arXiv:2605.29530v1 Announce Type: new
Abstract: Ethanol is conventionally perceived only as a pungent tastant, while the potential sweet properties of ethanol clusters have remained unrecognized. Here we show that ethanol tetramers exhibit a unique time dependent lingering sweetness, distinct from the immediate upupfront sweetness of conventional sweeteners. To objectively verify this phenomenon, we developed an impedance based bionic electronic tongue that monitors changes at a lipid polymer membrane interface. Sensory evaluation revealed increasing ethanol cluster content from 4.83% to 29.53% prolonged sweetness duration from 10s to 25s, whereas sweetness intensity remained low to moderate (1.0-1.7 on a 10 point scale). Using this electronic tongue, we achieved objective discrimination between upfront and lingering sweetness,full band phase angle scanning identified optimal frequencies of 0.933Hz (t=0.171s) for xylitol (fast binding and dissociation) and 0.03Hz (t=5.305s) for ethanol clusters (slow binding and dissociation). Phase angle responses correlated strongly with sensory sweetness duration. In natural aged Chinese Baijiu (1-20 years), low frequency phase angle signals positively correlated with sensory lingering sweetness scores, confirming that ethanol cluster accumulation during aging underpins the sweetness of Baijiu. This work expands the classic AH-B sweetness theory from single molecules to supramolecular clusters and provides an analytical platform for probing the temporal dimension of flavor perception.
Science Journals
arXiv:2605.29719v1 Announce Type: new
Abstract: The development of large-scale neuromorphic hardware has made practical implementations of threshold gate-based circuits a near-term possibility. The complexity advantages regarding traditional computing classes, as evidenced in the literature, have prompted us to tackle Epistasis Detection, one of the most computationally complex combinatorial problems in bioinformatics. We propose specially designed circuits that calculate the relative frequencies of all dataset combinations in an efficient pipelined fashion, taking advantage of co-located memory and configurable parallelism, obtaining complexity gains. Overall, we obtain the runtime to be bounded by the number of combinations to calculate, without any additional complexity overhead, contrary to classical approaches, using log-linear space. To accomplish this, we propose a data encoding and combination generation strategy using Leaky Integrate and Fire (LIF) neurons, that feeds a constant depth threshold gate population count circuit. Accounting for typical hardware characteristics, such as limited fan-in and variable precisions, we obtain logarithmic depth and log-cubic linear connections, for the population count circuit by composing developed unbounded fan-in constant depth threshold gate circuits to perform population count and binary array sum.
arXiv:2605.29615v1 Announce Type: new
Abstract: Vision-language models (VLMs) have made strong progress on high-level image-text alignment, yet their ability to perceive subtle visual differences remains limited. We study this problem in rendered web interfaces, where localized visual changes are both a diagnostic test of fine-grained perception and a practical requirement for GUI agents and design tools. We introduce \textbf{DiffSpot}, a code-driven benchmark for open-ended spot-the-difference on web interfaces. DiffSpot constructs controlled image pairs by mutating a single CSS property of a target element in self-contained HTML, re-rendering the page, and recording the changed property, element, and mutation magnitude. A grounding gate retains only pairs whose rendered pixel difference is confined to the target element. The benchmark contains 4{,}400 pairs, including 3{,}900 has-diff pairs balanced across 13 CSS-property operators and three difficulty tiers, plus 500 no-diff pairs for hallucination control. Evaluating 13 frontier VLMs zero-shot, we find that even the best model identifies only $40.7\%$ of true changes, with Hard-tier Recall below $23\%$ for every model. DiffSpot further shows that difficulty is strongly property-dependent: across CSS operators, neither pixel magnitude nor CLIP distance reliably predicts Recall.
arXiv:2605.29749v1 Announce Type: new
Abstract: Identifying subgroups of respondents in psychometric data is traditionally addressed with Latent Class Analysis, which requires the number of classes to be specified a priori and can perform poorly when strong inter-item correlations violate local independence assumptions. We propose a network-theoretic alternative based on community detection in subject-subject similarity networks. To suppress the systematic artifacts induced by the factor structure of the items, the similarity is computed in a low-dimensional factor-score space and the null model for modularity maximisation is obtained by removing the leading (global) mode of the similarity matrix, rather than via the standard Newman--Girvan model. The significance of a detected partition is then assessed against a column-wise resampling null through four complementary observables: the modularity, the differential entropy of the eigenvector point cloud at two neighbourhood scales, and the overlap of the within- and between-community similarity histograms. On a synthetic benchmark with controlled mixture signal, all four metrics correctly identify the homogeneous case as null-compatible -- including the demanding regime of a dataset dominated by a single factor -- and exhibit a graded departure from the null as the cluster separation grows. Applied to 14 widely used psychometric scales, the pipeline isolates a small group of datasets supporting a genuine and directly interpretable modular structure, while the remaining scales fall either in a mixed-signal regime or in one compatible with a single homogeneous community. The significance analysis is independent of the specific community-detection algorithm and provides an operational way to test for modular subject-level structure in questionnaire data.
