AI regulations are expected to prohibit machine learning models from using sensitive attributes during training. However, the latest Natural Language Processing (NLP) classifiers, which rely on deep learning, operate as black-box systems, complicating the detection and remediation of such misuse. Traditional bias mitigation methods in NLP aim for comparable performance across different groups based on attributes like gender or race but fail to address the underlying issue of reliance on protected attributes. To partly fix that, we introduce NLPGuard, a framework for mitigating the reliance on protected attributes in NLP classifiers. NLPGuard takes an unlabeled dataset, an existing NLP classifier, and its training data as input, producing a modified training dataset that significantly reduces dependence on protected attributes without compromising accuracy. NLPGuard is applied to three classification tasks: identifying toxic language, sentiment analysis, and occupation classification. Our evaluation shows that current NLP classifiers heavily depend on protected attributes, with up to $23\%$ of the most predictive words associated with these attributes. However, NLPGuard effectively reduces this reliance by up to $79\%$, while slightly improving accuracy.

The capacity of large language models (LLMs) to generate honest, harmless, and helpful responses heavily relies on the quality of user prompts. However, these prompts often tend to be brief and vague, thereby significantly limiting the full potential of LLMs. Moreover, harmful prompts can be meticulously crafted and manipulated by adversaries to jailbreak LLMs, inducing them to produce potentially toxic content. To enhance the capabilities of LLMs while maintaining strong robustness against harmful jailbreak inputs, this study proposes a transferable and pluggable framework that refines user prompts before they are input into LLMs. This strategy improves the quality of the queries, empowering LLMs to generate more truthful, benign and useful responses. Specifically, a lightweight query refinement model is introduced and trained using a specially designed reinforcement learning approach that incorporates multiple objectives to enhance particular capabilities of LLMs. Extensive experiments demonstrate that the refinement model not only improves the quality of responses but also strengthens their robustness against jailbreak attacks. Code is available at: https://github.com/Huangzisu/query-refinement .

This paper describes our approach to hierarchical multi-label detection of persuasion techniques in meme texts. Our model, developed as a part of the recent SemEval task, is based on fine-tuning individual language models (BERT, XLM-RoBERTa, and mBERT) and leveraging a mean-based ensemble model in addition to dataset augmentation through paraphrase generation from ChatGPT. The scope of the study encompasses enhancing model performance through innovative training techniques and data augmentation strategies. The problem addressed is the effective identification and classification of multiple persuasive techniques in meme texts, a task complicated by the diversity and complexity of such content. The objective of the paper is to improve detection accuracy by refining model training methods and examining the impact of balanced versus unbalanced training datasets. Novelty in the results and discussion lies in the finding that training with paraphrases enhances model performance, yet a balanced training set proves more advantageous than a larger unbalanced one. Additionally, the analysis reveals the potential pitfalls of indiscriminate incorporation of paraphrases from diverse distributions, which can introduce substantial noise. Results with the SemEval 2024 data confirm these insights, demonstrating improved model efficacy with the proposed methods.

A polyglot is a file that is valid in two or more formats. Polyglot files pose a problem for malware detection systems that route files to format-specific detectors/signatures, as well as file upload and sanitization tools. In this work we found that existing file-format and embedded-file detection tools, even those developed specifically for polyglot files, fail to reliably detect polyglot files used in the wild, leaving organizations vulnerable to attack. To address this issue, we studied the use of polyglot files by malicious actors in the wild, finding $30$ polyglot samples and $15$ attack chains that leveraged polyglot files. In this report, we highlight two well-known APTs whose cyber attack chains relied on polyglot files to bypass detection mechanisms. Using knowledge from our survey of polyglot usage in the wild -- the first of its kind -- we created a novel data set based on adversary techniques. We then trained a machine learning detection solution, PolyConv, using this data set. PolyConv achieves a precision-recall area-under-curve score of $0.999$ with an F1 score of $99.20$% for polyglot detection and $99.47$% for file-format identification, significantly outperforming all other tools tested. We developed a content disarmament and reconstruction tool, ImSan, that successfully sanitized $100$% of the tested image-based polyglots, which were the most common type found via the survey. Our work provides concrete tools and suggestions to enable defenders to better defend themselves against polyglot files, as well as directions for future work to create more robust file specifications and methods of disarmament.

The development of magnetic resonance imaging (MRI) for medical imaging has provided a leap forward in diagnosis, providing a safe, non-invasive alternative to techniques involving ionising radiation exposure for diagnostic purposes. It was described by Block and Purcel in 1946, and it was not until 1980 that the first clinical application of MRI became available. Since that time the MRI has gone through many advances and has altered the way diagnosing procedures are performed. Due to its ability to improve constantly, MRI has become a commonly used practice among several specialisations in medicine. Particularly starting 0.55T and 7T MRI technologies have pointed out enhanced preservation of image detail and advanced tissue characterisation. This review examines the integration of deep learning (DL) techniques into these MRI modalities, disseminating and exploring the study applications. It highlights how DL contributes to 0.55T and 7T MRI data, showcasing the potential of DL in improving and refining these technologies. The review ends with a brief overview of how MRI technology will evolve in the coming years.

