Machine learning has achieved state-of-the-art results in network intrusion detection; however, its performance significantly degrades when confronted by a new attack class— a zero-day attack. In simple terms, classical machine learning-based approaches are adept at identifying attack classes on which they have been previously trained, but struggle with those not included in their training data. One approach to addressing this shortcoming is to utilise anomaly detectors which train exclusively on benign data with the goal of generalising to all attack classes— both known and zero-day. However, this comes at the expense of a prohibitively high false positive rate. This work proposes a novel contrastive loss function which is able to maintain the advantages of other contrastive learning-based approaches (robustness to imbalanced data) but can also generalise to zero-day attacks. Unlike anomaly detectors, this model learns the distributions of benign traffic using both benign and known malign samples, i.e. other well-known attack classes (not including the zero-day class), and consequently, achieves significant performance improvements. The proposed approach is experimentally verified on the Lycos2017 dataset where it achieves an AUROC improvement of.000065 and.060883 over previous models in known and zero-day attack detection, respectively. Finally, the proposed method is extended to open-set recognition achieving OpenAUC improvements of.170883 over existing approaches.The implementation and experiments are open-sourced and available at: https://github.com/jackwilkie/CLOSR

Optimizing images as output of a neural network has been shown to introduce a powerful prior for image inverse tasks, capable of producing solutions of reasonable quality in a fully internal learning context, where no external datasets are involved. Two potential technical approaches involve fitting a coordinate-based Multilayer Perceptron (MLP), or a Convolutional Neural Network to produce the result image as output. The aim of this work is to evaluate the two counterparts, as well as a new framework proposed here, named Neural Knitwork, which maps pixel coordinates to local texture patches rather than singular pixel values. The utility of the proposed technique is demonstrated on the tasks of image inpainting, super-resolution, and denoising. It is shown that the Neural Knitwork can outperform the standard coordinate-based MLP baseline for the tasks of inpainting and denoising, and perform comparably for the super-resolution task.

Abstract The use of supervised Machine Learning (ML) to enhance Intrusion Detection Systems (IDS) has been the subject of significant research. Supervised ML is based upon learning by example, demanding significant volumes of representative instances for effective training and the need to retrain the model for every unseen cyber-attack class. However, retraining the models in-situ renders the network susceptible to attacks owing to the time-window required to acquire a sufficient volume of data. Although anomaly detection systems provide a coarse-grained defence against unseen attacks, these approaches are significantly less accurate and suffer from high false-positive rates. Here, a complementary approach referred to as “One-Shot Learning”, whereby a limited number of examples of a new attack-class is used to identify a new attack-class (out of many) is detailed. The model grants a new cyber-attack classification opportunity for classes that were not seen during training without retraining. A Siamese Network is trained to differentiate between classes based on pairs similarities, rather than features, allowing to identify new and previously unseen attacks. The performance of a pre-trained model to classify new attack-classes based only on one example is evaluated using three mainstream IDS datasets; CICIDS2017, NSL-KDD, and KDD Cup’99. The results confirm the adaptability of the model in classifying unseen attacks and the trade-off between performance and the need for distinctive class representations.

In critical infrastructure applications, timely and consistent fault detection and diagnosis is an increasingly important operational process, especially in the energy sector where safety is of the utmost importance. To realise this, engineers have to manually analyse data acquired from several assets using predefined diagnostic processes, but this is a time-consuming process requiring significant amounts of specialist expert knowledge. Data-driven approaches to support fault detection and diagnosis, and other similar problems, can produce accurate results comparable to what the engineers can achieve in a fraction of the time. However, the majority of these data-driven techniques are black box techniques and lack explainability which is often necessary for explaining decisions about critical assets in the power generation industry. Knowledge-based systems, such as rule-based expert systems have been shown to provide not only accurate decisions but also the explanation and reasoning behind these decisions in some related applications. However, there is a significant time cost associated with the development of knowledge-based systems, and in particular with the knowledge elicitation process, where the domain expert’s knowledge is formalised and is encoded into the system. This challenge is commonly referred to as the knowledge elicitation bottleneck. In this paper, we present a novel approach to performing the knowledge elicitation using a set of symbolic primitives (rise, fall, fluctuate, and stable) to parameterise typical time-series condition monitoring data. The knowledge is represented by using a common language that can easily be communicated with (and from) the domain experts. This allows for the quick and accurate elicitation of the domain experts knowledge, but also the formalisation and implementation of the knowledge into a rapidly produced diagnostic system. Further to this, due to the parametrisation of the knowledge, it is possible to iteratively improve the knowledge base by updating these parameters based on new unseen data. This approach was applied to the Tennessee Eastman dataset, which is simulated data of a real-world industrial process. It was found that by using this approach it was possible to accurately and quickly capture the knowledge required to detect several faults within the case study dataset, but also provided fully explained reasons why each fault was detected by relating the explanations to the symbolic primitives previously defined.

