algorithm can achieve better results compared with traditional clustering ensemble methods, especially when the number of component clusterings is large.
基金Supported by the National Natural Science Foundation of China(No.61301245,U1533104)
文摘In order to improve the efficiency of learning the triangular membership functions( TMFs) for mining fuzzy association rule( FAR) in dynamic database,a single-pass fuzzy c means( SPFCM)algorithm is combined with the real-coded CHC genetic model to incrementally learn the TMFs. The cluster centers resulting from SPFCM are regarded as the midpoint of TMFs. The population of CHC is generated randomly according to the cluster center and constraint conditions among TMFs. Then a new population for incremental learning is composed of the excellent chromosomes stored in the first genetic process and the chromosomes generated based on the cluster center adjusted by SPFCM. The experiments on real datasets show that the number of generations converging to the solution of the proposed approach is less than that of the existing batch learning approach. The quality of TMFs generated by the approach is comparable to that of the batch learning approach. Compared with the existing incremental learning strategy,the proposed approach is superior in terms of the quality of TMFs and time cost.
文摘In this study, we propose an incremental learning approach based on a machine-machine interaction via relative attribute feedbacks that exploit comparative relationships among top level image categories. One machine acts as ＇Student （S）＇ with initially limited information and it endeavors to capture the task domain gradually by questioning its mentor on a pool of unlabeled data. The other machine is ＇Teacher （T）＇ with the implicit knowledge for helping S on learning the class models. T initiates relative attributes as a communication channel by randomly sorting the classes on attribute space in an unsupervised manner. S starts modeling the categories in this intermediate level by using only a limited number of labeled data. Thereafter, it first selects an entropy-based sample from the pool of unlabeled data and triggers the conversation by propagating the selected image with its belief class in a query. Since T already knows the ground truth labels, it not only decides whether the belief is true or false, but it also provides an attribute-based feedback to S in each case without revealing the true label of the query sample if the belief is false. So the number of training data is increased virtually by dropping the falsely predicted sample back into the unlabeled pool. Next, S updates the attribute space which, in fact, has an impact on T＇s future responses, and then the category models are updated concurrently for the next run. We experience the weakly supervised algorithm on the real world datasets of faces and natural scenes in comparison with direct attribute prediction and semi-supervised learning approaches, and a noteworthy performance increase is achieved.
基金support from the Strategic Priority Research Program of the Chinese Academy of Sciences under Grant No.XDA27000000.
文摘Deep Convolution Neural Networks(DCNNs)can capture discriminative features from large datasets.However,how to incrementally learn new samples without forgetting old ones and recognize novel classes that arise in the dynamically changing world,e.g.,classifying newly discovered fish species,remains an open problem.We address an even more challenging and realistic setting of this problem where new class samples are insufficient,i.e.,Few-Shot Class-Incremental Learning(FSCIL).Current FSCIL methods augment the training data to alleviate the overfitting of novel classes.By contrast,we propose Filter Bank Networks(FBNs)that augment the learnable filters to capture fine-detailed features for adapting to future new classes.In the forward pass,FBNs augment each convolutional filter to a virtual filter bank containing the canonical one,i.e.,itself,and multiple transformed versions.During back-propagation,FBNs explicitly stimulate fine-detailed features to emerge and collectively align all gradients of each filter bank to learn the canonical one.FBNs capture pattern variants that do not yet exist in the pretraining session,thus making it easy to incorporate new classes in the incremental learning phase.Moreover,FBNs introduce model-level prior knowledge to efficiently utilize the limited few-shot data.Extensive experiments on MNIST,CIFAR100,CUB200,andMini-ImageNet datasets show that FBNs consistently outperformthe baseline by a significantmargin,reporting new state-of-the-art FSCIL results.In addition,we contribute a challenging FSCIL benchmark,Fishshot1K,which contains 8261 underwater images covering 1000 ocean fish species.The code is included in the supplementary materials.
