Computer-use agents (CUAs) promise to automate complex tasks across operating systems (OS) and the web, but remain vulnerable to indirect prompt injection. Current evaluations of this threat either lack support realistic but controlled environments or ignore hybrid web-OS attack scenarios involving both interfaces. To address this, we propose RedTeamCUA, an adversarial testing framework featuring a novel hybrid sandbox that integrates a VM-based OS environment with Docker-based web platforms. Our sandbox supports key features tailored for red teaming, such as flexible adversarial scenario configuration, and a setting that decouples adversarial evaluation from navigational limitations of CUAs by initializing tests directly at the point of an adversarial injection. Using RedTeamCUA, we develop RTC-Bench, a comprehensive benchmark with 864 examples that investigate realistic, hybrid web-OS attack scenarios and fundamental security vulnerabilities. Benchmarking current frontier CUAs identifies significant vulnerabilities: Claude 3.7 Sonnet | CUA demonstrates an ASR of 42.9%, while Operator, the most secure CUA evaluated, still exhibits an ASR of 7.6%. Notably, CUAs often attempt to execute adversarial tasks with an Attempt Rate as high as 92.5%, although failing to complete them due to capability limitations. Nevertheless, we observe concerning high ASRs in realistic end-to-end settings, with the strongest-to-date Claude 4.5 Sonnet | CUA exhibiting the highest ASR of 60%, indicating that CUA threats can already result in tangible risks to users and computer systems. Overall, RedTeamCUA provides an essential framework for advancing realistic, controlled, and systematic analysis of CUA vulnerabilities, highlighting the urgent need for robust defenses to indirect prompt injection prior to real-world deployment.

Language model developers typically filter out high-risk content -- such as toxic or copyrighted text -- from their pre-training data to prevent models from generating similar outputs. However, removing such data altogether limits models' ability to recognize and appropriately respond to harmful or sensitive content. In this paper, we introduce Selective Loss to Understand but Not Generate (SLUNG), a pre-training paradigm through which models learn to understand high-risk data without learning to generate it. Instead of uniformly applying the next-token prediction loss, SLUNG selectively avoids incentivizing the generation of high-risk tokens while ensuring they remain within the model's context window. As the model learns to predict low-risk tokens that follow high-risk ones, it is forced to understand the high-risk content. Through our experiments, we show that SLUNG consistently improves models' understanding of high-risk data (e.g., ability to recognize toxic content) without increasing its generation (e.g., toxicity of model responses). Overall, our SLUNG paradigm enables models to benefit from high-risk text that would otherwise be filtered out.

Robot vision has greatly benefited from advancements in multimodal fusion techniques and vision-language models (VLMs). We adopt a task-oriented perspective to systematically review the applications and advancements of multimodal fusion methods and VLMs in the field of robot vision. For semantic scene understanding tasks, we categorize fusion approaches into encoder-decoder frameworks, attention-based architectures, and graph neural networks. Meanwhile, we also analyze the architectural characteristics and practical implementations of these fusion strategies in key tasks such as simultaneous localization and mapping (SLAM), 3D object detection, navigation, and manipulation. We compare the evolutionary paths and applicability of VLMs based on large language models (LLMs) with traditional multimodal fusion methods.Additionally, we conduct an in-depth analysis of commonly used datasets, evaluating their applicability and challenges in real-world robotic scenarios. Building on this analysis, we identify key challenges in current research, including cross-modal alignment, efficient fusion, real-time deployment, and domain adaptation. We propose future directions such as self-supervised learning for robust multimodal representations, structured spatial memory and environment modeling to enhance spatial intelligence, and the integration of adversarial robustness and human feedback mechanisms to enable ethically aligned system deployment. Through a comprehensive review, comparative analysis, and forward-looking discussion, we provide a valuable reference for advancing multimodal perception and interaction in robotic vision. A comprehensive list of studies in this survey is available at https://github.com/Xiaofeng-Han-Res/MF-RV.