The last few decades have led to the rise of research focused on propulsion and control systems for bio-inspired unmanned underwater vehicles (UUVs), which provide more maneuverable alternatives to traditional UUVs in underwater missions. Recent work has explored the use of time-series neural network surrogate models to predict thrust and power from vehicle design and fin kinematics. We develop a search-based inverse model that leverages kinematics-to-thrust and kinematics-to-power neural network models for control system design. Our inverse model finds a set of fin kinematics with the multi-objective goal of reaching a target thrust under power constraints while creating a smooth kinematics transition between flapping cycles. We demonstrate how a control system integrating this inverse model can make online, cycle-to-cycle adjustments to prioritize different system objectives, with improvements in increasing thrust generation or reducing power consumption of any given movement upwards of 0.5 N and 3.0 W in a range of 2.2 N and 9.0 W. As propulsive efficiency is of utmost importance for flapping-fin UUVs in order to extend their range and endurance for essential operations but lacks prior research, we develop a non-dimensional figure of merit (FOM), derived from measures of propulsive efficiency, that is able to evaluate different fin designs and kinematics, and allow for comparison with other bio-inspired platforms. We use the developed FOM to analyze optimal gaits and compare the performance between different fin materials, providing a better understanding of how fin materials affect thrust generation and propulsive efficiency and allowing us to inform control systems and weight for efficiency on the developed inverse gait-selector model.

Wi-Fi devices can effectively be used as passive radar systems that sense what happens in the surroundings and can even discern human activity. We propose, for the first time, a principled architecture which employs Variational Auto-Encoders for estimating a latent distribution responsible for generating the data, and Evidential Deep Learning for its ability to sense out-of-distribution activities. We verify that the fused data processed by different antennas of the same Wi-Fi receiver results in increased accuracy of human activity recognition compared with the most recent benchmarks, while still being informative when facing out-of-distribution samples and enabling semantic interpretation of latent variables in terms of physical phenomena. The results of this paper are a first contribution toward the ultimate goal of providing a flexible, semantic characterisation of black-swan events, i.e., events for which we have limited to no training data.

Existing efforts in safeguarding LLMs are limited in actively exposing the vulnerabilities of the target LLM and readily adapting to newly emerging safety risks. To address this, we present Purple-teaming LLMs with Adversarial Defender training (PAD), a pipeline designed to safeguard LLMs by novelly incorporating the red-teaming (attack) and blue-teaming (safety training) techniques. In PAD, we automatically collect conversational data that cover the vulnerabilities of an LLM around specific safety risks in a self-play manner, where the attacker aims to elicit unsafe responses and the defender generates safe responses to these attacks. We then update both modules in a generative adversarial network style by training the attacker to elicit more unsafe responses and updating the defender to identify them and explain the unsafe reason. Experimental results demonstrate that PAD significantly outperforms existing baselines in both finding effective attacks and establishing a robust safe guardrail. Furthermore, our findings indicate that PAD excels in striking a balance between safety and overall model quality. We also reveal key challenges in safeguarding LLMs, including defending multi-turn attacks and the need for more delicate strategies to identify specific risks.

This paper expands the damped block Newton (dBN) method introduced recently in [4] for 1D diffusion-reaction equations and least-squares data fitting problems. To determine the linear parameters (the weights and bias of the output layer) of the neural network (NN), the dBN method requires solving systems of linear equations involving the mass matrix. While the mass matrix for local hat basis functions is tri-diagonal and well-conditioned, the mass matrix for NNs is dense and ill-conditioned. For example, the condition number of the NN mass matrix for quasi-uniform meshes is at least ${\cal O}(n^4)$. We present a factorization of the mass matrix that enables solving the systems of linear equations in ${\cal O}(n)$ operations. To determine the non-linear parameters (the weights and bias of the hidden layer), one step of a damped Newton method is employed at each iteration. A Gauss-Newton method is used in place of Newton for the instances in which the Hessian matrices are singular. This modified dBN is referred to as dBGN. For both methods, the computational cost per iteration is ${\cal O}(n)$. Numerical results demonstrate the ability dBN and dBGN to efficiently achieve accurate results and outperform BFGS for select examples.

Developers try to evaluate whether an AI system can be misused by adversaries before releasing it; for example, they might test whether a model enables cyberoffense, user manipulation, or bioterrorism. In this work, we show that individually testing models for misuse is inadequate; adversaries can misuse combinations of models even when each individual model is safe. The adversary accomplishes this by first decomposing tasks into subtasks, then solving each subtask with the best-suited model. For example, an adversary might solve challenging-but-benign subtasks with an aligned frontier model, and easy-but-malicious subtasks with a weaker misaligned model. We study two decomposition methods: manual decomposition where a human identifies a natural decomposition of a task, and automated decomposition where a weak model generates benign tasks for a frontier model to solve, then uses the solutions in-context to solve the original task. Using these decompositions, we empirically show that adversaries can create vulnerable code, explicit images, python scripts for hacking, and manipulative tweets at much higher rates with combinations of models than either individual model. Our work suggests that even perfectly-aligned frontier systems can enable misuse without ever producing malicious outputs, and that red-teaming efforts should extend beyond single models in isolation.