文摘This article presents hourly load forecasting by using an incremental learning model called Online Sequential Extreme Learning Machine(OSELM),which can learn and adapt automatically according to new arrival input.However,the use of OS-ELM requires a sufficient amount of initial training sample data,which makes OS-ELM inoperable if sufficiently accurate sample data cannot be obtained.To solve this problem,a synthesis of the initial training sample is proposed.The synthesis of the initial sample is achieved by taking the first data received at the start of working and adding random noises to that data to create new and sufficient samples.Then the synthesis samples are used to initial train the OS-ELM.This proposed method is compared with Fully Online Extreme Learning Machine(FOS-ELM),which is an incremental learning model that also does not require the initial training samples.Both the proposed method and FOS-ELM are used for hourly load forecasting from the Hourly Energy Consumption dataset.Experiments have shown that the proposed method with a wide range of noise levels,can forecast hourly load more accurately than the FOS-ELM.
基金Taif University Researchers Supporting Project Number(TURSP-2020/126),Taif University,Taif,Saudi Arabia.
文摘At this current time,data stream classification plays a key role in big data analytics due to its enormous growth.Most of the existing classification methods used ensemble learning,which is trustworthy but these methods are not effective to face the issues of learning from imbalanced big data,it also supposes that all data are pre-classified.Another weakness of current methods is that it takes a long evaluation time when the target data stream contains a high number of features.The main objective of this research is to develop a new method for incremental learning based on the proposed ant lion fuzzy-generative adversarial network model.The proposed model is implemented in spark architecture.For each data stream,the class output is computed at slave nodes by training a generative adversarial network with the back propagation error based on fuzzy bound computation.This method overcomes the limitations of existing methods as it can classify data streams that are slightly or completely unlabeled data and providing high scalability and efficiency.The results show that the proposed model outperforms stateof-the-art performance in terms of accuracy(0.861)precision(0.9328)and minimal MSE(0.0416).
基金Supported by SJTU-HUAWEI TECH Cybersecurity Innovation Lab。
文摘Background With the development of information technology,there is a significant increase in the number of network traffic logs mixed with various types of cyberattacks.Traditional intrusion detection systems(IDSs)are limited in detecting new inconstant patterns and identifying malicious traffic traces in real time.Therefore,there is an urgent need to implement more effective intrusion detection technologies to protect computer security.Methods In this study,we designed a hybrid IDS by combining our incremental learning model(KANSOINN)and active learning to learn new log patterns and detect various network anomalies in real time.Conclusions Experimental results on the NSLKDD dataset showed that KAN-SOINN can be continuously improved and effectively detect malicious logs.Meanwhile,comparative experiments proved that using a hybrid query strategy in active learning can improve the model learning efficiency.
基金sponsored by the National Natural Science Foundation of China under Grants 62271264,61972207,and 42175194the Project through the Priority Academic Program Development(PAPD)of Jiangsu Higher Education Institution.
文摘Intrusion detection is a hot field in the direction of network security.Classical intrusion detection systems are usually based on supervised machine learning models.These offline-trained models usually have better performance in the initial stages of system construction.However,due to the diversity and rapid development of intrusion techniques,the trained models are often difficult to detect new attacks.In addition,very little noisy data in the training process often has a considerable impact on the performance of the intrusion detection system.This paper proposes an intrusion detection system based on active incremental learning with the adaptive capability to solve these problems.IDS consists of two modules,namely the improved incremental stacking ensemble learning detection method called Multi-Stacking model and the active learning query module.The stacking model can cope well with concept drift due to the diversity and generalization selection of its base classifiers,but the accuracy does not meet the requirements.The Multi-Stacking model improves the accuracy of the model by adding a voting layer on the basis of the original stacking.The active learning query module improves the detection of known attacks through the committee algorithm,and the improved KNN algorithm can better help detect unknown attacks.We have tested the latest industrial IoT dataset with satisfactory results.
