Advances in computational intelligence最新文献

Feature selection can effectively eliminate irrelevant or redundant features without changing features semantics, so as to improve the performance of learning and reduce the training time. In most of the existing feature selection methods based on rough sets, eliminating the redundant features between features and decisions, and deleting the redundant features between features are performed separately. This will greatly increase the search time of feature subset. To quickly remove redundant features, we define a series of feature evaluation functions that consider both the consistency between features and decisions, and redundancy between features, then propose a novel feature selection method based on min-redundancy and max-consistency. Firstly, we define the consistency of features with respect to decisions and the redundancy between features from neighborhood information granules. Then we propose a combined criterion to measure the importance of features and design a feature selection algorithm based on minimal-redundancy-maximal-consistency (mRMC). Finally, on UCI data sets, mRMC is compared with three other popular feature selection algorithms based on neighborhood idea, from classification accuracy, the number of selected features and running time. The experimental comparison shows that mRMC can quickly delete redundant features and select useful features while ensuring classification accuracy.

The existing monotonic decision tree algorithms are based on a linearly ordered constraint that certain attributes are monotonously consistent with the decision, which could be called monotonic attributes, whereas others, called non-monotonic attributes. In practice, monotonic and non-monotonic attributes coexist in most classification tasks, and some attribute values are even evaluated as interval numbers. In this paper, we proposed a fuzzy rank-inconsistent rate based on probability degree to judge the monotonicity of interval numbers. Furthermore, we devised a hybrid model composed of monotonic and non-monotonic attributes to construct a mixed monotone decision tree for interval-valued data. Experiments on artificial and real-world data sets show that the proposed hybrid model is effective.

Continual learning models are known to suffer from catastrophic forgetting. Existing regularization methods to countering forgetting operate by penalizing large changes to learned parameters. A significant downside to these methods, however, is that, by effectively freezing model parameters, they gradually suspend the capacity of a model to learn new tasks. In this paper, we explore an alternative approach to the continual learning problem that aims to circumvent this downside. In particular, we ask the question: instead of forcing continual learning models to remember the past, can we modify the learning process from the start, such that the learned representations are less susceptible to forgetting? To this end, we explore multiple methods that could potentially encourage durable representations. We demonstrate empirically that the use of unsupervised auxiliary tasks achieves significant reduction in parameter re-optimization across tasks, and consequently reduces forgetting, without explicitly penalizing forgetting. Moreover, we propose a distance metric to track internal model dynamics across tasks, and use it to gain insight into the workings of our proposed approach, as well as other recently proposed methods.

Social recommendation has become an important technique of various online commerce platforms, which aims to predict the user preference based on the social network and the interactive network. Social recommendation, which can naturally integrate social information and interactive structure, has been demonstrated to be powerful in solving data sparsity and cold-start problems. Although some of the existing methods have been proven effective, the following two insights are often neglected. First, except for the explicit connections, social information contains implicit connections, e.g., indirect social relations. Indirect social relations can effectively improve the quality of recommendation when users only have few direct social relations. Second, the strength of social influence between users is different. In other words, users have different degrees of trust in different friends. These insights motivate us to propose a novel social recommendation model SIER (short for Social Influence-based Effective Recommendation) in this paper, which incorporates interactive information and social information into personal latent factors learning for social influence-based recommendation. Specifically, user preferences are captured in behavior history and social relations, i.e., user latent factors are shared in interactive network and social network. In particular, we utilize an overlapping community detection method to sufficiently capture the implicit relations in the social network. Extensive experiments on two real-world datasets demonstrate the effectiveness of the proposed method.

In classification tasks, unlabeled data bring the uncertainty in the learning process, which may result in the degradation of the performance. In this paper, we propose a novel semi-supervised inception neural network ensemble-based architecture to achieve missing label imputation. The main idea of the proposed architecture is to use smaller ensembles within a larger ensemble to involve diverse ways of missing label imputation and internal transformation of feature representation, towards enhancing the prediction accuracy. Following the process of imputing the missing labels of unlabeled data, the human-labeled data and the data with imputed labels are used together as a training set for the credible classifiers learning. Meanwhile, we discuss how this proposed approach is more effective as compared to the traditional ensemble learning approaches. Our proposed approach is evaluated on different well-known benchmark data sets, and the experimental results show the effectiveness of the proposed method. In addition, the approach is validated by statistical analysis using Wilcoxon signed rank test and the results indicate statistical significance of the performance improvement in comparison with other methods.

