Intrusion detection poses a significant challenge within expansive and persistently interconnected environments. As malicious code continues to advance and sophisticated attack methodologies proliferate, various advanced deep learning-based detection approaches have been proposed. Nevertheless, the complexity and accuracy of intrusion detection models still need further enhancement to render them more adaptable to diverse system categories, particularly within resource-constrained devices, such as those embedded in edge computing systems. This research introduces a three-stage training paradigm, augmented by an enhanced pruning methodology and model compression techniques. The objective is to elevate the system's effectiveness, concurrently maintaining a high level of accuracy for intrusion detection. Empirical assessments conducted on the UNSW-NB15 dataset evince that this solution notably reduces the model's dimensions, while upholding accuracy levels equivalent to similar proposals.

Ensuring safety is paramount in the field of collaborative robotics to mitigate the risks of human injury and environmental damage. Apart from collision avoidance, it is crucial for robots to rapidly detect and respond to unexpected collisions. While several learning-based collision detection methods have been introduced as alternatives to purely model-based detection techniques, there is currently a lack of such methods designed for collaborative robots equipped with variable stiffness actuators. Moreover, there is potential for further enhancing the network's robustness and improving the efficiency of data training. In this paper, we propose a new network, the Modularized Attention-Dilated Convolutional Neural Network (MAD-CNN), for collision detection in robots equipped with variable stiffness actuators. Our model incorporates a dual inductive bias mechanism and an attention module to enhance data efficiency and improve robustness. In particular, MAD-CNN is trained using only a four-minute collision dataset focusing on the highest level of joint stiffness. Despite limited training data, MAD-CNN robustly detects all collisions with minimal detection delay across various stiffness conditions. Moreover, it exhibits a higher level of collision sensitivity, which is beneficial for effectively handling false positives, which is a common issue in learning-based methods. Experimental results demonstrate that the proposed MAD-CNN model outperforms existing state-of-the-art models in terms of collision sensitivity and robustness.

In the rapidly evolving field of speech generative models, there is a pressing need to ensure audio authenticity against the risks of voice cloning. We present AudioSeal, the first audio watermarking technique designed specifically for localized detection of AI-generated speech. AudioSeal employs a generator/detector architecture trained jointly with a localization loss to enable localized watermark detection up to the sample level, and a novel perceptual loss inspired by auditory masking, that enables AudioSeal to achieve better imperceptibility. AudioSeal achieves state-of-the-art performance in terms of robustness to real life audio manipulations and imperceptibility based on automatic and human evaluation metrics. Additionally, AudioSeal is designed with a fast, single-pass detector, that significantly surpasses existing models in speed - achieving detection up to two orders of magnitude faster, making it ideal for large-scale and real-time applications.

The increased application of machine learning (ML) in sensitive domains requires protecting the training data through privacy frameworks, such as differential privacy (DP). DP requires to specify a uniform privacy level $\varepsilon$ that expresses the maximum privacy loss that each data point in the entire dataset is willing to tolerate. Yet, in practice, different data points often have different privacy requirements. Having to set one uniform privacy level is usually too restrictive, often forcing a learner to guarantee the stringent privacy requirement, at a large cost to accuracy. To overcome this limitation, we introduce our novel Personalized-DP Output Perturbation method (PDP-OP) that enables to train Ridge regression models with individual per data point privacy levels. We provide rigorous privacy proofs for our PDP-OP as well as accuracy guarantees for the resulting model. This work is the first to provide such theoretical accuracy guarantees when it comes to personalized DP in machine learning, whereas previous work only provided empirical evaluations. We empirically evaluate PDP-OP on synthetic and real datasets and with diverse privacy distributions. We show that by enabling each data point to specify their own privacy requirement, we can significantly improve the privacy-accuracy trade-offs in DP. We also show that PDP-OP outperforms the personalized privacy techniques of Jorgensen et al. (2015).

The structure of data organization is widely recognized as having a substantial influence on the efficacy of machine learning algorithms, particularly in binary classification tasks. Our research provides a theoretical framework suggesting that the maximum potential of binary classifiers on a given dataset is primarily constrained by the inherent qualities of the data. Through both theoretical reasoning and empirical examination, we employed standard objective functions, evaluative metrics, and binary classifiers to arrive at two principal conclusions. Firstly, we show that the theoretical upper bound of binary classification performance on actual datasets can be theoretically attained. This upper boundary represents a calculable equilibrium between the learning loss and the metric of evaluation. Secondly, we have computed the precise upper bounds for three commonly used evaluation metrics, uncovering a fundamental uniformity with our overarching thesis: the upper bound is intricately linked to the dataset's characteristics, independent of the classifier in use. Additionally, our subsequent analysis uncovers a detailed relationship between the upper limit of performance and the level of class overlap within the binary classification data. This relationship is instrumental for pinpointing the most effective feature subsets for use in feature engineering.

