arXiv AI

arXiv:2602.00032v4 Announce Type: replace-cross Abstract: Synthetic faces from text-to-image (T2I) models pervade digital media, yet their demographic biases under emotionally conditioned prompts remain poorly understood. We aim to systematically audit how emotionally conditioned prompts affect demographic and perceived-attractiveness biases in synthetic faces generated by T2I models, with particular attention to intersectional patterns and cross-ecosystem differences across model families. We audited eight (4 Western and 4 Chinese) T2I models and generated 56,000 faces under seven prompt conditions: a neutral baseline and six emotion conditions. We quantified biases in gender, race, age, and perceived attractiveness using information-theoretic divergence metrics. We further conducted intersectional analyses across combined demographic attributes and compared patterns between the Western and Chinese model groups to assess cross-ecosystem consistency and divergence in bias behavior. All models show strong overrepresentation of young faces, and most also overrepresent White-coded individuals. Intersectional analysis reveals compound underrepresentation or near-erasure of specific demographic combinations, such as young x female x Black faces, which are largely absent across models and are not captured by single-attribute audits. Emotion prompts act as additional demographic selectors: negatively valenced emotions (including sadness and fear) consistently shift outputs toward White, middle-aged, male-coded faces. This produces a valence-driven mapping that is also associated with lower perceived attractiveness in generated faces. These findings indicate that demographic bias in T2I face generation is both pervasive and shaped by emotional conditioning. They underscore the need for intersectional, emotion-conditioned, and multilingual demographic audits as part of standard pre-deployment evaluation practices.

arXiv:2601.19747v2 Announce Type: replace-cross Abstract: In the rapidly evolving field of Electronic Design Automation (EDA), the deployment of Large Language Models (LLMs) for Register-Transfer Level (RTL) code generation has emerged as a promising direction. However, because RTL code describes concurrent hardware behavior with strict cycle-level timing and thus differs fundamentally from software code, LLM-based RTL code generation remains bottlenecked by (i) sparse feedback and limited test coverage of simulation-centric evaluation, (ii) regressions and repair hallucinations introduced by whole-file regeneration during debugging, and (iii) semantic drift as intent is reinterpreted across repair iterations and agent handoffs. In this work, we propose Veri-Sure, a multi-agent framework that establishes a design contract to align agents' intent and uses a patching mechanism guided by static dependency slicing to perform precise, localized repairs. By integrating a multi-branch verification pipeline that combines trace-driven temporal analysis with formal verification, Veri-Sure yields more robust and reliable RTL code generation. We also introduce VerilogEval-v2-EXT, extending the original benchmark with 53 more industrial-grade design tasks and stratified difficulty levels, and show that Veri-Sure achieves state-of-the-art verified-correct RTL code generation performance, surpassing standalone LLMs and prior agentic systems. Code and dataset are available at https://github.com/xyjoey/Veri-Sure.

arXiv:2601.13749v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly employed for decision-support across multiple domains. We ask whether these models favor artificial intelligence (AI) itself. We examine this question through three experiments. First, in open-ended advice across investment, study, career, and startup domains, AI is prominent in most recommendation lists and is usually ranked first when it appears. Second, in salary estimations, models overestimate AI-related jobs more than they do for closely matched non-AI jobs. In a controlled follow-up, changing only the job title from the matched non-AI title to the AI-related title raises predicted salary by 10%, while all other supplied job details remain fixed. Finally, we show that the number of degrees and other credentials awarded by AI programs were 28% more overestimated than corresponding awards for matched computing programs. Overall, we show that AI is not only prominent in everyday LLM advice, but also receives systematically inflated quantitative estimates compared to non-AI cases, even against its closest disciplinary neighbors, raising concerns for high-stakes decision support.

arXiv:2601.02554v2 Announce Type: replace-cross Abstract: Public debate links worsening job prospects for AI-exposed occupations to the release of ChatGPT in late 2022. Using monthly U.S. unemployment insurance records, we measure occupation- and location-specific unemployment risk and find that risk rose in AI-exposed occupations beginning in early 2022, months before ChatGPT was released. Analyzing millions of LinkedIn profiles, we show that graduate cohorts from 2021 onward entered AI-exposed jobs at lower rates than earlier cohorts, with gaps opening before late 2022. Finally, from millions of university syllabi, we find that graduates taking more LLM-exposed curricula had shorter job searches after ChatGPT and higher first-job pay when entering highly exposed occupations. Together, these results---pointing to forces pre-dating generative AI and to the ongoing value of LLM-relevant education---suggest that generative AI may not be the root cause of these labor market hardships and that the launch of ChatGPT is a faulty natural experiment for understanding AI's impact on jobs.