文摘Recently, deep convolutional neural networks (DCNNs) have achieved remarkable results in image classification tasks. Despite convolutional networks’ great successes, their training process relies on a large amount of data prepared in advance, which is often challenging in real-world applications, such as streaming data and concept drift. For this reason, incremental learning (continual learning) has attracted increasing attention from scholars. However, incremental learning is associated with the challenge of catastrophic forgetting: the performance on previous tasks drastically degrades after learning a new task. In this paper, we propose a new strategy to alleviate catastrophic forgetting when neural networks are trained in continual domains. Specifically, two components are applied: data translation based on transfer learning and knowledge distillation. The former translates a portion of new data to reconstruct the partial data distribution of the old domain. The latter uses an old model as a teacher to guide a new model. The experimental results on three datasets have shown that our work can effectively alleviate catastrophic forgetting by a combination of the two methods aforementioned.
基金supported by the National Natural Science Foundation of China(Grant Nos.62272048,61976056,and 62371069)BUPT Excellent Ph.D.Students Foundation(Grant No.CX20241055)。
文摘In scenarios where a large amount of data needs to be learned,incremental learning can make full use of old knowledge,signif-icantly reduce the computational cost of the overall learning process,and maintain high performance.In this paper,taking the MaxCut problem as our example,we introduce the idea of incremental learning into quantum computing,and propose a Quantum Proactive Incremental Learning algorithm(QPIL).Instead of a one-off training of quantum circuit,QPIL contains a multi-phase training on gradually-increased subgraphs of all vertices,proactively reducing large-scale problems to smaller ones to solve in steps,providing an efficient solution for MaxCut.Specifically,some vertices and corresponding edges are randomly selected for training to obtain optimized parameters of the quantum circuit at first.Then,in each incremental phase,the remaining vertices and corresponding edges are gradually added and the parameters obtained from the previous phase are reused in the parameter initialization of the current phase.We perform experiments on 120 different small-scale graphs,and it shows that QPIL performs superior to prevalent quantum and classical baselines in terms of approximation ratio(AR),time cost,anti-forgetting,and solv-ing stability.In particular,QPIL’s AR surpasses 20%of mainstream quantum baselines,and the time cost is less than 1/5 of them.The idea of QPIL is expected to inspire efficient and high-quality solutions in large-scale MaxCut and other combinatorial optimization problems.
基金the Deputyship for Research&Innovation,Ministry of Education in Saudi Arabia for funding this research work through project number RI-44-0833.
文摘The field of biometric identification has seen significant advancements over the years,with research focusing on enhancing the accuracy and security of these systems.One of the key developments is the integration of deep learning techniques in biometric systems.However,despite these advancements,certain challenges persist.One of the most significant challenges is scalability over growing complexity.Traditional methods either require maintaining and securing a growing database,introducing serious security challenges,or relying on retraining the entiremodelwhen new data is introduced-a process that can be computationally expensive and complex.This challenge underscores the need for more efficient methods to scale securely.To this end,we introduce a novel approach that addresses these challenges by integrating multimodal biometrics,cancelable biometrics,and incremental learning techniques.This work is among the first attempts to seamlessly incorporate deep cancelable biometrics with dynamic architectural updates,applied incrementally to the deep learning model as new users are enrolled,achieving high performance with minimal catastrophic forgetting.By leveraging a One-Dimensional Convolutional Neural Network(1D-CNN)architecture combined with a hybrid incremental learning approach,our system achieves high recognition accuracy,averaging 98.98% over incrementing datasets,while ensuring user privacy through cancelable templates generated via a pre-trained CNN model and random projection.The approach demonstrates remarkable adaptability,utilizing the least intrusive biometric traits like facial features and fingerprints,ensuring not only robust performance but also long-term serviceability.
基金supported by the Pre-research Project on Civil Aerospace Technologies of China National Space Administration(No.D010301).
