Joung Woo Ryu

A Collaborative Recommendation Based on Neural Networks

Collaborative filtering is one of the widely used methods for recommendation. It recommends an item to a user based on the reference users’ preferences for the target item or the target user’s preferences for the reference items. In this paper, we propose a neural network based collaborative filtering method. Our method builds a model by learning correlation between users or items using a multi-layer perceptron. We also investigate selection of the reference users or items based on similarity to improve performance. We finally demonstrate that our method outperforms the existing methods through experiments using the EachMovie data.

Optimized fuzzy classification using genetic algorithm

Fuzzy rules are suitable for describing uncertain phenomena and natural for human understanding and they are, in general, efficient for classification. In addition, fuzzy rules allow us to effectively classify data having non-axis-parallel decision boundaries, which is difficult for the conventional attribute-based methods. In this paper, we propose an optimized fuzzy rule generation method for classification both in accuracy and comprehensibility (or rule complexity). We investigate the use of genetic algorithm to determine an optimal set of membership functions for quantitative data. In our method, for a given set of membership functions a fuzzy decision tree is constructed and its accuracy and rule complexity are evaluated, which are combined into the fitness function to be optimized. We have experimented our algorithm with several benchmark data sets. The experiment results show that our method is more efficient in performance and comprehensibility of rules compared with the existing methods including C4.5 and FID3.1 (Fuzzy ID3).

Efficient Fuzzy Rules For Classification

Fuzzy rules are suitable for describing uncertain phenomena and natural for human understanding and they are, in general, efficient for classification. In this paper, we propose an optimized fuzzy rule generation method for classification both in accuracy and comprehensibility (or rule complexity). We investigate the use of genetic algorithm to determine an optimal set of membership functions for quantitative data. In our method, for a given set of membership functions a fuzzy decision tree is constructed and its accuracy and rule complexity are evaluated, which are combined into the fitness function to be optimized. We have experimented our algorithm with several benchmark data sets. The experiment results show that our method is more efficient in performance and compactness of rules compared with the existing methods

Efficiently Maintaining the Performance of an Ensemble Classifier in Streaming Data

In data stream environments, classifiers are generally refined based on regular time interval or fixed number of streaming data. Also, the correct labels of all unlabeled streaming data are typically used in the refine process. Such an approach is not feasible in many real world applications where data labeled by human experts should be used to improve classifiers. In this paper, we select data for refining a classifier from streaming data in an online process. Our selection methodology uses training data, and is applied to build an ensemble of classifiers over streaming data. We compared the results of our ensemble approach and of a conventional ensemble approach where new classifiers for an ensemble are periodically generated. In experiments with ten benchmark data sets including three real streaming data sets, our ensemble approach generated an average of 2.4% classifiers using an average of 10.0% labeled data for the conventional ensemble approach, and produced comparable classification accuracy.

An Efficient Method of Building an Ensemble of Classifiers in Streaming Data

To efficiently refine a classifier in streaming data such as sensor data and web log data we have to decide whether each streaming unlabeled datum is selected or not. The exiting methods refine a classifier based on a regular time interval. They refine a classifier even if the classification accuracy of the classifier is high. Also it uses a classifier even if the classification accuracy is low. In this paper, our ensemble method selects data in an online process that should be labeled. The selected data are used to build new classifiers of an ensemble. Our selection methodology uses training data that are applied to generate an ensemble of classifiers over streaming data. We compared the results of our ensemble approach and of a conventional ensemble approach where new classifiers for an ensemble are periodically generated. In experiments with ten benchmark data sets including three real streaming data sets, our ensemble approach generated 12.9% new classifiers for the chunk-based ensemble approach using partially labeled samples, and used an average of 10% labeled samples for the ten data sets. In all the experiments, our ensemble approach produced comparable classification accuracy. We showed that our approach can efficiently maintain the performance of an ensemble over streaming data.

