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
According to the index early warning method, a commercial bank loans risk early warning system based on BP neural networks is proposed. The warning signal is mainly involved with the financial situation signal of loan...According to the index early warning method, a commercial bank loans risk early warning system based on BP neural networks is proposed. The warning signal is mainly involved with the financial situation signal of loaning corporation. Except the structure description of the system structure the demonstration of attemptive designing is also elaborated.展开更多
With the advent of Internet financial innovation,many commercial banks quietly have started to enter into the Ecommercial in order to prevent oligarchs from eroding financial market.From the perspective of industrial ...With the advent of Internet financial innovation,many commercial banks quietly have started to enter into the Ecommercial in order to prevent oligarchs from eroding financial market.From the perspective of industrial division,this paper reveals the nature of a phenomenon that E-commercial enterprises and banks have stepped into each other's field,which E-commerce of banks can give full play to network effects.Then it uses game theory to analyze the motions of banks to involve into E-commerce and the short-term competitive equilibrium of large incumbent Ecommercial enterprises as well.For individual rationality,the dominant strategy of banks and E-commercial enterprises is(enter,counterattack).Considering network externalities,it constructs a competing model on banks and incumbent E-commercial enterprises and simulates competitive trends and balanced results of their behaviors,which illustrates that banks can obtain network effect after choosing E-commerce strategy.展开更多
Nowadays the optimal allocation of distributed generation (DG) in the distribution network becomes the popular research area in restructuring of power system. The capacitor banks introduced in the distribution network...Nowadays the optimal allocation of distributed generation (DG) in the distribution network becomes the popular research area in restructuring of power system. The capacitor banks introduced in the distribution networks for reactive power compensation also have the capacity to minimize the real and reactive power losses occurred in the system. Hence, this research integrates the allocation of renewable energy DG and capacitor banks in the radial distribution network to minimize the real power loss occurred in the system. A two-stage methodology is used for simultaneous allocation of renewable DG and capacitor banks. The optimum location of renewable energy DG and capacitor banks is determined using the distributed generation sitting index (DGSI) ranking method and the optimum sizing of DG and capacitor banks is found out for simultaneous placement using weight improved particle swarm optimization algorithm (WIPSO) and self adaptive differential evolution algorithm (SADE). This two-stage methodology reduces the burden of SADE and WIPSO algorithm, by using the DGSI index in determining the optimal location. Hence the computational time gets reduced which makes them suitable for online applications. By using the above methodology, a comprehensive performance analysis is done on IEEE 33 bus and 69 bus RDNs and the results are discussed in detail.展开更多
Intelligent Decision Support System (IISS) for Bank Loans Risk Classification (BLRC), based on the way of integration Artificial Neural Network (ANN) and Expert System (ES), is proposed. According to the feature of BL...Intelligent Decision Support System (IISS) for Bank Loans Risk Classification (BLRC), based on the way of integration Artificial Neural Network (ANN) and Expert System (ES), is proposed. According to the feature of BLRC, the key financial and non-financial factors are analyzed. Meanwhile, ES and Model Base (MB) which contain ANN are designed . The general framework,interaction and integration of the system are given. In addition, how the system realizes BLRC is elucidated in detail.展开更多
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.
基金Supported by the National Science Foundation of China(Approved NO.79770086)
文摘According to the index early warning method, a commercial bank loans risk early warning system based on BP neural networks is proposed. The warning signal is mainly involved with the financial situation signal of loaning corporation. Except the structure description of the system structure the demonstration of attemptive designing is also elaborated.
基金National Natural Science Foundation of China(No.71501094)National Social Science Foundation of China(No.15BJY160)
文摘With the advent of Internet financial innovation,many commercial banks quietly have started to enter into the Ecommercial in order to prevent oligarchs from eroding financial market.From the perspective of industrial division,this paper reveals the nature of a phenomenon that E-commercial enterprises and banks have stepped into each other's field,which E-commerce of banks can give full play to network effects.Then it uses game theory to analyze the motions of banks to involve into E-commerce and the short-term competitive equilibrium of large incumbent Ecommercial enterprises as well.For individual rationality,the dominant strategy of banks and E-commercial enterprises is(enter,counterattack).Considering network externalities,it constructs a competing model on banks and incumbent E-commercial enterprises and simulates competitive trends and balanced results of their behaviors,which illustrates that banks can obtain network effect after choosing E-commerce strategy.
文摘Nowadays the optimal allocation of distributed generation (DG) in the distribution network becomes the popular research area in restructuring of power system. The capacitor banks introduced in the distribution networks for reactive power compensation also have the capacity to minimize the real and reactive power losses occurred in the system. Hence, this research integrates the allocation of renewable energy DG and capacitor banks in the radial distribution network to minimize the real power loss occurred in the system. A two-stage methodology is used for simultaneous allocation of renewable DG and capacitor banks. The optimum location of renewable energy DG and capacitor banks is determined using the distributed generation sitting index (DGSI) ranking method and the optimum sizing of DG and capacitor banks is found out for simultaneous placement using weight improved particle swarm optimization algorithm (WIPSO) and self adaptive differential evolution algorithm (SADE). This two-stage methodology reduces the burden of SADE and WIPSO algorithm, by using the DGSI index in determining the optimal location. Hence the computational time gets reduced which makes them suitable for online applications. By using the above methodology, a comprehensive performance analysis is done on IEEE 33 bus and 69 bus RDNs and the results are discussed in detail.
基金the National Natural Science Fund of China(Approved No.79779986)
文摘Intelligent Decision Support System (IISS) for Bank Loans Risk Classification (BLRC), based on the way of integration Artificial Neural Network (ANN) and Expert System (ES), is proposed. According to the feature of BLRC, the key financial and non-financial factors are analyzed. Meanwhile, ES and Model Base (MB) which contain ANN are designed . The general framework,interaction and integration of the system are given. In addition, how the system realizes BLRC is elucidated in detail.