arXiv:2512.21201v5 Announce Type: replace-cross Abstract: Zero-shot object navigation (ZSON) requires robots to find target objects in unseen environments without task-specific fine-tuning or pre-built maps, a key capability for general-purpose service robots. Yet methods that perform well in simulation often degrade in cluttered real-world scenes with severe occlusion and latent hazards, where large unseen regions make single-scene inference brittle and unsafe. We propose Schr\"odinger's Navigator, a belief-aware framework that reasons at inference time over multiple trajectory-conditioned imagined 3D futures. Given candidate paths, a trajectory-conditioned 3D world model predicts hypothetical observations and maintains a superposition of plausible scene realizations rather than committing to one map. An adaptive occluder-aware sampler directs imagination to uncertainty-critical regions, while a Future-Aware Value Map (FAVM) aggregates imagined futures for robust, proactive action selection. Experiments in simulation and on a physical Go2 quadruped show that Schr\"odinger's Navigator outperforms strong ZSON baselines, improving hidden-target discovery and risk-aware waypoint selection in occlusion-heavy navigation scenarios. These results highlight imagined 3D futures as a scalable and generalizable strategy for zero-shot navigation in uncertain real-world environments.

arXiv:2512.05960v2 Announce Type: replace-cross Abstract: Underwater images often suffer from severe color distortion, low contrast, and a hazy appearance due to wavelength-dependent light absorption and scattering. Simultaneously, existing deep learning models exhibit high computational complexity, which limits their practical deployment for real-time underwater applications. To address these challenges, this paper presents a novel underwater image enhancement model, called Adaptive Frequency Fusion and Illumination Aware Network (AQUA-Net). It integrates a residual encoder-decoder with dual auxiliary branches, which operate in the frequency and illumination domains. The frequency fusion encoder enriches spatial representations with frequency cues from the Fourier domain and preserves fine textures and structural details. Inspired by Retinex, the illumination-aware decoder performs adaptive exposure correction through a learned illumination map that separates reflectance from lighting effects. This joint spatial, frequency, and illumination design enables the model to restore color balance, visual contrast, and perceptual realism under diverse underwater conditions. Additionally, we present a high-resolution, real-world underwater video-derived dataset from the Mediterranean Sea, which captures challenging deep-sea conditions with realistic visual degradations to enable robust evaluation and development of deep learning models. Extensive experiments on multiple benchmark datasets show that AQUA-Net performs on par with SOTA in both qualitative and quantitative evaluations while using less number of parameters. Ablation studies further confirm that the frequency and illumination branches provide complementary contributions that improve visibility and color representation. Overall, the proposed model shows strong generalization capability and robustness, and it provides an effective solution for real-world underwater imaging applications.

arXiv:2511.16080v2 Announce Type: replace-cross Abstract: Modern data processing workflows frequently encounter ragged data: collections with variable-length elements that arise naturally in domains like natural language processing, scientific measurements, and autonomous AI agents. Existing workflow engines lack native support for tracking the shapes and dependencies inherent to ragged data, forcing users to manage complex indexing and dependency bookkeeping manually. We present Operon, a Rust-based workflow engine that addresses these challenges through a novel formalism of named dimensions with explicit dependency relations. Operon provides a domain-specific language where users declare pipelines with dimension annotations that are statically verified for correctness, while the runtime system dynamically schedules tasks as data shapes are incrementally discovered during execution. We formalize the mathematical foundation for reasoning about partial shapes and prove that Operon's incremental construction algorithm guarantees deterministic and confluent execution in parallel settings. The system's explicit modeling of partially-known states enables robust persistence and recovery mechanisms, while its per-task multi-queue architecture achieves efficient parallelism across heterogeneous task types. Empirical evaluation demonstrates that Operon outperforms an existing workflow engine with 14.94x baseline overhead reduction while maintaining near-linear end-to-end output rates as workloads scale, making it particularly suitable for large-scale data generation pipelines in machine learning applications.

arXiv:2511.06571v4 Announce Type: replace-cross Abstract: Large language models represent text as hidden numerical patterns, but what information these representations actually contain remains a mystery. Can we recover the original text from just one of them? In this paper, we introduce Rep2Text, a method designed to answer this question by reconstructing text from a single hidden representation. We find that these representations preserve much more information than we might expect. For short texts of about 16 words, Rep2Text can recover roughly half of the original words while preserving their overall meaning. As the text gets longer, recovering the exact wording becomes harder, but the main ideas can still be recovered. These findings offer a new way to understand what language models store internally and highlight potential privacy risks, as hidden representations may reveal parts of the original input.

