{"title":"On Improving the Effectiveness of Adversarial Training","authors":"Yi Qin, Ryan Hunt, Chuan Yue","doi":"10.1145/3309182.3309190","DOIUrl":"https://doi.org/10.1145/3309182.3309190","url":null,"abstract":"Machine learning models, including neural networks, are vulnerable to adversarial examples, which are adversarial inputs generated from legitimate examples by applying small perturbations to fool machine learning models to misclassify. Algorithms that are used to generate adversarial examples are called adversarial example generation methods. As the state-of-the-art defense approach, adversarial training improves the robustness of machine learning models by augmenting the training data with adversarial examples. However, adversarial training is far from being perfect yet, and a deeper understanding of it is always needed for further improving its effectiveness. In this paper, we propose to investigate two research questions. The first question is: whether Method-Based Ensemble Adversarial Training (MBEAT) could be beneficial, i.e., whether leveraging the adversarial examples generated by multiple methods could help increase the effectiveness of adversarial training. The second question is: whether Round Gap Of Adversarial Training (RGOAT) could exist, i.e., whether a neural network model adversarially trained in one round would not be robust against the adversarial examples further generated from this model itself. We design an adversarial training experimental framework to answer these two research questions. We find that MBEAT is indeed beneficial, indicating that it has some important value in practice. We also find that RGOAT indeed exists, indicating that adversarial training should be an iterative process.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"125 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115038971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Computing the Identification Capability of SQL Queries for Privacy Comparison","authors":"Muhammad Imran Khan, S. Foley, B. O’Sullivan","doi":"10.1145/3309182.3309188","DOIUrl":"https://doi.org/10.1145/3309182.3309188","url":null,"abstract":"Comparing SQL queries is an interesting area of research, with applications of query similarity in different domains; for instance, anomaly-based intrusion detection systems. Comparing two SQL queries in terms of privacy - privacy comparison as well as computing a quantitative value for the identification capability of an SQL query is desirable. In this paper, we compute the identification capability of an SQL query and subsequently we propose an approach to compare two SQL queries in terms of privacy by introducing the notion of privacy equivalence, less-private and more-private relations. Additionally, an edge-labelled directed acyclic graph style privacy-aware attribute relationship diagram is proposed that facilitates the privacy comparison.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122898382","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Spears Against Shields: Are Defenders Winning the Phishing War?","authors":"Ayman El Aassal, Rakesh M. Verma","doi":"10.1145/3309182.3309191","DOIUrl":"https://doi.org/10.1145/3309182.3309191","url":null,"abstract":"Phishing is the act of using deceptive methods to lure users into taking harmful and dangerous actions against themselves and/or the company they work for. Hackers have been using this method to bypass various security systems, steal personal information, make sensitive data public, and all of this while raking millions of dollars. On the attack side, hackers are using phishing kits and employ all possible phishing techniques to build a successful campaign. On the defense side, we find spam/phishing filters and malicious website detection in addition to anti-phishing simulation training. In this study, we assess the current phishing landscape by testing the tools used on both sides of the war. We generated 1,000 phishing emails containing phishing links semi-automatically using natural language generation technology and tested five popular tools with anti-phishing modules. The number of undetected emails ranged from 77 to 927. We also evaluate several anti-phishing training technologies and reveal their shortcomings. Our results suggest that both anti-phishing filters and current training tools have a long way to go and thus improving these defense mechanisms is still essential.