Safety measures need to be systemically investigated to what extent they evaluate the intended performance of Deep Neural Networks (DNNs) for critical applications. Due to a lack of verification methods for high-dimensional DNNs, a trade-off is needed between accepted performance and handling of out-of-distribution (OOD) samples. This work evaluates rejecting outputs from semantic segmentation DNNs by applying a Mahalanobis distance (MD) based on the most probable class-conditional Gaussian distribution for the predicted class as an OOD score. The evaluation follows three DNNs trained on the Cityscapes dataset and tested on four automotive datasets and finds that classification risk can drastically be reduced at the cost of pixel coverage, even when applied on unseen datasets. The applicability of our findings will support legitimizing safety measures and motivate their usage when arguing for safe usage of DNNs in automotive perception.

The growing popularity of Android requires malware detection systems that can keep up with the pace of new software being released. According to a recent study, a new piece of malware appears online every 12 seconds. To address this, we treat Android malware detection as a streaming data problem and explore the use of active online learning as a means of mitigating the problem of labelling applications in a timely and cost-effective manner. Our resulting framework achieves accuracies of up to 96\%, requires as little of 24\% of the training data to be labelled, and compensates for concept drift that occurs between the release and labelling of an application. We also consider the broader practicalities of online learning within Android malware detection, and systematically explore the trade-offs between using different static, dynamic and hybrid feature sets to classify malware.

Cybersecurity remains a critical challenge in the digital age, with network traffic flow anomaly detection being a key pivotal instrument in the fight against cyber threats. In this study, we address the prevalent issue of data integrity in network traffic datasets, which are instrumental in developing machine learning (ML) models for anomaly detection. We introduce two refined versions of the CICIDS-2017 dataset, NFS-2023-nTE and NFS-2023-TE, processed using NFStream to ensure methodologically sound flow expiration and labeling. Our research contrasts the performance of the Random Forest (RF) algorithm across the original CICIDS-2017, its refined counterparts WTMC-2021 and CRiSIS-2022, and our NFStream-generated datasets, in both binary and multi-class classification contexts. We observe that the RF model exhibits exceptional robustness, achieving consistent high-performance metrics irrespective of the underlying dataset quality, which prompts a critical discussion on the actual impact of data integrity on ML efficacy. Our study underscores the importance of continual refinement and methodological rigor in dataset generation for network security research. As the landscape of network threats evolves, so must the tools and techniques used to detect and analyze them.

Large Language Models (LLMs), exemplified by ChatGPT, have significantly reshaped text generation, particularly in the realm of writing assistance. While ethical considerations underscore the importance of transparently acknowledging LLM use, especially in scientific communication, genuine acknowledgment remains infrequent. A potential avenue to encourage accurate acknowledging of LLM-assisted writing involves employing automated detectors. Our evaluation of four cutting-edge LLM-generated text detectors reveals their suboptimal performance compared to a simple ad-hoc detector designed to identify abrupt writing style changes around the time of LLM proliferation. We contend that the development of specialized detectors exclusively dedicated to LLM-assisted writing detection is necessary. Such detectors could play a crucial role in fostering more authentic recognition of LLM involvement in scientific communication, addressing the current challenges in acknowledgment practices.

Graph neural networks (GNNs) have achieved remarkable performance on graph-structured data. However, GNNs may inherit prejudice from the training data and make discriminatory predictions based on sensitive attributes, such as gender and race. Recently, there has been an increasing interest in ensuring fairness on GNNs, but all of them are under the assumption that the training and testing data are under the same distribution, i.e., training data and testing data are from the same graph. Will graph fairness performance decrease under distribution shifts? How does distribution shifts affect graph fairness learning? All these open questions are largely unexplored from a theoretical perspective. To answer these questions, we first theoretically identify the factors that determine bias on a graph. Subsequently, we explore the factors influencing fairness on testing graphs, with a noteworthy factor being the representation distances of certain groups between the training and testing graph. Motivated by our theoretical analysis, we propose our framework FatraGNN. Specifically, to guarantee fairness performance on unknown testing graphs, we propose a graph generator to produce numerous graphs with significant bias and under different distributions. Then we minimize the representation distances for each certain group between the training graph and generated graphs. This empowers our model to achieve high classification and fairness performance even on generated graphs with significant bias, thereby effectively handling unknown testing graphs. Experiments on real-world and semi-synthetic datasets demonstrate the effectiveness of our model in terms of both accuracy and fairness.