Large language models (LLMs) are increasingly used to automate or augment penetration testing, but their effectiveness and reliability across attack phases remain unclear. We present a comprehensive evaluation of multiple LLM-based agents, from single-agent to modular designs, across realistic penetration testing scenarios, measuring empirical performance and recurring failure patterns. We also isolate the impact of five core functional capabilities via targeted augmentations: Global Context Memory (GCM), Inter-Agent Messaging (IAM), Context-Conditioned Invocation (CCI), Adaptive Planning (AP), and Real-Time Monitoring (RTM). These interventions support, respectively: (i) context coherence and retention, (ii) inter-component coordination and state management, (iii) tool use accuracy and selective execution, (iv) multi-step strategic planning, error detection, and recovery, and (v) real-time dynamic responsiveness. Our results show that while some architectures natively exhibit subsets of these properties, targeted augmentations substantially improve modular agent performance, especially in complex, multi-step, and real-time penetration testing tasks.

Current unlearning and safety training methods consistently fail to remove dangerous knowledge from language models. We identify the root cause - unlearning targets representations which are too general - and develop a highly selective technique that unlearns robustly while preserving general performance. Our method performs PCA on activations and module-output gradients to identify subspaces containing common representations, then collapses these subspaces before computing unlearning updates, a technique we term Collapse of Irrelevant Representations (CIR). This avoids unlearning general knowledge and targets only representations specific to the facts being unlearned. When unlearning bio-and cyber-hazardous facts from Llama-3.1-8B, we achieve over 30 greater reduction in post-attack accuracy than the best baseline (Circuit Breakers), while disrupting general performance 30 less, and using less than 3 GPU-seconds per fact. Thus, by disentangling harmful and benign capabilities at the level of representations, CIR enables robust and non-disruptive unlearning. Our code is available at: github.com/filyp/unlearning During pre-training, large language models (LLM) learn hazardous capabilities useful for bioterrorism and cybercrime (Li et al., 2024). They even acquire information about their own safety controls, which could enable future models to circumvent them (Roger, 2024; Greenblatt et al., 2024). Popular safety training approaches (RLHF, DPO) do not eliminate unwanted capabilities, but rather teach the models to stop using them (Lee et al., 2024). These concealed capabilities can be resurfaced via jailbreak attacks (Zou et al., 2023) or even accidentally through benign fine-tuning (Qi et al., 2023).

We study question answering in the domain of radio regulations, a legally sensitive and high-stakes area. We propose a telecom-specific Retrieval-Augmented Generation (RAG) pipeline and introduce, to our knowledge, the first multiple-choice evaluation set for this domain, constructed from authoritative sources using automated filtering and human validation. To assess retrieval quality, we define a domain-specific retrieval metric, under which our retriever achieves approximately 97% accuracy. Beyond retrieval, our approach consistently improves generation accuracy across all tested models. In particular, while naively inserting documents without structured retrieval yields only marginal gains for GPT-4o (less than 1%), applying our pipeline results in nearly a 12% relative improvement. These findings demonstrate that carefully targeted grounding provides a simple yet strong baseline and an effective domain-specific solution for regulatory question answering. All code and evaluation scripts, along with our derived question-answer dataset, are available at https://github.com/Zakaria010/Radio-RAG.

A common challenge with supervised machine learning (ML) in high energy physics (HEP) is the reliance on simulations for labeled data, which can often mismodel the underlying collision or detector response. To help mitigate this problem of domain shift, we propose RINO (Renormalization Group Invariance with No Labels), a self-supervised learning approach that can instead pretrain models directly on collision data, learning embeddings invariant to renormalization group flow scales. In this work, we pretrain a transformer-based model on jets originating from quantum chromodynamic (QCD) interactions from the JetClass dataset, emulating real QCD-dominated experimental data, and then finetune on the JetNet dataset -- emulating simulations -- for the task of identifying jets originating from top quark decays. RINO demonstrates improved generalization from the JetNet training data to JetClass data compared to supervised training on JetNet from scratch, demonstrating the potential for RINO pretraining on real collision data followed by fine-tuning on small, high-quality MC datasets, to improve the robustness of ML models in HEP.

Automated defect detection from UAV imagery of transmission lines is a challenging task due to the small size, ambiguity, and complex backgrounds of defects. This paper proposes TinyDef-DETR, a DETR-based framework designed to achieve accurate and efficient detection of transmission line defects from UAV-acquired images. The model integrates four major components: an edge-enhanced ResNet backbone to strengthen boundary-sensitive representations, a stride-free space-to-depth module to enable detail-preserving downsampling, a cross-stage dual-domain multi-scale attention mechanism to jointly model global context and local cues, and a Focaler-Wise-SIoU regression loss to improve the localization of small and difficult objects. Together, these designs effectively mitigate the limitations of conventional detectors. Extensive experiments on both public and real-world datasets demonstrate that TinyDef-DETR achieves superior detection performance and strong generalization capability, while maintaining modest computational overhead. The accuracy and efficiency of TinyDef-DETR make it a suitable method for UAV-based transmission line defect detection, particularly in scenarios involving small and ambiguous objects.