arXiv:2605.29794v1 Announce Type: new
Abstract: LLM agents now draw on growing skill libraries to handle complex tasks. However, injecting more skills does not always improve task completion and can even degrade it. Existing methods still treat skill injection as a static step, selecting skills with fixed criteria, fixing the budget in advance, and leaving descriptions unchanged. We argue that this static treatment can undermine the utility of skills, because which skills are exposed, how many are included, and how they are presented all affect downstream performance. We propose SkillsInjector, a two-stage adaptive method that jointly addresses these decisions. First, a context planner learns execution-grounded skill preferences and admits an adaptive number of skills for each task. A set-aware renderer then tailors how selected descriptions are presented relative to their co-injected neighbors. Across tau2-bench, SkillsBench, and ALFWorld, SkillsInjector achieves the highest score, improving over the strongest baseline by 3.9, 6.1, and 7.3 percentage points, respectively. Ablation studies show that skill selection, adaptive budgeting, and set-aware rendering each contribute to the gain. These results show that skill-augmented agents benefit from optimizing the injected context itself. Code will be released upon publication
arXiv:2605.30284v1 Announce Type: new
Abstract: Scientific discovery is an inherently creative and uncertain process, requiring reasoning beyond the recall of known knowledge. While many benchmarks have been proposed to evaluate large language model (LLM) performance on deep research tasks via multi-hop retrieval, their innovative reasoning abilities essential for true scientific discovery remain largely untested. We introduce a benchmark framework for evaluating model performance in scientific discovery and reasoning, building up from a raw problem to the classical null hypothesis test. In our framework, models initially receive only the topic and research question from a recent paper, with technical details progressively revealed. At each stage of information disclosure, the model is tasked with generating hypotheses that address the research question, which is compared with the conclusions from the original paper and evaluated via automated semantic similarity of constituent atomic claims. This progressive evaluation of semantic divergence from ground-truth conclusions enables assessment of a model's innovativeness (under minimal information) to grounded reasoning capabilities (under full experimental details), both critical for using LLMs for scientific discovery purposes. Our framework provides a foundation for systematically evaluating scientific reasoning and discovery capabilities in LLMs, crucial for advancing the development of next-generation AI scientist/co-scientist systems. Specifically, here we evaluate GPT-5, GPT-5.4, Gemini 2.5 pro, and Gemini 3.1 pro preview across 45 papers spanning bioactive materials, mechanical materials, and nanomaterials. We find that GPT-5.4 and Gemini 3.1 pro outperform their previous generation counterparts as expected, and GPT-5.4 in particular maintains 0.7 F1 score alignment with ground truth conclusions even under minimal context.
arXiv:2605.30330v1 Announce Type: new
Abstract: Diffusion models have excellent capacity to model complex distributions of natural data, which has made them a popular and effective choice for posterior sampling in imaging inverse problems. Existing methods can incorporate any measurement model at inference time but must use an inexact approximation for the likelihood at intermediate timesteps for computational tractability. Although these approximations can often work well empirically, their downstream effect on the sampled posterior is poorly understood and can result in unexplained failures. To understand when, why, and how these likelihood approximations propagate to erroneous posterior distributions, we introduce a finite-sample perspective on posterior sampling that approximates the posterior to arbitrary precision as training set size tends towards infinity, for any forward model and prior distribution. Using this finite-sample lens, we observe that popular posterior sampling approximations tend to under- or over-estimate the spread of the posterior at intermediate timesteps, causing downstream consequences including sensitivity to early stopping time, inaccurate relative weighting of posterior modes, and hallucination, both of prior modes that are not in the posterior and likelihood modes that are not supported by the prior. Moreover, we find that the cause of these posterior errors requires neither a nonlinear measurement model nor a multimodal posterior, but can arise solely due to a multimodal prior and inaccurate posterior spread at intermediate sampling times. Our finite-sample posterior sampling approach is agnostic to the type of likelihood approximation and the type of (linear or nonlinear) forward model, and can thus serve as a drop-in diagnostic to evaluate the accuracy and failure modes of existing and future posterior samplers.