Digital Marketplace is a market-based framework for trading mobile communications services at a service level. It is well suited for managing a future mobile communications environment where the one-to-one mapping between network operators and subscribers no longer holds, and the subscribers are given the option to select a network operator and a wireless access technology that matches their preferences best at a service level. As with any market-based framework, it is important to analyse the selling mechanism from the economic perspective. In this paper, we address the deficiencies of previous research on the economic equilibrium of the Digital Marketplace. We achieve this by proposing an approximation to the equilibrium of the Digital Marketplace through the use of an auction with common prior.

While there has been interest in modeling the group behavior of herds or flocks, much of this work has focused on simulating their collective spatial motion patterns, which have not accounted for individuality in the herd and instead assume a homogenized role for all members or subgroups of the herd. Animal behavior experts have noted that domestic animals exhibit behaviors that are indicative of social hierarchy: leader/follower type behaviors are present as well as dominance and subordination, aggression and rank order, and specific social affiliations may also exist. Both wild and domestic cattle are social species, and group behaviors are likely to be influenced by the expression of specific social interactions. In this paper, global positioning system coordinate fixes gathered from a herd of beef cows tracked in open fields over several days at a time are utilized to learn a model that focuses on the interactions within the herd as well as its overall movement. Using these data, in this way, explores the validity of existing group behavior models against actual herding behaviors. Domain knowledge, location geography, and human observations are utilized to explain the causes of these deviations from this idealized behavior.

Cognitive sensor network (CogSeNet) is an intelligent based wireless sensor network system which relies on cognitive processes to provide a dynamic capability in configuring wireless sensor network. CogSeNet is a network structure formed by sensor nodes equipped with cognitive modules allowing them to observe their operating environment and allowing a decision can be derived. A series of actions result so that the nodes can adapt and achieve certain goals by an overall policy. These goals can be as simple as to provide robust connectivity or as complex as negotiate additional resources from neighbouring network groups to forward mission-critical data. In this paper, the concept of cognitive sensor network is reviewed and a preliminary case study is illustrated.

In wireless sensor networks, end-to-end routing paths need to be established when nodes want to communicate with the desired destination. For nodes assumed to be static, many routing protocols such as Directed Diffusion have been proposed to meet this requirement efficiently. The performance of such routing protocols is relative to the given network connectivity. This paper addresses mobile sensor nodes taking into account the diversity of scattered node density and investigates how physical conditions impact on network connectivity which in turn influences routing performance. Three analysis metrics: path availability, path duration, and interavailable path time are proposed to quantify the impact of different physical conditions on network connectivity. Simulation results show that the network connectivity varies significantly as a function of different physical conditions.

Polarisation independent strained bulk RSOAs have been fabricated in the S-band for the first time. In system tests at 1.25 Gbit/s, the devices have been shown to operate over >60 nm with large return Path Loss Capabilities (PLC). The devices are capable of providing >25 dB PLC over a 60 nm span and hence the combination of two RSOAs provides PLC of >25 dB over a wavelength interval of >120 nm (1470 nm to 1590 nm). The temperature performance of the C/L band device has also been studied and a PLC of >20 dB has been shown at 60°C.

This paper proposes a novel multi-channel medium access control (MAC) that is based on CSMA/CA mechanism. The objective of this research is to evaluate the possibility of increasing network capacity and addressing collision problem of wireless sensor network comprising of low spec physical hardware. The proposed protocol can be implemented into wireless sensor network platform i.e. MICAz, Tmote-Sky without any additional hardware or firmware. Experiment results show that the proposed protocol can effectively improve the network performance compared with traditional single channel MAC.

In recent years, wireless sensor networks have been widely used in many types of monitoring applications. Conventionally, all sensor nodes are treated equally when they try to gain access to the air channel for packet transmission. For some applications such as in wild fire monitoring, sensor nodes

This paper examines the issue of instrumentation for civil structures and in particular discusses the role of sensing systems in the evolution of the smart building concept. The sheer size of most structures of interest presents its own technological challenge in designing a suitable sensing architecture which will also conveniently address the area to be covered and will also provide sufficient sensing points to adequately characterize the structure under test. In addition, the measurement problem itself, which comprises both process monitoring and in use assessment must be defined. The former is very much process specific and may range from assessing whether concrete is dry to ensuring that protective materials are properly installed. The latter is almost entirely concerned with providing an alarm for either the onset of unacceptable corrosion within the structure or the appearance of physical damage for example due to foundation failure. This paper presents an analysis of the principal issues which the civil engineering smart structure must address and derives generic system specifications. It continues to argue that fiber optic solutions are the most appropriate sensing technologies and examines a number of distributed fiber optic technologies which enable both physical and chemical parameters to be addressed throughout a large structure.