Reinforcement learning (RL) has become central to training large language models (LLMs), yet the field lacks predictive scaling methodologies comparable to those established for pre-training. Despite rapidly rising compute budgets, there is no principled understanding of how to evaluate algorithmic improvements for scaling RL compute. We present the first large-scale systematic study, amounting to more than 400,000 GPU-hours, that defines a principled framework for analyzing and predicting RL scaling in LLMs. We fit sigmoidal compute-performance curves for RL training and ablate a wide range of common design choices to analyze their effects on asymptotic performance and compute efficiency. We observe: (1) Not all recipes yield similar asymptotic performance, (2) Details such as loss aggregation, normalization, curriculum, and off-policy algorithm primarily modulate compute efficiency without materially shifting the asymptote, and (3) Stable, scalable recipes follow predictable scaling trajectories, enabling extrapolation from smaller-scale runs. Combining these insights, we propose a best-practice recipe, ScaleRL, and demonstrate its effectiveness by successfully scaling and predicting validation performance on a single RL run scaled up to 100,000 GPU-hours. Our work provides both a scientific framework for analyzing scaling in RL and a practical recipe that brings RL training closer to the predictability long achieved in pre-training.

Large language model (LLM)-based reasoning systems have recently achieved gold medal-level performance in the IMO 2025 competition, writing mathematical proofs where, to receive full credit, each step must be not only correct but also sufficiently supported. To train LLM-based reasoners in such challenging, open-ended settings, strong verifiers capable of catching step-level mistakes are necessary prerequisites. We introduce Hard2V erify, a human-annotated, step-level verification benchmark produced with over 500 hours of human labor. Hard2V erify is designed to rigorously assess step-level verifiers at the frontier: V erifiers must provide step-level annotations or identify the first error in responses generated by frontier LLMs for very recent, challenging, and open-ended math questions. We evaluate 29 generative critics and process reward models, demonstrating that, beyond a few standouts, open-source verifiers lag closed source models. We subsequently analyze what drives poor performance in step-level verification, the impacts of scaling verifier compute, as well as fundamental questions such as self-verification and verification-generation dynamics.Figure 1: Comparison of models evaluated on both ProcessBench (Zheng et al., 2024a) and our Hard2V erify benchmark. Past benchmarks do not sufficiently evaluate in the frontier-level math settings that Hard2V erify does; On the same error identification task, Qwen2.5-Math-PRM-72B Mathematical reasoning serves as a gold-standard evaluation setting for benchmarking reasoning progress in large language models (LLMs). Over the past half-decade, benchmarks have been introduced to assess LLMs at the grade-school (Cobbe et al., 2021), high-school (Hendrycks et al., 2021), university (Zhang et al., 2023), and competition math level (MMA, 2025; He et al., 2024a; Gao et al., 2024). However, the progress of mathematical reasoning ability of LLMs has outpaced benchmark creation, with every subsequent release of a frontier LLM saturating new benchmarks, most recently with GPT -5 Pro achieving 96.5%+ on AIME 2024. As a result, recent efforts (Glazer et al., 2024; Phan et al., 2025) have written novel, unseen mathematical questions to test LLMs. 1 This paradigm requires training data with solutions that are easily verifiable, i.e., have solutions that can be easily checked against a known ground-truth by string matching or symbolic checkers. Math benchmarks, for the most part, also adopt the verifiable setup, where a model response is considered correct if its final answer matches the established ground-truth.

Recent works have proposed accelerating the wall-clock training time of actor-critic methods via the use of large-scale environment parallelization; unfortunately, these can sometimes still require large number of environment interactions to achieve a desired level of performance. Noting that well-structured representations can improve the generalization and sample efficiency of deep reinforcement learning (RL) agents, we propose the use of simplicial embeddings: lightweight representation layers that constrain embeddings to simplicial structures. This geometric inductive bias results in sparse and discrete features that stabilize critic bootstrapping and strengthen policy gradients. When applied to FastTD3, FastSAC, and PPO, simplicial embeddings consistently improve sample efficiency and final performance across a variety of continuous- and discrete-control environments, without any loss in runtime speed.

