Anomaly Detection Using Autoencoders in High Performance Computing Systems

Borghesi, Andrea; Bartolini, Andrea; Lombardi, Michele; Milano, Michela; Benini, Luca

doi:10.1609/aaai.v33i01.33019428

Cited by 160 publications

(107 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Moreover, in [25], [26], [27] several Convolutional Neural Network (CNNs) anomaly detection techniques were proposed. Additionally, several deep learning formulation for network anomaly detection employ the intelligent scheme of Stacked Autoencoders (SAE) [28], [29], [30]. Despite the fact that deep learning architectures are demanding, since a great amount of data and internal iterations are required, these methods usually outperform traditional supervised and unsupervised intrusion detection approaches.…”

Section: B Network Traffic Anomaly Detectionmentioning

confidence: 99%

Incidents Information Sharing Platform for Distributed Attack Detection

Fotiadou¹,

Velivassaki

Voulkidis³

et al. 2020

IEEE Open J. Commun. Soc.

View full text Add to dashboard Cite

Intrusion detection plays a critical role in cyber-security domain since malicious attacks cause irreparable damages to cyber-systems. In this work, we propose the I2SP prototype, which is a novel Information Sharing Platform, able to gather, pre-process, model, and distribute network-traffic information. Within the I2SP prototype we build several challenging deep feature learning models for network-traffic intrusion detection. The learnt representations will be utilized for classifying each new network measurement into its corresponding threat level. We evaluate our prototype's performance by conducting case studies using cyber-security data extracted from the Malware Information Sharing Platform (MISP)-API. To the best of our knowledge, we are the first that combine the MISP-API in order to construct an information sharing mechanism that supports multiple novel deep feature learning architectures for intrusion detection. Experimental results justify that the proposed deep feature learning techniques are able to predict accurately MISP threat-levels. INDEX TERMS Malware information sharing platform, network intrusion detection, anomaly detection, deep feature learning, convolutional neural networks, long-short memory neural networks, stacked-sparse autoencoders.

show abstract

Section: B Network Traffic Anomaly Detectionmentioning

confidence: 99%

Incidents Information Sharing Platform for Distributed Attack Detection

Fotiadou¹,

Velivassaki

Voulkidis³

et al. 2020

IEEE Open J. Commun. Soc.

View full text Add to dashboard Cite

show abstract

“…Given a set of features, K-means aims to partition data points into K-clusters. The underlying rationale is that each data point in a cluster is more similar to points of its own cluster than to points from other clusters [38].  Deep learning: these algorithms reproduce the structure and the functions of biological neural networks [39].…”

Section: A Machine Learning and Its Applicationsmentioning

confidence: 99%

Cyberbullying Detection Through Machine Learning: Can Technology Help to Prevent Internet Bullying?

Angelis*¹,

Perasso²

2020

IJMH

View full text Add to dashboard Cite

Cyberbullying constitutes a threat to adolescents’ psychosocial wellbeing that developed alongside technological progress. Detecting online bullying cases is still an issue because most of victims and bystanders do not timely report cyberbullying episodes to adults. Therefore, automatized technologies may play a critical role in detecting cyberbullying through the use of Machine Learning (ML). ML covers a broad range of techniques that enables systems to quickly access and learn from data, and to make decisions about complex problems. This contribution aims at deepening the role of ML in cyberbullying detection and prevention. Specifically, the following issues are addressed: i. identifying the features most frequently considered to develop ML models predicting cyberbullying; ii. identifying the most used ML algorithms and their evaluation methods; iii. understanding the implication of ML for prevention; iv. highlighting the main theoretical and methodological issues of ML algorithms in predicting cyberbullying. To answer these research questions, a systematic review of literature reviews, from a total of n=186 records from online databanks, has been conducted. Ten literature reviews have been elected to analyze and discuss evidence about ML preventative potential against cyberbullying. Most of the models used content-based features to predict cyberbullying. The majority of these features includes words written in social network posts, whereas Support Vector Machine, Naïve Bayes, and Convolutional Neural Networks are the most used alghorithms. Methodological and technical issues have been critically discussed. ML represents an innovative preventative strategy that may optimize and integrate educational programs for adolescents and be the starting point of the development of technology-based automatized detection strategies. Future research is challenged to develop algorithms capable of detecting cyberbullying from several multimedia sources.

show abstract

“…To minimize the requirement of prior knowledge, machine learning techniques, especially autoencoder-based architectures, are widely applied to equipment AD for machines in various domains. Borghesi et al [ 17 ] adopted a set of general autoencoders to learn normal behaviors of high performance computing systems and exploits learned autoencoders to identify abnormal behaviors. Oh and Yun [ 18 ] specifically used the residual error of a general autoencoder to identify the anomaly in a surface-mounted device (SMD) machine.…”

Section: Introductionmentioning

confidence: 99%

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Chen

Chang

Liao³

2020

Sensors

View full text Add to dashboard Cite

In-line anomaly detection (AD) not only identifies the needs for semiconductor equipment maintenance but also indicates potential line yield problems. Prompt AD based on available equipment sensory data (ESD) facilitates proactive yield and operations management. However, ESD items are highly diversified and drastically scale up along with the increased use of sensors. Even veteran engineers lack knowledge about ESD items for automated AD. This paper presents a novel Spectral and Time Autoencoder Learning for Anomaly Detection (STALAD) framework. The design consists of four innovations: (1) identification of cycle series and spectral transformation (CSST) from ESD, (2) unsupervised learning from CSST of ESD by exploiting Stacked AutoEncoders, (3) hypothesis test for AD based on the difference between the learned normal data and the tested sample data, (4) dynamic procedure control enabling periodic and parallel learning and testing. Applications to ESD of an HDP-CVD tool demonstrate that STALAD learns normality without engineers’ prior knowledge, is tolerant to some abnormal data in training input, performs correct AD, and is efficient and adaptive for fab applications. Complementary to the current practice of using control wafer monitoring for AD, STALAD may facilitate early detection of equipment anomaly and assessment of impacts to process quality.

show abstract

Anomaly Detection Using Autoencoders in High Performance Computing Systems

Cited by 160 publications

References 12 publications

Incidents Information Sharing Platform for Distributed Attack Detection

Incidents Information Sharing Platform for Distributed Attack Detection

Cyberbullying Detection Through Machine Learning: Can Technology Help to Prevent Internet Bullying?

Equipment Anomaly Detection for Semiconductor Manufacturing by Exploiting Unsupervised Learning from Sensory Data

Contact Info

Product

Resources

About