Summary
Efficient management of big data becomes challenging in recent decades. Online Feature Selection (OFS) is one type of online learning in contrast to batch learning, allowing a classifier to have small and fixed number of features. The major aim of this work is to introduce an OFS algorithm supported on meta‐heuristic algorithm that exploits the MapReduce paradigm. A novel Hybrid Multi‐Objective Firefly and Simulated Annealing (HMOFSA) algorithm is proposed to select optimal set of features. Therefore, as a first step, the original big dataset is decomposed into blocks of examples in the map phase. Subsequently, HMOFSA algorithm is employed to choose the selected features from examples. After that, the attained partial outcomes will be combined into a final vector of features in the reduce phase and evaluated using Kernel Support Vector Machine (KSVM) classifier. The mentioned OFS approach is analyzed with the help of the well‐known classifiers (Logistic Regression, KSVM and Naïve Bayes) developed within the Spark framework. Experiments were conducted on big datasets, containing 66 million samples and 2000 attributes that confirm the proficiency of proposed work. The proposed KSVM classifier results are measured in terms of the metrics like Precision, Recall, Geometric‐mean (G‐mean), F‐measure, and accuracy.
All-to-all broadcast communication is to disseminate a unique message from each node to every other node in a network. This is a fundamental problem in multiprocessor systems and telecommunication networks that need to collect information about other nodes in the network regularly in order to manage network resources efficiently. It is known that in wavelength division multiplexing (WDM) optical networks with single hop routing, the number of wavelengths needed for realizing all-to-all broadcast is quite large, even for a moderately sized network. One possible method to reduce the number of wavelengths is to adopt a network with drop-and-continue capable nodes. In such networks, a message from a source node can be dropped only at a limited number of destination nodes along a light path due to power loss of the dropping optical signals at the destination nodes. In this paper, a new wavelength assignment method is proposed to establish all-to-all broadcast in unidirectional ring and bi-directional ring networks which contain drop-and-continue capable nodes. An upper bound on the number of wavelengths required to establish all-to-all broadcast is also derived for unidirectional ring and bi-directional ring.
Template security of biometric systems is a vital issue and needs critical focus. The importance lies in the fact that unlike passwords, stolen biometric templates cannot be revoked. Hence, the biometric templates cannot be stored in plain format and needs strong protection against any forgery. In this paper, we present a technique to generate face and palm vein-based fuzzy vault for multi-biometric cryptosystem. Here, initially the input images are pre-processed using various processes to make images fit for further processing. In our proposed method, the features are extracted from the processed hand and palm vein images by finding out unique common points. The chaff points are added to the already extracted points to obtain the combined feature vector. The secret key points which are generated based on the user key input (by using proposed method) are added to the combined feature vector to have the fuzzy vault. For decoding, the multi-modal biometric template from palm vein and hand vein image is constructed and is combined with the stored fuzzy vault to generate the final key. Finally, the experimentation is conducted using the palm vein and hand vein database. The evaluation metrics employed are FMR (False Match Ratio) and GMR (Genuine Match Ratio). From the metric values obtained for the proposed system, we can infer that the system has performed well. The results are also compared to our earlier technique and the results shows that the proposed technique has achieved better results having GAR of 94% without noise and 88% with noise.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.