Casting out Demons: Sanitizing Training Data for Anomaly Sensors

By Gabriela F. Cretu-Ciocarlie, Angelos Stavrou, Michael E. Locasto, Salvatore J. Stolfo, Angelos Dennis Keromytis. 2008

257 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Anomaly & Outlier Detection, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Networks and Communications

In Anomaly & Outlier Detection

In Signal Processing

Citations in the corpus

Advances and Open Problems in Federated Learning (2019, methodology): “scuss this in more detail in at the end of Section 5.1.3. Data sanitization and network pruning Defenses designed speciﬁcally for data poisoning attacks frequently rely on “data sanitization” methods [123], which aim to remove poisoned or otherwise anomalous 65 data. More recent work has developed i”
Local Model Poisoning Attacks to Byzantine-Robust Federated Learning (2019, background): “One category of defenses [4, 15, 56, 59] detects malicious data based on their (negative) impact on the performance of the learnt model.”
Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning (2017, background): “, attack detection and removal) [111, 116, 118], and robust learning (i.”
Attack of the Tails: Yes, You Really Can Backdoor Federated Learning (2020, background): “Typical defenses against data poisoning attacks involve Data Sanitization [46] which uses outlier detection [47], however Koh et al. [34] show that these defenses can be overcome.”
Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural Networks (2019, background): “Defenses against poisoning attack mostly focus on sanitizing the training set and removing poisoned samples [32], [33], [34], [35], [36], [37].”
Stronger data poisoning attacks break data sanitization defenses (2018, background): “nicantly dierent decision boundary by an attacker who injects just a small amount of poisoned data (Figure 1-Middle). A common defense against data poisoning attacks are data sanitization defenses (Cretu et al., 2008), which use anomaly detectors to lter out training points that look suspicious (see”
Certified Defenses for Data Poisoning Attacks (2017, methodology): “In this paper, we consider data sanitization defenses (Cretu et al., 2008), which examine the full dataset and try to remove the poisoned points, for example by deleting outliers.”
Manipulating Machine Learning: Poisoning Attacks and Countermeasures for Regression Learning (2018, background): “[12] discuss sanitization methods against timebased anomaly detectors in which multiple micro-models are built and compared over time to identify poisoned data.”
When Does Machine Learning FAIL? Generalized Transferability for Evasion and Poisoning Attacks (2018, background): “The Micromodels defense was proposed for cleaning training data for network intrusion detectors [13].”
Security Evaluation of Pattern Classifiers under Attack (2014, methodology): “, the data sanitization method of [27]).”
Adversarial Feature Selection Against Evasion Attacks (2016, background): “, a form of outlier detection) [6], [32], multiple classifier systems [33], and robust statistics [26].”
Security analysis of online centroid anomaly detection (2010, background): “Another related technique is sanitization proposed by Cretu et al. (2008). It deals with the general problem of cleaning the training data from potential attacks.”
Support vector machines under adversarial label contamination (2015, background): “, a form of outlier detection) [5, 6, 28], multiple classifier systems [29], and robust statistics [7].”
Towards Making Systems Forget with Machine Unlearning (2015, background): “[34] introduced a sanitization phase in the machine learning process to filter polluted data.”
Design of robust classifiers for adversarial environments (2011, background): “we point out that less works have been devoted to causative attacks, which are less common than exploratory ones in real applications (see, for instance [4], [15], [16]).”
When a Tree Falls: Using Diversity in Ensemble Classifiers to Identify Evasion in Malware Detectors (2016, background): “Many studies have addressed the importance of data sanitization or adversarial influence at training time [5], [13], [23], [30].”
Efficient Repair of Polluted Machine Learning Systems via Causal Unlearning (2018, background): “tential polluted emails [17, 34, 39] and make the model robust.”
Latent Backdoor Attacks on Deep Neural Networks (2019, background): “Defenses against poisoning attacks mostly focus on sanitizing training data and removing poisoned samples [6, 15, 21, 36, 44, 47].”
Resilient Linear Classification: An Approach to Deal with Attacks on Training Data (2017, unknown): “…for defending against e.g., commonly required for fault tolerant systems , are challenging to Moreover, adversarial learning literature ( e.g., [9, 17, 19]) data-driven CPS, it is necessary to consider techniques for resilient machine learning that can defend against cyber-physical attacks”
Summoning Demons: The Pursuit of Exploitable Bugs in Machine Learning (2017, background): “Cretu et al. [7] discuss the process and importance of “cas-ting out demons,” sanitizing ML training datasets for anomaly detection (AD) sensors.”