‘Security Theater’: On the Vulnerability of Classifiers to Exploratory Attacks

The increasing scale and sophistication of cyberattacks has led to the adoption of machine learning based classification techniques, at the core of cybersecurity systems. These techniques promise scale and accuracy, which traditional rule or signature based methods cannot. However, classifiers operating in adversarial domains are vulnerable to evasion attacks by an adversary, who is capable of learning the behavior of the system by employing intelligently crafted probes. Classification accuracy in such domains provides a false sense of security, as detection can easily be evaded by carefully perturbing the input samples. In this paper, a generic data driven framework is presented, to analyze the vulnerability of classification systems to black box probing based attacks. The framework uses an exploration exploitation based strategy, to understand an adversarys point of view of the attack defense cycle. The adversary assumes a black box model of the defenders classifier and can launch indiscriminate attacks on it, without information of the defenders model type, training data or the domain of application. Experimental evaluation on 10 real world datasets demonstrates that even models having high perceived accuracy (>90%), by a defender, can be effectively circumvented with a high evasion rate (>95%, on average). The detailed attack algorithms, adversarial model and empirical evaluation, serve.

An Efficient Coevolutionary Algorithm Using Dynamic Species Control

A coevolutionary algorithm is an extention of the conventional genetic algorithm that incorporates the strategy of divide and conquer in developing a complex solution in the form of interacting co-adapted subcomponents. In this paper we propose an efficient coevolutionary algorithm dynamically controlling species splitting and merging. Our algorithm conducts efficient local search in the reduced search space by splitting species for independent variables while it conducts global search by merging species for interdependent variables. We have experimented the proposed algorithm with some benchmarking function optimization problems and the inventory control problem, and have shown that the algorithm outperforms the existing coevolutionary algorithms.

Speech recognition with multi-modal features based on neural networks

Recent researches have been focusing on fusion of audio and visual features for reliable speech recognition in noisy environments. In this paper, we propose a neural network based model of robust speech recognition by integrating audio, visual, and contextual information. Bimodal Neural Network (BMNN) is a multi-layer perceptron of 4 layers, which combines audio and visual features of speech to compensate loss of audio information caused by noise. In order to improve the accuracy of speech recognition in noisy environments, we also propose a post-processing based on contextual information which are sequential patterns of words spoken by a user. Our experimental results show that our model outperforms any single mode models. Particularly, when we use the contextual information, we can obtain over 90% recognition accuracy even in noisy environments, which is a significant improvement compared with the state of art in speech recognition.

Optimized fuzzy decision tree using genetic algorithm

Fuzzy rules are suitable for describing uncertain phenomena and natural for human understanding and they are, in general, efficient for classification. In addition, fuzzy rules allow us to effectively classify data having non-axis-parallel decision boundaries, which is difficult for the conventional attribute-based methods. In this paper, we propose an optimized fuzzy rule generation method for classification both in accuracy and comprehensibility (or rule complexity). We investigate the use of genetic algorithm to determine an optimal set of membership functions for quantitative data. In our method, for a given set of membership functions a fuzzy decision tree is constructed and its accuracy and rule complexity are evaluated, which are combined into the fitness function to be optimized. We have experimented our algorithm with several benchmark data sets. The experiment results show that our method is more efficient in performance and complexity of rules compared with the existing methods.

Speech recognition by integrating audio, visual and contextual features based on neural networks

Recent researches have been focusing on fusion of audio and visual features for reliable speech recognition in noisy environments. In this paper, we propose a neural network based model of robust speech recognition by integrating audio, visual, and contextual information. Bimodal Neural Network (BMNN) is a multi-layer perceptron of 4 layers, which combines audio and visual features of speech to compensate loss of audio information caused by noise. In order to improve the accuracy of speech recognition in noisy environments, we also propose a post-processing based on contextual information which are sequential patterns of words spoken by a user. Our experimental results show that our model outperforms any single mode models. Particularly, when we use the contextual information, we can obtain over 90% recognition accuracy even in noisy environments, which is a significant improvement compared with the state of art in speech recognition.