文摘Learning domain-invariant feature representations is critical to alleviate the distribution differences between training and testing domains.The existing mainstream domain generalization approaches primarily pursue to align the across-domain distributions to extract the transferable feature representations.However,these representations may be insufficient and unstable.Moreover,these networks may also undergo catastrophic forgetting because the previous learned knowledge is replaced by the new learned knowledge.To cope with these issues,we propose a novel causality-based contrastive incremental learning model for domain generalization,which mainly includes three components:(1)intra-domain causal factorization,(2)inter-domain Mahalanobis similarity metric,and(3)contrastive knowledge distillation.The model extracts intra and inter domain-invariant knowledge to improve model generalization.Specifically,we first introduce a causal factorization to extract intra-domain invariant knowledge.Then,we design a Mahalanobis similarity metric to extract common inter-domain invariant knowledge.Finally,we propose a contrastive knowledge distillation with exponential moving average to distill model parameters in a smooth way to preserve the previous learned knowledge and mitigate model forgetting.Extensive experiments on several domain generalization benchmarks prove that our model achieves the state-of-the-art results,which sufficiently show the effectiveness of our model.
基金financially supported by the National Natural Science Foundation of China(Grant Nos.52378392,52408356)Foal Eagle Program Youth Top-notch Talent Project of Fujian Province,China(Grant No.00387088).
文摘Current machine learning models for predicting geological conditions during earth pressure balance(EPB)shield tunneling predominantly rely on accurate geological conditions as model label inputs.This study introduces an innovative approach for the real-time prediction of geological conditions in EPB shield tunneling by utilizing an unsupervised incremental learning model that integrates deep temporal clustering(DTC)with elastic weight consolidation(EWC).The model was trained and tested using data from an EPB shield tunneling project in Nanjing,China.Results demonstrate that the DTC model outperforms nine comparison models by clustering the entire dataset into four distinct groups representing various geological conditions without requiring labeled data.Additionally,integrating EWC into the DTC model significantly enhances its continuous learning capabilities,enabling automatic parameter updates with incoming data and facilitating the real-time recognition of geological conditions.Feature importance was evaluated using the feature elimination method and the Shapley additive explanations(SHAP)method,underscoring the critical roles of earth chamber pressure and cutterhead rotation speed in predicting geological conditions.The proposed EWC-DTC model demonstrates practical utility for EPB shield tunneling in complex environments.
基金the Shandong Province National Natural Science Foundation of China(No.ZR2023QE036)the Natural Science Foundation of Jiangsu Province(No.BK20210600)for their financial support.
文摘Digital twin technology used to realize the interactive mapping between digital model and physical entity in virtual space plays a crucial role in promoting the transformation of battery management to digitalization and intelligence.The key to achieving a digital twin is developing a virtual model that can accurately reflect the physical object.However,the intricate time-varying and nonlinear reaction characteristics within the battery often pose challenges in simulating complex operational conditions and maintaining high accuracy throughout the full life cycle.Incremental learning algorithms are suitable for online streaming data processing and can adapt to concept drift of the data stream by receiving new data online without retraining the entire model from scratch.This paper employs a simplified electrochemical model of the battery and integrates an Aggregated Mondrian Forest with incremental learning capabilities to construct a hybrid mechanism-data battery model.The developed hybrid model can acquire battery data and train in real time during battery operation to realize co-evolution with the physical battery to ensure the voltage prediction accuracy during the full life cycle of the battery.
文摘The visions of Industry 4.0 and 5.0 have reinforced the industrial environment.They have also made artificial intelligence incorporated as a major facilitator.Diagnosing machine faults has become a solid foundation for automatically recognizing machine failure,and thus timely maintenance can ensure safe operations.Transfer learning is a promising solution that can enhance the machine fault diagnosis model by borrowing pre-trained knowledge from the source model and applying it to the target model,which typically involves two datasets.In response to the availability of multiple datasets,this paper proposes using selective and adaptive incremental transfer learning(SA-ITL),which fuses three algorithms,namely,the hybrid selective algorithm,the transferability enhancement algorithm,and the incremental transfer learning algorithm.It is a selective algorithm that enables selecting and ordering appropriate datasets for transfer learning and selecting useful knowledge to avoid negative transfer.The algorithm also adaptively adjusts the portion of training data to balance the learning rate and training time.The proposed algorithm is evaluated and analyzed using ten benchmark datasets.Compared with other algorithms from existing works,SA-ITL improves the accuracy of all datasets.Ablation studies present the accuracy enhancements of the SA-ITL,including the hybrid selective algorithm(1.22%-3.82%),transferability enhancement algorithm(1.91%-4.15%),and incremental transfer learning algorithm(0.605%-2.68%).These also show the benefits of enhancing the target model with heterogeneous image datasets that widen the range of domain selection between source and target domains.
