Recently, different deep learning based techniques have been proposed to detect intrusions and anomalies in the information systems. Convolutional neural networks are often used to reveal hidden spatial relations between features, however their application requires certain data preprocessing techniques that transform tabular non-spatial data to matrices. This paper studies different approaches to tabular data transformation to images and analyzes their impact on the efficiency of attack detection including the ability to detect novel and unseen attacks. Experiments are conducted using the CICIDS2017 dataset, which describes network traffic flows as numerical vectors and contains different types of attacks. The conducted research allowed us to conclude on areas of applicability of tabular data transformation to images considering advantages and limitations of such preprocessing.
Demanding data processing on small or medium sized mobile robots is used in many research projects and starts distributing in more and more applications. This is also the case for agricultural robotics where previously single heavy machines were used. Several recent agricultural robotic systems (in research and in industry) (1) are used outdoors and need to be energy efficient, (2) have to perform complex tasks, e.g., using machine learning (ML) computation like Convolutional Neural Network (CNN), also on high-dimensional multi-spectral image data, and (3) are used in multiple small distributed systems rather than single large vehicles. These properties lead to the contradicting demands for the computation hardware of high computational power and low power consumption and small size. Previous work has shown that using Field Programmable Gate Arrays (FPGA) can lead to a higher energy efficiency compared to standard, e.g., CPU-based ML computation. Furthermore, using a hybrid FPGA with CPU cores and programmable user logic in the same device can lead to a one-chip solution for mobile systems or IoT devices. Presented in this work in progress paper are first results of FPGA-based implementations of ML (CNN) inference on RGB and multi-spectral image data, as part of an agricultural mobile robot research project.
The increase in connectivity capabilities, resources, and data availability, has undoubtedly brought many advantages in gaining access to services quickly, but it also made possible numerous and sophisticated cybersecurity attacks affecting nowadays companies, national infrastructures, organizations, and, ultimately, users across the globe. Some cyberattacks, namely the zero-day attacks are difficult to counter, because by the time such attacks are discovered and countermeasures are implemented and deployed, other unknown attack variants might occur. Thus, in recent years, anomaly-based Intrusion Detection Systems (IDS) using machine learning (ML) and deep learning (DL) techniques have been proposed to mitigate such attacks, namely the “unknown” attacks. An anomaly-based IDS performs traffic analysis by exploiting supervised or unsupervised ML and DL algorithms and raises alerts if a suspicious pattern is encountered. In this paper, we use an anomaly-based security attack detection model exploiting the unsupervised ML autoencoder model to detect variants of the Heartbleed attack affecting the famous Transport Layer Security (TLS) protocol. By using the CIC-IDS2017 dataset and a custom Heartbleed dataset, we evaluate our model for detecting the Heartbleed attack. The results are encouraging, since the proposed autoencoder-based model recognizes Heartbleed TLS anomalies and distinguishes them from the benign traffic in 85
Our overall research topics are self-organising multi-agent systems (MAS). Here, we study how to design MAS architectures, which implement adaptivity using the MAPE pattern (monitor, analyse, plan, execute). We investigate the approach of organisation-oriented MAS (Org-MAS), where one considers large-scale and agent-independent aspects of MAS, like roles, norms, positions, interaction protocols etc. When considering adaptivity in the context of Org-MAS we are confronted with two perspectives: On the one side we have the micro-level, where the agents adapt their behaviour; and on the other side we have the macro-level, where the organisation changes the network of positions, role obligations etc. We like to point out that these two perspectives are not independent from each other. This mutual adaption of the agents and the system’s structure is known as the micro-macro link. We consider the adequate modelling of the micro-macro link as an essential precondition for the design of MAPE-loops in Org-MAS since due to this mutual dependencies, the Org-MAS is more than the sum of its parts and a model that does not take them into account (or even relies one perspective only) will provide less accurate predictions about adaption effects. So, our research challenge is (1) to integrate these two perspectives – micro and macro – into the MAPE-loop with equal rights and (2) to take the micro-macro link intoi account. Here, we define the Sonar-MAPE-loop, i.e. a design for a micro-macro aware adaption.
Privacy is a well-researched area in the context of Federated Learning. Typically, ensuring privacy means that individual data used for local training cannot be reconstructed by other local learners or a central server. Thus, it is the individual data points that should be private, but not the entire distribution of locally available data. In many cases, this makes sense because each data point comes from a different individual while all data points originate from a common global distribution. In this position paper, we address a more challenging task where the privacy of each local data distribution must be preserved. This is relevant for use cases where there is a one-to-one mapping from local learners to users, such as when each local learner is part of a personalized assistant on a smartphone. We provide a definition of this problem case, describe the challenges that need to be addressed, and formulate a possible approach to solve the problem.