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122435045","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Dark Web Content Analysis and Visualization","authors":"Sugiu Takaaki, Inomata Atsuo","doi":"10.1145/3309182.3309189","DOIUrl":"https://doi.org/10.1145/3309182.3309189","url":null,"abstract":"The Dark Web, which is a vast array of encrypted online content and websites that can be only be accessed by the use of anonymizing tools such as The Onion Router (Tor), is currently a topic of serious concern because it has become a growth area for clandestine entities catering to forbidden activities and services such as illegal drugs and weapons, child pornography, sensitive information such as stolen credit card information, distributed denial of services (DDoS) tools such as Booter, and other antisocial activities. As a result, it is now being investigated by researchers, law enforcement agencies, security companies, and academic institutions around the world. However, there is not an efficient investigation method and it takes a lot of time. In an effort to gain a better understanding of the Dark Web, we analyzed a large amount of Onion domain name data obtained from using the \"Ichidan\" search engine and the \"Fresh Onions\" open source Tor. More specifically, we collected everything we could gather into an acquired Onion domain and then downloaded the top page. The Onion domain was then classified into six categories, including classification not possible, and a new domain was created from the downloaded top page text. Additionally, we implemented a simulator that is a directed graph created from the uncovered hyperlinks and connection states to the Dark Web were achieved with simplicity. We then attempted to determine the relationships and characteristics of each instance Dark Web content by reflecting on the graphed classification results.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"56 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125913484","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Preserving Privacy in Data Analytics","authors":"Lila Ghemri","doi":"10.1145/3309182.3311786","DOIUrl":"https://doi.org/10.1145/3309182.3311786","url":null,"abstract":"Data Analytics is becoming an essential business tool for many data intensive companies and organizations. However, the increased use of such methods comes with the threat of data disclosure. Privacy-preserving methods have been developed with varying degrees of efficiency with the main goal of protecting individuals' privacy. This tutorial aims at presenting models and techniques of preserving privacy in machine learning and data mining.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"46 2","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132546013","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Efficient k-means Using Triangle Inequality on Spark for Cyber Security Analytics","authors":"Ambika Shrestha Chitrakar, Slobodan V. Petrovic","doi":"10.1145/3309182.3309187","DOIUrl":"https://doi.org/10.1145/3309182.3309187","url":null,"abstract":"With the advancement in technology and the increase in the number of digital sources, data quantity increases every day and, consequently, the cyber security related data quantity. Traditional security systems such as Intrusion Detection Systems (IDS) are not capable of handling such a growing amount of data set in real time. Cyber security analytics is an alternative solution to such traditional security systems, which can use big data analytics techniques to provide a faster and scalable framework to handle a large amount of cyber security related data in real time. k-means clustering is one of the commonly used clustering algorithms in cyber security analytics aimed at dividing security related data into groups of similar entities, which in turn can help in gaining important insights about the known and unknown attack patterns. This technique helps a security analyst to focus on the data specific to some clusters only for the analysis. To improve performance, k-means can exploit the triangle inequality to skip many point-center distance computations, without affecting the clustering results. In this paper, we re-formulate the parallel version of Elkan's k-means with triangle inequality (k-meansTI) algorithm, implement this algorithm on Apache Spark, and use it to classify Web attacks in different clusters. The paper also provides the speed comparison of our parallel k-meansTI on Spark with the Spark ML k-means clustering algorithm.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130308675","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Clustering for Security Challenges","authors":"Rakesh M. Verma, Srivathsan Srinivasagopalan","doi":"10.1145/3309182.3309184","DOIUrl":"https://doi.org/10.1145/3309182.3309184","url":null,"abstract":"Data science, which includes statistics, probability, data mining, machine learning and natural language processing, is on a major uptrend. Clustering is a key data science method, which does not require labeled data. This tutorial will cover clustering for security challenges. We expect to cover both classical clustering techniques and some of the important, newer clustering algorithms. Examples will be drawn from clustering of malware datasets and phishing emails. Throughout the discussion, we will highlight some of the key unique needs of the security domain that are relevant for clustering techniques.","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"163 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115154543","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","authors":"","doi":"10.1145/3309182","DOIUrl":"https://doi.org/10.1145/3309182","url":null,"abstract":"","PeriodicalId":424378,"journal":{"name":"Proceedings of the ACM International Workshop on Security and Privacy Analytics","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124611410","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}