基金supported by the National Natural Science Key Foundation of China(69974021)
文摘A new incremental support vector machine （SVM） algorithm is proposed which is based on multiple kernel learning. Through introducing multiple kernel learning into the SVM incremental learning, large scale data set learning problem can be solved effectively. Furthermore, different punishments are adopted in allusion to the training subset and the acquired support vectors, which may help to improve the performance of SVM. Simulation results indicate that the proposed algorithm can not only solve the model selection problem in SVM incremental learning, but also improve the classification or prediction precision.
基金financially supported by the National Key R&D Program of China(No.2018YFA0702504)the National Natural Science Foundation of China(No.42174152 and No.41974140)+1 种基金the Science Foundation of China University of Petroleum,Beijing(No.2462020YXZZ008 and No.2462020QZDX003)the Strategic Cooperation Technology Projects of CNPC and CUPB(No.ZLZX2020-03).
文摘Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize facies identification with high efficiency and accuracy;however,it depends on the usage of a large amount of well-labeled data.To solve this issue,we propose herein an incremental semi-supervised method for intelligent facies identification.Our method considers the continuity of the lateral variation of strata and uses cosine similarity to quantify the similarity of the seismic data feature domain.The maximum-diff erence sample in the neighborhood of the currently used training data is then found to reasonably expand the training sets.This process continuously increases the amount of training data and learns its distribution.We integrate old knowledge while absorbing new ones to realize incremental semi-supervised learning and achieve the purpose of evolving the network models.In this work,accuracy and confusion matrix are employed to jointly control the predicted results of the model from both overall and partial aspects.The obtained values are then applied to a three-dimensional(3D)real dataset and used to quantitatively evaluate the results.Using unlabeled data,our proposed method acquires more accurate and stable testing results compared to conventional supervised learning algorithms that only use well-labeled data.A considerable improvement for small-sample categories is also observed.Using less than 1%of the training data,the proposed method can achieve an average accuracy of over 95%on the 3D dataset.In contrast,the conventional supervised learning algorithm achieved only approximately 85%.
基金Supported by the National Natural Science Foundation of China(60939003)
文摘The maintenance of an aero-engine usually includes three levels,and the maintenance cost and period greatly differ depending on the different maintenance levels.To plan a reasonable maintenance budget program, airlines would like to predict the maintenance level of aero-engine before repairing in terms of performance parameters,which can provide more economic benefits.The maintenance level decision rules are mined using the historical maintenance data of a civil aero-engine based on the rough set theory,and a variety of possible models of updating rules produced by newly increased maintenance cases added to the historical maintenance case database are investigated by the means of incremental machine learning.The continuously updated rules can provide reasonable guidance suggestions for engineers and decision support for planning a maintenance budget program before repairing. The results of an example show that the decision rules become more typical and robust,and they are more accurate to predict the maintenance level of an aero-engine module as the maintenance data increase,which illustrates the feasibility of the represented method.
基金the National Natural Science Foundation of China(No.50775219)
文摘In view of the incremental learning problem of manifold learning algorithm, an adaptive neighborhood incremental principal component analysis(PCA) and locality preserving projection(LPP) manifold learning algorithm is presented, and the incremental learning principle of algorithm is introduced. For incremental sample data, the adjacency and covariance matrices are incrementally updated by the existing samples; then the dimensionality reduction results of the incremental samples are estimated by the dimensionality reduction results of the existing samples; finally, the dimensionality reduction results of the incremental and existing samples are updated by subspace iteration method. The adaptive neighborhood incremental PCA-LPP manifold learning algorithm is applied to processing of gearbox fault signals. The dimensionality reduction results by incremental learning have very small error, compared with those by batch learning. Spatial aggregation of the incremental samples is basically stable, and fault identification rate is increased.