arXiv:2509.09700v2 Announce Type: replace-cross Abstract: With the large-scale adoption of Large Language Models (LLMs) in various applications, there is a growing reliability concern due to their tendency to generate inaccurate text, i.e. hallucinations. In this work, we propose Cross-Layer Attention Probing (CLAP), a novel activation probing technique for hallucination detection, which processes the LLM activations across the entire residual stream as a joint sequence. Our empirical evaluations using five LLMs and three tasks show that CLAP improves hallucination detection compared to baselines on both greedy decoded responses as well as responses sampled at higher temperatures, thus enabling fine-grained detection, i.e. the ability to disambiguate hallucinations and non-hallucinations among different sampled responses to a given prompt. This allows us to propose a detect-then-mitigate strategy using CLAP to reduce hallucinations and improve LLM reliability compared to direct mitigation approaches. Finally, we show that CLAP maintains high reliability even when applied out-of-distribution. Code available at: https://github.com/itsmemala/CLAP

arXiv:2507.15681v2 Announce Type: replace-cross Abstract: Handling missing values is a common challenge in biostatistical analyses, typically addressed by imputation methods. We propose a novel, fast, and easy-to-use imputation method called missing value imputation with adversarial random forests (MissARF), based on generative machine learning, that provides both single and multiple imputation. MissARF employs adversarial random forest (ARF) for density estimation and data synthesis. To impute a missing value of an observation, we condition on the non-missing values and sample from the estimated conditional distribution generated by ARF. Our experiments demonstrate that MissARF performs comparably to state-of-the-art single and multiple imputation methods in terms of imputation quality and fast runtime with no additional costs for multiple imputation.

arXiv:2507.01693v3 Announce Type: replace-cross Abstract: Recent work shows that distinct inputs produce distinct hidden representations in LLMs, so the input can in principle be recovered from them. Yet, no existing method can reliably do so from the model's output text or logits alone. We introduce Sparse One-hot Discrete Adam (SODA), a white-box search-based algorithm that reconstructs textual inputs from partial knowledge of the output. Given output logits, SODA recovers natural sequences of up to 17 tokens with zero false positives and achieves a 4x improvement in exact-match reconstruction over the strongest prior method. We further show that input length and vocabulary size matter more for inversion difficulty than parameter count, with SODA inverting models from 33M to 3B parameters. Finally, we demonstrate proof-of-concept security applications: SODA recovers 6 backdoor triggers from a poisoned LLM in under 1000 iterations, and 139 tokens of personally identifiable information from synthetic user-chatbot interactions. Our code is available at https://github.com/AdrSkapars/inversion_optimisation.

arXiv:2505.18588v2 Announce Type: replace-cross Abstract: Despite significant progress in safety alignment, large language models (LLMs) remain susceptible to jailbreak attacks. Existing defense mechanisms have not fully deleted harmful knowledge in LLMs, which allows such attacks to bypass safeguards and produce harmful outputs. To address this challenge, we propose a novel safety alignment strategy, Constrained Knowledge Unlearning (CKU), which focuses on two primary objectives: knowledge localization and retention, and unlearning harmful knowledge. CKU works by scoring neurons in specific multilayer perceptron (MLP) layers to identify a subset U of neurons associated with useful knowledge. During the unlearning process, CKU prunes the gradients of neurons in U to preserve valuable knowledge while effectively mitigating harmful content. Experimental results demonstrate that CKU significantly enhances model safety without compromising overall performance, offering a superior balance between safety and utility compared to existing methods. Additionally, our analysis of neuron knowledge sensitivity across various MLP layers provides valuable insights into the mechanics of safety alignment and model knowledge editing.