Automated visual inspection in medical-device manufacturing faces unique challenges, including extremely low defect rates, limited annotated data, hardware restrictions on production lines, and the need for validated, explainable artificial-intelligence systems. This paper presents two attention-guided autoencoder architectures that address these constraints through complementary anomaly-detection strategies. The first employs a multi-scale structural-similarity (4-MS-SSIM) index for inline inspection, enabling interpretable, real-time defect detection on constrained hardware. The second applies a Mahalanobis-distance analysis of randomly reduced latent features for efficient feature-space monitoring and lifecycle verification. Both approaches share a lightweight backbone optimised for high-resolution imagery for typical manufacturing conditions. Evaluations on the Surface Seal Image (SSI) dataset-representing sterile-barrier packaging inspection-demonstrate that the proposed methods outperform reference baselines, including MOCCA, CPCAE, and RAG-PaDiM, under realistic industrial constraints. Cross-domain validation on the MVTec-Zipper benchmark confirms comparable accuracy to state-of-the-art anomaly-detection methods. The dual-mode framework integrates inline anomaly detection and supervisory monitoring, advancing explainable AI architectures toward greater reliability, observability, and lifecycle monitoring in safety-critical manufacturing environments. To facilitate reproducibility, the source code developed for the experiments has been released in the project repository, while the datasets were obtained from publicly available sources.

Multimodal large language models achieve strong performance across diverse tasks but remain prone to hallucinations, where outputs are not grounded in visual inputs. This issue can be attributed to two main biases: text-visual bias, the overreliance on prompts and prior outputs, and co-occurrence bias, spurious correlations between frequently paired objects. We propose Gradient-based Influence-Aware Constrained Decoding (GACD), an inference-based method, that addresses both biases without auxiliary models, and is readily applicable to existing models without finetuning. The core of our approach is bias estimation, which uses first-order Taylor gradients to understand the contribution of individual tokens-visual features and text tokens-to the current output. Based on this analysis, GACD mitigates hallucinations through two components: (1) suppressing spurious visual features correlated with the output objects, and (2) rebalancing cross-modal contributions by strengthening visual features relative to text. Experiments across multiple benchmarks demonstrate that GACD effectively reduces hallucinations and improves the visual grounding of MLLM outputs.

Reinforcement learning (RL) using foundation models for policy approximations in multi-turn tasks remains challenging. We identify two main limitations related to sparse reward settings and policy gradient updates, based on which we formulate a key insight: updates from positive samples with high returns typically do not require policy regularisation, whereas updates from negative samples, reflecting undesirable behaviour, can harm model performance. This paper introduces Succeed or Learn Slowly (SoLS), a novel off-policy RL algorithm evaluated on mobile app control tasks. SoLS improves sample efficiency when fine-tuning foundation models for user interface navigation via a modified off-policy actor-critic approach, applying direct policy updates for positive samples and conservative, regularised updates for negative ones to prevent model degradation. We augment SoLS with Successful Transition Replay (STR), which prioritises learning from successful interactions, further improving sample efficiency. We evaluate SoLS on the AndroidWorld benchmark, where it significantly outperforms existing methods (at least 17% relative increase), including prompt-engineering and RL approaches, while requiring substantially fewer computational resources than GPT-4o-based methods with 5-60x faster inference.

Alzheimer's disease and related dementias(ADRD) affect nearly five million older adults in the United States, yet more than half remain undiagnosed. Speech-based natural language processing(NLP) offers a scalable approach for detecting early cognitive decline through subtle linguistic markers that may precede clinical diagnosis. This study develops and evaluates a speech-based screening pipeline integrating transformer embeddings with handcrafted linguistic features, synthetic augmentation using large language models(LLMs), and benchmarking of unimodal and multimodal classifiers. External validation assessed generalizability to a MCI-only cohort. Transcripts were drawn from the ADReSSo 2021 benchmark dataset(n=237, Pitt Corpus) and the DementiaBank Delaware corpus(n=205, MCI vs. controls). Ten transformer models were tested under three fine-tuning strategies. A late-fusion model combined embeddings from the top transformer with 110 linguistic features. Five LLMs(LLaMA8B/70B, MedAlpaca7B, Ministral8B,GPT-4o) generated label-conditioned synthetic speech for augmentation, and three multimodal LLMs(GPT-4o,Qwen-Omni,Phi-4) were evaluated in zero-shot and fine-tuned modes. On ADReSSo, the fusion model achieved F1=83.3(AUC=89.5), outperforming transformer-only and linguistic baselines. MedAlpaca7B augmentation(2x) improved F1=85.7, though larger scales reduced gains. Fine-tuning boosted unimodal LLMs(MedAlpaca7B F1=47.7=>78.7), while multimodal models performed lower (Phi-4=71.6;GPT-4o=67.6). On Delaware, the fusion plus 1x MedAlpaca7B model achieved F1=72.8(AUC=69.6). Integrating transformer and linguistic features enhances ADRD detection. LLM-based augmentation improves data efficiency but yields diminishing returns, while current multimodal models remain limited. Validation on an independent MCI cohort supports the pipeline's potential for scalable, clinically relevant early screening.

Large Language Models (LLMs) have shown improved generation performance through retrieval-augmented generation (RAG) following the retriever-reader paradigm, which supplements model inputs with externally retrieved knowledge. However, prior work often evaluates RAG holistically, assessing the retriever and reader jointly, making it difficult to isolate the true contribution of retrieval, particularly given the prompt sensitivity of LLMs used as readers. We move beyond perplexity and introduce Spectrum Projection Score (SPS), a lightweight and supervision-free metric that allows the reader to gauge the semantic alignment of a retrieved summary with its hidden representation by comparing the area formed by generated tokens from the summary, and the principal directions of subspace in the reader and to measure the relevance. Building on SPS we present xCompress, an inference-time controller framework that dynamically samples, ranks, and compresses retrieval summary candidates. Extensive experiments on five QA benchmarks with four open-sourced LLMs show that SPS not only enhances performance across a range of tasks but also provides a principled perspective on the interaction between retrieval and generation.