Despite the success of Reinforcement Learning from Human Feedback (RLHF) in aligning language models with human values, reward hacking-or reward over-optimization-remains a major challenge. We identify two key obstacles to its mitigation: (1) reward misgeneralization in reward modeling, where reward models overfit to spurious, preference-irrelevant features; and (2) the lack of suitable regularization during RL optimization, as existing token-level constraints often over-restrict the policy space. To address these issues, we propose InfoRM, an information-theoretic reward modeling framework based on the Information Bottleneck (IB) principle, which filters out preference-irrelevant information to alleviate reward misgeneralization. We further observe that reward-hacked responses manifest as pronounced outliers in InfoRM's IB latent space, measured by Mahalanobis distance from the SFT-induced distribution. Motivated by this, we introduce IBL, a distribution-level regularization that penalizes such deviations, effectively expanding the optimization landscape while maintaining alignment. We prove that IBL is theoretically equivalent to the pessimistic RL objective within the IB latent space. Finally, we present Mahalanobis Outlier Probability (MOP), a statistical metric for quantifying reward hacking severity, enabling principled hyperparameter tuning and online mitigation such as early stopping. Extensive experiments across diverse LLMs and datasets confirm the generality of our findings, the effectiveness of InfoRM and IBL, and the reliability of MOP as a diagnostic tool-collectively advancing the state of RLHF.

Federated continual learning (FCL) enables models to learn new tasks across multiple distributed clients, protecting privacy and without forgetting previously acquired knowledge. However, current methods face challenges balancing performance, privacy preservation, and communication efficiency. We introduce a Distributed Online LoRA for Federated INcremental learning method DOLFIN, a novel approach combining Vision Transformers with low-rank adapters designed to efficiently and stably learn new tasks in federated environments. Our method leverages LoRA for minimal communication overhead and incorporates DualGradient Projection Memory (DualGPM) to prevent forgetting. Evaluated on CIFAR-100, ImageNet-R, ImageNet-A, and CUB-200 under two Dirichlet heterogeneity settings, DOLFIN consistently surpasses six strong baselines in final average accuracy while matching their memory footprint. Orthogonal low-rank adapters offer an effective and scalable solution for privacy-preserving continual learning in federated settings.

Question Answering (QA) on narrative text poses a unique challenge to current systems, requiring a deep understanding of long, complex documents. However, the reliability of NarrativeQA, the most widely used benchmark in this domain, is hindered by noisy documents and flawed QA pairs. In this work, we introduce LiteraryQA, a high-quality subset of NarrativeQA focused on literary works. Using a human- and LLM-validated pipeline, we identify and correct low-quality QA samples while removing extraneous text from source documents. We then carry out a meta-evaluation of automatic metrics to clarify how systems should be evaluated on LiteraryQA. This analysis reveals that all n-gram-based metrics have a low system-level correlation to human judgment, while LLM-as-a-Judge evaluations, even with small open-weight models, can strongly agree with the ranking identified by humans. Finally, we benchmark a set of long-context LLMs on LiteraryQA. We release our code and data at https://github.com/SapienzaNLP/LiteraryQA.

This thesis investigates the application of near-infrared hyperspectral imaging (NIR-HSI) for food quality analysis. The investigation is conducted through four studies operating with five research hypotheses. For several analyses, the studies compare models based on convolutional neural networks (CNNs) and partial least squares (PLS). Generally, joint spatio-spectral analysis with CNNs outperforms spatial analysis with CNNs and spectral analysis with PLS when modeling parameters where chemical and physical visual information are relevant. When modeling chemical parameters with a 2-dimensional (2D) CNN, augmenting the CNN with an initial layer dedicated to performing spectral convolution enhances its predictive performance by learning a spectral preprocessing similar to that applied by domain experts. Still, PLS-based spectral modeling performs equally well for analysis of the mean content of chemical parameters in samples and is the recommended approach. Modeling the spatial distribution of chemical parameters with NIR-HSI is limited by the ability to obtain spatially resolved reference values. Therefore, a study used bulk mean references for chemical map generation of fat content in pork bellies. A PLS-based approach gave non-smooth chemical maps and pixel-wise predictions outside the range of 0-100\%. Conversely, a 2D CNN augmented with a spectral convolution layer mitigated all issues arising with PLS. The final study attempted to model barley's germinative capacity by analyzing NIR spectra, RGB images, and NIR-HSI images. However, the results were inconclusive due to the dataset's low degree of germination. Additionally, this thesis has led to the development of two open-sourced Python packages. The first facilitates fast PLS-based modeling, while the second facilitates very fast cross-validation of PLS and other classical machine learning models with a new algorithm.