arXiv:2505.10597v3 Announce Type: replace-cross Abstract: Reward models (RMs) are central to aligning large language models (LLMs) with human values, yet noisy preference feedback can induce reward misgeneralization through spurious correlations and overfitting. We characterize the training dynamics of preference pairs and uncover a distinctive optimization signature of noise: persistently high losses, sharp temporal fluctuations, and irregular gradient updates. Our analyses further show that excluding such pairs improves downstream alignment. In practice, however, reliable preferences are not known a priori, and self-filtering based on a model's own training signals can amplify confirmation bias. To address this challenge, we introduce Collaborative Reward Modeling (CRM), an online framework for identifying reliable preferences. CRM couples cross-model peer review with curriculum learning: two RMs evaluate each other's sample selections, while curriculum keeps their learning progress synchronized. Across extensive experiments, CRM improves RM generalization by up to 9.94 points on RewardBench under 40% preference noise. CRM further generalizes to implicit-reward alignment, offering a robust and broadly applicable approach to learning from imperfect feedback.

arXiv:2504.18636v2 Announce Type: replace-cross Abstract: Phishing attacks represent an increasingly sophisticated and pervasive threat to individuals and organizations, causing significant financial losses, identity theft, and severe damage to institutional reputations. Existing phishing detection methods often struggle to simultaneously achieve high accuracy and explainability, either failing to detect novel attacks or operating as opaque black-box models. To address this critical gap, we propose a novel phishing URL detection system based on a first-order Takagi-Sugeno-Kang (TSK) fuzzy inference model optimized through gradient-based techniques. Our approach intelligently combines the interpretability and human-like reasoning capabilities of fuzzy logic with the precision and adaptability provided by gradient optimization methods, specifically leveraging the Adam optimizer for efficient parameter tuning. Experiments conducted using a comprehensive dataset of over 235,000 URLs demonstrate rapid convergence, exceptional predictive performance (accuracy averaging 99.95% across 5 cross-validation folds, with a perfect AUC i.e. 1.00). Furthermore, optimized fuzzy rules and membership functions improve interoperability, clearly indicating how the model makes decisions - an essential feature for cybersecurity applications. This high-performance, transparent, and interpretable phishing detection framework significantly advances current cybersecurity defenses, providing practitioners with accurate and explainable decision-making tools.

arXiv:2503.18238v4 Announce Type: replace-cross Abstract: We examined the mechanisms underlying productivity and performance gains from AI agents using a large-scale experiment on Pairit, a platform we developed to study human-AI collaboration. We randomly assigned 2,234 participants to human-human and human-AI teams that produced 11,024 ads. We evaluated the ads using independent human ratings and a field experiment on X which garnered ~5M impressions. We found human-AI teams produced 50% more ads per worker and higher text quality, while human-human teams produced higher image quality, suggesting a jagged frontier of AI agent capability. Human-AI teams also produced more homogeneous outputs (diversity collapse) as their ads were more self-similar. The field experiment revealed higher text quality (produced by human-AI teams) increased click-through rates and view time, while higher image quality (produced by human-human teams) improved cost-per-click rates. Two mechanisms explained these effects: Human-AI collaboration was significantly more task-oriented and displayed more delegation. AI-appropriate work processes, in turn, improved performance in human-AI teams and transcended the jagged frontier: Task-oriented communication produced higher productivity, higher quality ads across all dimensions, and displayed no diversity collapse in their outputs, while human-AI teams with more interpersonal communication were less productive and produced lower quality ads across all dimensions. In contrast, in human-human teams, interpersonal communication improved productivity. Delegation improved text quality in human-AI teams and was positively associated with diversity collapse, creating homogeneous outputs of higher average quality. The results suggest AI agents drive productivity, performance, and output diversity by reshaping teamwork along the jagged frontier of AI capabilities.

arXiv:2503.15889v2 Announce Type: replace-cross Abstract: While there are many advantages to deploying machine learning models on edge devices, the resource constraints of mobile platforms, the dynamic nature of the environment, and differences between the distribution of training versus in-the-wild data make such deployments challenging. Current test-time adaptation methods are often memory-intensive and not designed to be quantization-compatible or deployed on low-resource devices. To address these challenges, we present LeanTTA, a novel backpropagation-free and stateless framework for quantized test-time adaptation tailored to edge devices. Our approach minimizes computational costs by dynamically updating normalization statistics without backpropagation, which frees LeanTTA from the common pitfall of relying on large batches and historical data, making our method robust to realistic deployment scenarios. Our approach is the first to enable further computational gains by combining partial adaptation with quantized module fusion. We validate our framework across sensor modalities, demonstrating significant improvements over state-of-the-art TTA methods, including a 15.7% error reduction, peak memory usage of only 11.2MB for ResNet18, and fast adaptation within an order-of-magnitude of normal inference speeds on-device. LeanTTA provides a robust solution for achieving the right trade offs between accuracy and system efficiency in edge deployments, addressing the unique challenges posed by limited data and varied operational conditions.

