Replicator Neural Networks Research Articles

Flow-based intrusion detection on software-defined networks: a multivariate time series anomaly detection approach

In this study, we present and implement the SAnDet (SDN anomaly detector) architecture, an anomaly-based intrusion detection system designed to take advantage of the capabilities offered by software-defined networking (SDN) architecture, as a controller application. The SAnDet system is composed of three modules: statistics collection, anomaly detection, and anomaly prevention. In particular, we utilize replicator neural networks (RNN), which is a specialized variant of the autoencoder, and the LSTM-based encoder–decoder (EncDecAD) method, which is a special type of long short-term memory (LSTM) network that has demonstrated a strong performance on data series particularly, to identify unknown attacks using flow features collected from OpenFlow switches. In our experiments, we utilize flow-based features extracted from network traffic data containing various types of attacks as input to our models in the form of time series. We evaluate the performance of our methods using the accuracy and area under the receiver operating characteristic curve (AUC) metrics. Our experimental results demonstrate that EncDecAD outperforms RNN and that our approach offers several benefits over previously conducted research.

Exploring the research landscape of data warehousing and mining based on DaWaK Conference full-text articles

The international conference on Data Warehousing and Knowledge Discovery (DaWaK) has become a pivotal place to exchange experiences and knowledge among researchers and practitioners in big data analytics. The conference has been essential to data warehousing and data analytics for the last 21 years (1999–2019). This study explored the knowledge structure embedded in the DaWaK Conference papers and examined the research trends over time. It also analyzed the performance of published papers, authors, and their affiliations and countries and visualized a collaboration network in DaWaK. We applied several text mining techniques, including co-word analysis, topic modeling, co-author network analysis, and network visualization. The study’s findings indicate that the core topics are data mining techniques, algorithm performance, and information systems. The popular topic trends are associated with database encryption, whereas the topics related to online analytical processing (OLAP) technology are in decline. The research metrics results demonstrate that the DaWaK papers were cited 6,262 times, with an h-index of 34 for the 722 DaWaK papers. The article titled “Outlier Detection Using Replicator Neural Networks” reached the most citations (177), and the most productive author was Bellatreche, Ladjel (15 papers). Nanyang Technological University is the most frequently mentioned as the author’s affiliation, the United States is the country with the largest number of authors, and the National Science Foundation was the largest funding agency that supported the DaWaK researchers. Moreover, the authorship network of Bellatreche, Ladjel is the largest collaboration network in the DaWaK scholar community. The outcomes of this study would be beneficial for comprehending the knowledge in data warehousing and the relevant cross-disciplinary areas of research and collaboration networks in this field.

Variance error of multi-classification based anomaly detection for time series data

For abnormal detection of time series data, the supervised anomaly detection methods require labeled data. While the range of outlier factors used by the existing semi-supervised methods varies with data, model and time, the threshold for determining abnormality is difficult to obtain, in addition, the computational cost of the way to calculate outlier factors from other data points in the data set is also very large. These make such methods difficult to practically apply. This paper proposes a framework named LSTM-VE which uses clustering combined with visualization method to roughly label normal data, and then uses the normal data to train long short-term memory (LSTM) neural network for semi-supervised anomaly detection. The variance error (VE) of the normal data category classification probability sequence is used as outlier factor. The framework enables anomaly detection based on deep learning to be practically applied and using VE avoids the shortcomings of existing outlier factors and gains a better performance. In addition, the framework is easy to expand because the LSTM neural network can be replaced with other classification models. Experiments on the labeled and real unlabeled data sets prove that the framework is better than replicator neural networks with reconstruction error (RNN-RS) and has good scalability as well as practicability.

Case Study: Models for Detecting Low Oil Pressure Anomalies on Commercial Vehicles

We present a case study of anomaly detection using commercial vehicle data (from a single vehicle collected over a six-month interval) and propose a failure-event analysis. Our analysis allows performance comparison of anomaly detection models in the absence of sufficient anomalies to compute the Receiver Operating Characteristic curve. Several heuristically-guided data-driven models were considered to capture the relationship among three main engine signals (oil pressure, temperature, and speed). These models include regression-based approaches and distance-based approaches; the former use the residual’s z-score as the detection metric, while the latter use a Mahalanobis distance or similar measure as the metric. The selected regression-based models (Boosted Regression Trees, Feed-Forward Neural Networks, and Gridded Regression tables) outperformed the selected distance-based approaches (Gaussian Mixtures and Replicator Neural Networks). Both groups of models were superior to existing Diagnostic Trouble Codes. The Gridded Regression tables and Boosted Regression Trees exhibited the best overall metric performance. We report a surprising behavior of one of the models: locally- optimal Gaussian Mixture Models often had zero detection performance, with such models occurring in at least 25% of the iterations with seven or more Gaussians in the mixture. To overcome the problem, we propose a regularization method that employs a heuristic filter for rejecting Gaussian Mixtures with non-discriminative components.

Generating multiple spectrum compatible accelerograms using stochastic neural networks

AbstractA new neural‐network‐based methodology for generating artificial earthquake spectrum compatible accelerograms from response spectra was proposed in 1997, in which, the learning capabilities of neural networks were used to develop the knowledge of the inverse mapping from the response spectra to earthquake accelerograms. Recently, this methodology has been further extended and enhanced. This paper presents a new stochastic neural network that is capable of generating multiple earthquake accelerograms from a single‐response spectrum. A new stochastic feature to the neural network has been combined with a new scheme for data compression using the replicator neural networks developed in the original method. A benefit of this extended methodology is gaining efficiency in compressing the earthquake accelerograms and extracting their characteristics. The proposed method produces a stochastic ensemble of earthquake accelerograms from any response spectra or design spectra. An example is presented that used 100 recorded accelerograms to train the neural network and several design spectra and response spectra to test this improved methodology. Copyright © 2001 John Wiley & Sons, Ltd.

Replicator Neural Networks for Universal Optimal Source Coding

Replicator neural networks self-organize by using their inputs as desired outputs; they internally form a compressed representation for the input data. A theorem shows that a class of replicator networks can, through the minimization of mean squared reconstruction error (for instance, by training on raw data examples), carry out optimal data compression for arbitrary data vector sources. Data manifolds, a new general model of data sources, are then introduced and a second theorem shows that, in a practically important limiting case, optimal-compression replicator networks operate by creating an essentially unique natural coordinate system for the manifold.