Abstract Federated learning enables multiple users to build a joint model by sharing their model updates (gradients), while their raw data remains local on their devices. In contrast to the common belief that this provides privacy benefits, we here add to the very recent results on privacy risks when sharing gradients. Specifically, we investigate Label Leakage from Gradients (LLG), a novel attack to extract the labels of the users’ training data from their shared gradients. The attack exploits the direction and magnitude of gradients to determine the presence or absence of any label. LLG is simple yet effective, capable of leaking potential sensitive information represented by labels, and scales well to arbitrary batch sizes and multiple classes. We mathematically and empirically demonstrate the validity of the attack under different settings. Moreover, empirical results show that LLG successfully extracts labels with high accuracy at the early stages of model training. We also discuss different defense mechanisms against such leakage. Our findings suggest that gradient compression is a practical technique to mitigate the attack.

Most research in the field of network intrusion detection heavily relies on datasets. Datasets in this field, however, are scarce and difficult to reproduce. To compare, evaluate, and test related work, researchers usually need the same datasets or at least datasets with similar characteristics as the ones used in related work. In this work, we present concepts and the Intrusion Detection Dataset Toolkit (ID2T) to alleviate the problem of reproducing datasets with desired characteristics to enable an accurate replication of scientific results. Intrusion Detection Dataset Toolkit (ID2T) facilitates the creation of labeled datasets by injecting synthetic attacks into background traffic. The injected synthetic attacks created by ID2T blend with the background traffic by mimicking the background traffic’s properties. This article has three core contributions. First, we present a comprehensive survey on intrusion detection datasets. In the survey, we propose a classification to group the negative qualities found in the datasets. Second, the architecture of ID2T is revised, improved, and expanded in comparison to previous work. The architectural changes enable ID2T to inject recent and advanced attacks, such as the EternalBlue exploit or a peer-to-peer botnet. ID2T’s functionality provides a set of tests, known as TIDED, that helps identify potential defects in the background traffic into which attacks are injected. Third, we illustrate how ID2T is used in different use-case scenarios to replicate scientific results with the help of reproducible datasets. ID2T is open source software and is made available to the community to expand its arsenal of attacks and capabilities.

The process of designing and evaluating distributed Cyber-Physical Systems (CPSs) is not a trivial task. There are many challenges to tackle such as managing distributed resources, enabling communication between components, and choosing performance metrics to evaluate the “goodness” of the system. Smart Grids (SGs) are prominent representatives of CPSs, a particular type of Critical Infrastructure (CI), whose organizational model is becoming more distributed and dynamic. Due to this paradigm shift, new control and management mechanisms need to be identified and tested to guarantee uninterrupted operation. However, novel approaches cannot always be tested against real networks as the economic cost and risk can be high. In contrast, modeling and simulation techniques are viable evaluation mechanisms that support the continuous evolution of CIs. In this paper, we present an Open Source time-discrete simulation software, called HOLEG, that models and evaluates SGs. The software is based on the Holon analogy, a bio-inspired approach that enables systems resilience through flexible reconfiguration mechanisms. The presented software provides features that enable the integration and execution of optimization algorithms along with their evaluation. To demonstrate HOLEG, a case study is presented where a heuristic algorithm is implemented to minimize wasted energy while preventing network destabilization.

In distributed storage systems, documents are shared among multiple Cloud providers and stored within their respective storage servers. In social secret sharing-based distributed storage systems, shares of the documents are allocated according to the trustworthiness of the storage servers. This paper proposes a trust mechanism using machine learning techniques to compute evidence-based trust values. Our mechanism mitigates the effect of colluding storage servers. More precisely, it becomes possible to detect unreliable evidence and establish countermeasures in order to discourage the collusion of storage servers. Furthermore, this trust mechanism is applied to the social secret sharing protocol AS^3, showing that this new evidence-based trust mechanism enhances the protection of the stored documents.

Intrusion Detection Systems (IDSs) are an important defense tool against the sophisticated and ever-growing network attacks. With this in mind, the research community has been immersed in the field of IDSs over the past years more than before. Still, assessing and comparing performance between different systems and algorithms remains one of the biggest challenges in this research area. IDSs need to be evaluated and compared against high quality datasets; nevertheless, the existing ones have become outdated or lack many essential requirements. We present the Intrusion Detection Dataset Toolkit (ID2T), an approach for creating out-of-the-box labeled datasets that contain user defined attacks. In this paper, we discuss the essential requirements needed to create synthetic, yet realistic, datasets with user defined attacks. We also present typical problems found in synthetic datasets and propose a software architecture for building tools that can cope with the most typical problems. A publicly available prototype, is implemented and evaluated. The evaluation comprises a performance analysis and a quality assessment of the generated datasets. We show that our tool can handle large amounts of network traffic and that it can generate synthetic datasets without the problems or shortcomings we identified in other datasets.

New attack surfaces are emerging with the rise of Industrial Control System (ICS) devices exposed on the Internet. ICS devices must be protected in a holistic and efficient manner; especially when these are supporting critical infrastructure. Taking this issue into account, cyber-security research is recently being focused on providing early detection and warning mechanisms for ICSs. In this paper we present a novel honeypot capable of detecting multi-stage attacks targeting ICS networks. Upon detecting a multi-stage attack, our honeypot can generate signatures so that misuse Intrusion Detection Systems (IDSs) can subsequently thwart attacks of the same type. Our experimental results indicate that our honeypot and the signatures it generates provide good detection accuracy and that the Bro IDS can successfully use the signatures to prevent future attacks.

Defending key network infrastructure, such as Internet backbone links or the communication channels of critical infrastructure, is paramount, yet challenging. The inherently complex nature and quantity of network data impedes detecting attacks in real world settings. In this paper, we utilize features of network flows, characterized by their entropy, together with an extended version of the original Replicator Neural Network (RNN) and deep learning techniques to learn models of normality. This combination allows us to apply anomaly-based intrusion detection on arbitrarily large amounts of data and, consequently, large networks. Our approach is unsupervised and requires no labeled data. It also accurately detects network-wide anomalies without presuming that the training data is completely free of attacks. The evaluation of our intrusion detection method, on top of real network data, indicates that it can accurately detect resource exhaustion attacks and network profiling techniques of varying intensities. The developed method is efficient because a normality model can be learned by training an RNN within a few seconds only.

Due to the increasing quantity and sophistication of cyber-attacks, Intrusion Detection Systems (IDSs) are nowadays considered mandatory security mechanisms for protecting critical networks. Research on cyber-security is moving from such isolated IDSs towards Collaborative IDSs (CIDSs) in order to protect large-scale networks. In CIDSs, a number of IDS sensors work together for creating a holistic picture of the monitored network. Our contribution in this paper is a novel distributed and scalable CIDS, called SkipMon. Our system supports, both, the idea of locality and privacy preserving communication by means of exchanging compact alert data. Furthermore, we propose a mechanism for interconnecting sensors that experience similar traffic patterns. The experimental results suggest that our CIDS, with our technique of connecting monitoring nodes that experience similar traffic, is scalable and offers a good accuracy rate compared to a centralized system with full knowledge of the participating sensors' data.