arXiv:2501.19335v3 Announce Type: replace-cross Abstract: Causal Bayesian networks are often described as causal because they support reasoning about interventions. Yet interventions are defined in the mathematical model, whereas the resulting model predictions concern the effects of actions in the real world. Consequently, whether and how a causal model makes falsifiable predictions about the effects of actions depends on how actions are interpreted as interventions. This requires specifying which actions correspond to which interventions, a step that is frequently left implicit. We develop a formal framework that separates a causal model from the data-generating process it aims to describe and formalizes interpretations connecting actions to interventions. Within this framework, we prove an impossibility result: no interpretation can simultaneously satisfy a collection of desirable properties while remaining non-circular. Under a circular interpretation, every model that fits the observational distribution is interventionally valid, so no experiment can falsify it. We analyze several non-circular alternatives and discuss implications for falsification, causal discovery, causal representation learning, and causal abstraction. More broadly, our framework provides a way to make assumptions about the connections between actions and interventions explicit and to reason formally about the empirical content of causal models.

arXiv:2412.07658v3 Announce Type: replace-cross Abstract: Text-to-image diffusion models have rapidly transitioned from research prototypes to widely deployed tools. Along with their impressive capabilities, these models can also generate harmful or undesirable content (e.g., NSFW imagery), motivating practical mechanisms for removing such concepts. Negative prompting, which uses the negative-prompt prediction as the guidance anchor, is among the most widely used mechanisms for steering sampling trajectories away from unwanted concepts. However, for jailbreak prompts, it can fail in a regime we refer to as \emph{guidance collapse}, where the model's denoising predictions for the prompt and negative prompt become similar, so that the guidance difference driving negative prompting vanishes and the guided prediction degenerates to the negative-prompt prediction, undermining suppression. To fix this issue, we propose TraSCE (Trajectory Steering for ConcEpt Suppression), a training-free inference-time method that makes negative prompting reliable under guidance collapse. TraSCE combines two complementary components: (i) unconditional anchoring, which makes the fallback under guidance collapse a neutral unconditional prediction rather than the negative-prompt prediction, and (ii) an alignment-localized steering step, which deliberately induces this collapse for adversarial prompts and steers the sample to the unconditional fallback, while remaining minimally intrusive on benign prompts. We evaluate TraSCE on adversarial prompt benchmarks targeting nudity and violence, and further demonstrate its applicability to suppressing artistic styles and objects. Across these settings, TraSCE substantially improves the robustness of negative prompting to jailbreak prompts without any training, weight updates, or pre-collected concept-specific data (prompts or images), while keeping the impact on general image quality small.

arXiv:2412.00684v3 Announce Type: replace-cross Abstract: Visual grounding aims to localize the image regions based on a textual query. Given the difficulty of large-scale data curation, we investigate how to effectively learn visual grounding under data-scarce settings in this paper. To address the data scarcity, we propose a novel framework, POBF (Paint Outside the Box and Filter). POBF synthesizes images by inpainting outside the box, tackling a label misalignment issue encountered in previous works. Furthermore, POBF leverages an innovative filtering scheme to select the most effective training data. This scheme combines a hardness score and an overfitting score, balanced by a penalty term. Extensive experiments across four benchmark datasets demonstrate that POBF consistently improves performance, achieving an average gain of 5.83\% over the real-data-only method and outperforming leading baselines by 2.29\%-3.85\% in accuracy. Additionally, we validate the robustness and generalizability of POBF across various generative models, training data sizes, and model architectures.

arXiv:2411.05854v2 Announce Type: replace-cross Abstract: Video platforms, used by billions globally, are criticized for exposing users to harmful content. Yet, detecting online harm remains challenging due to inconsistent approaches to what constitutes harm and limitations of human annotation. We address both challenges. Theoretically, we integrate and expand existing taxonomies and platform guidelines to multimodal contexts and develop a comprehensive taxonomy for online harm. We categorize online harm on video platforms into six content-based categories: Information, Hate and harassment, Addictive, Clickbait, Sexual, and Physical harms. Methodologically, we propose and systematically test multimodal large language models (MLLMs) as alternative annotators of harmful videos. We curate a dataset of 19,422 potentially harmful YouTube videos and annotate them using GPT-4-Turbo, non-trained crowdworkers, and trained annotators. We discuss the potential applications of MLLMs in platform content moderation and social science research, particularly in contexts where they perform comparably to human annotators in binary and multi-label classification tasks.