Self-supervised learning based on mutual information makes good use of classification models and label information produced by clustering tasks to train networks parameters, and then updates the downstream clustering assignment with respect to maximizing mutual information between label information. This kind of methods have attracted more and more attention and obtained better progress, but there is still a larger improvement space compared with the methods of supervised learning, especially on the challenge image datasets. To this end, a self-supervised deep clustering method by maximizing mutual information is proposed (bi-MIM-SSC), where deep convolutional network is employed as a feature encoder. The first term is to maximize mutual information between output-feature pairs for importing more semantic meaning to the output features. The second term is to maximize mutual information between an input image and its feature generated by the encoder for keeping the useful information of an original image in latent space as possible. Furthermore, pre-training is carried out to further enhance the representation ability of the encoder, and the auxiliary over-clustering is added in clustering network. The performance of the proposed method bi-MIM-SSC is compared with other clustering methods on the CIFAR10, CIFAR100 and STL10 datasets. Experimental results demonstrate that the proposed bi-MIM-SSC method has better feature representation ability and provide better clustering results.

Extreme learning machine (ELM) possesses merits of rapid learning speed and good generalization ability. However, due to the random initialization of connection weights, the network outputs of ELM are usually unstable. Similar to ELM, one-class ELM (OCELM) also has the disadvantage of output instability. To enhance the stability and generalization performance of OCELM, a selective ensemble of OCELMs based on rotation transformation is proposed. First, principal component analysis (PCA)-based rotation transformation is utilized to construct different transformed training sets. Furthermore, several component OCELMs are trained independently on these training sets. Second, a dissimilarity measure based on angle cosine is used to evaluate the dissimilarity between each pair of OCELMs. The diversity of each component OCELM in the obtained ensemble can be further achieved. Thereafter, the component OCELMs with lower value of diversity are removed from the original ensemble. Finally, the voting strategy is utilized to determine that testing samples belong to the target class or the non-target class. Experimental results on 15 UCI benchmark data sets and one handwritten digit data set show that the proposed method is superior to its related approaches.

Security and safety are a necessity for automated teller machines (ATM). The ATM security system is implemented using the Internet of things (IoT) and GPS (global positioning system). The main idea of this project is to develop an ATM surveillance and security system. In this project, when any physical attack against the ATM takes place, then information about the attack is sent using IoT and also alerts the surrounding area using a buzzer, at the same time the entire data from the sensors is sent to the developed mobile application and puts alert message to the bank officials. The officials who have control over the mobile application can control the Door through their mobile to lock from their location remotely. To prevent the escape of the thief chloroform connected to the controller through relay can also be sprayed inside the ATM by the officials remotely from their place using the mobile app. The Camera (ESP32) is used for live video coverage and to monitor the activity inside the ATM. The Camera will not only record the activity but also, transmit will live video taken inside the ATM and the ATM location as latitude and longitude are tracked using GPS. The system is connected to the Blynk mobile application. The sensor and GPS data are read by the microcontroller and these data are sent to the Blynk application. With the help of the Blynk application, the official who has access to it can control the relays and the respective devices connected to the relay to turn it ON or OFF. It can be used in many real-time applications such as banks, homes, and industries.

Wheat is the most important source of food on earth and vital for food safety. It contains 75–80% carbohydrates, 9–18% protein, fiber, many vitamins (especially B vitamins), calcium, iron, and many macronutrients and micronutrients. According to data from the International Grain Council (IGC), wheat has continued to be the most important food grain source for humans in the world. Therefore, determining wheat production behavior has a very important role in food security. In this study, we have modeled and forecasted the production of wheat for 6 years from 2020 to 2025 using ARIMA and Holt’s linear trend models in Afghanistan, Bangladesh, Bhutan, China, India, Nepal, and Pakistan, which are all countries in the South Asian region. Since there is an expectation of a decrease in wheat production in some of these countries, this study can provide these countries with the information they need to take appropriate decisions to prevent the occurrence of food problems in the future and to help deal with food security. Moreover, this projection helps with policy implications and planning.

Discounted 0-1 knapsack problem (D0-1KP) has been proved to be NP-hard, thus a lot of researches focus on designing non-deterministic algorithms to solve it. Group theory-based optimization algorithm (GTOA), as a recently proposed evolutionary algorithm (EA), can provide satisfactory results to D0-1KP. GTOA introduces important theories of algebra, i.e., group theory, to describe combinatorial optimization problems, and applies the classic operations in group theory to design operators for EA. In order to generate a better solution according to a set of existing solutions during each evolutionary iteration, an important operator called random linear combination operator (RLCO) is designed. However, the practical meaning of applying the operations in group theory is hard to explain, and the proposed RLCO is lack of interpretability, causing difficulties in analyzing and improving the algorithm. In this paper, to improve the interpretability and further enhance the performance, we propose a new operator named random xor operator (RXO), and interpret it from the view point of bitwise operation. By replacing RLCO with RXO, a new GTOA algorithm is realized for D0-1KP. Experimental results demonstrate that it can provide very competitive performance.