arXiv:2605.30341v1 Announce Type: new
Abstract: Studying scalable methods for visual generative modeling requires large, accessible, and stable datasets. We introduce GPIC, a Giant Permissive Image Corpus of approximately 28 trillion pixels. GPIC comprises diverse internet images captioned by a state-of-the-art vision-language model, including 100M training, 200K validation, and 1M test examples. Moreover, all GPIC images are permissively licensed for both research and commercial use. GPIC is safety-filtered, deduplicated, and centrally hosted on Hugging Face. We provide a benchmarking protocol for generative modeling on GPIC. Finally, we provide a reference baseline for pixel-space flow matching on GPIC. Our dataset, benchmark, and models are available at https://huggingface.co/datasets/stanford-vision-lab/gpic. Evaluation toolkit and code are available at https://gpic.stanford.edu
arXiv:2605.30349v1 Announce Type: new
Abstract: Autoregressive video diffusion models generate streaming video by producing frames sequentially, conditioning each chunk on previously generated content. These models are structurally anchored to the first frame: its key-value representation occupies a privileged position in the attention cache and serves as the primary scene reference throughout generation. As the cleanest and most error-free position in the cache, this anchor draws disproportionate attention, suppressing video dynamics, and locking scene composition to the initial viewpoint even as the scene naturally evolves. The result is a temporally shallow video in which motion, camera movement, and scene progression are dampened in favor of static consistency. To address this, we replace the static anchor with an adaptive state, a hidden latent that the model denoises alongside content at every chunk but never renders. Rather than referencing a frozen first frame, the model generates its own scene anchor at each step by attending to both the previous state and the current content, producing a reference that evolves with the generated content. Unlike standard video generation, which encodes an absolute notion of time, our formulation treats time as relative: every generation step sees the same positional structure regardless of how far generation has progressed, and the state transition is identical at every chunk. Together, these properties introduce a recurrence into the generation process, where denoising serves as the transition function, and the KV cache serves as the carrier, requiring no external module. Experiments demonstrate that the adaptive state substantially improves video dynamics, enabling richer motion and natural scene progression within generated videos.
arXiv:2605.30214v1 Announce Type: new
Abstract: Third-person singular pronouns have long been used to study stereotypical biases in language models and to test their abilities to reason about reference. More recently, the interplay between reasoning and bias has been investigated with the task of pronoun fidelity, which assesses models' abilities to correctly reuse a previously-specified pronoun for a discourse entity, independent of other potentially distracting discourse entities mentioned in between. However, such research focuses on English, which is a language with limited grammatical gender and almost no gender agreement. In this paper we contribute a novel, large-scale dataset, GRUFF, to measure pronoun fidelity in German, covering four different gender agreement systems in nouns, and four sets of pronouns. With this dataset, we show that LLMs show strong grammatical agreement for masculine and feminine entities in the absence of explicit context, but not for neopronouns xier and en. Models are generally not robust to distractors, but encoder-only models are more robust in German than in English, reflecting the importance of grammatical gender. Finally, we show that occupational stereotypes in this context are poorly correlated across grammatical cases, and across most models, except ones with closely related architectures. We release all code and data to encourage further work on gender-inclusive language and referential reasoning in German.
arXiv:2605.30272v1 Announce Type: new
Abstract: Physics-informed neural networks (PINNs) formulate the solution of partial differential equations as residual minimization problems over neural network parameterizations. Although highly flexible, optimization of PINNs using modern variants of Stochastic Gradient Descent algorithms is expensive. On the other hand, iterative computation of PINN parameterization using the Gauss-Newton method suffers from convergence difficulties, dense Jacobian structures, and poor conditioning that limit the effectiveness of second-order optimization methods. In this work, we introduce IGA-ODIL, a spline-based residual minimization framework combining ideas from Optimizing DIscrete Loss (ODIL), robust variational residual minimization, and Isogeometric Analysis (IGA). Instead of neural-network parameterizations of PINNs, the unknown solution is represented by smooth B-spline basis functions, leading to sparse structured Jacobians and efficient Gauss--Newton optimization. We also derive robust residual formulations based on weighted Gram operators, making the loss function related with the true error. The resulting systems inherit locality, sparsity, and approximation-theoretic properties of classical finite element and isogeometric methods while preserving the residual-learning philosophy of scientific machine learning. The proposed methodology is evaluated on several benchmark problems, including Poisson equations, convection-dominated advection--diffusion equations, Helmholtz problems with highly oscillatory solutions, nonlinear Allen--Cahn equations, and inverse Helmholtz parameter identification. Numerical experiments demonstrate orders-of-magnitude speedups compared with PINNs and CRVPINNs while maintaining high accuracy and robustness.
arXiv:2509.23571v3 Announce Type: replace
Abstract: As cyber threats continue to grow in scale and sophistication, blue team defenders increasingly require advanced tools to proactively detect and mitigate risks. Large Language Models (LLMs) offer promising capabilities for enhancing threat analysis. However, their effectiveness in real-world blue team threat-hunting scenarios remains insufficiently explored. This paper presents CyberTeam, a benchmark designed to guide LLMs in blue teaming practice. CyberTeam constructs a standardized workflow in two stages. First, it models realistic threat-hunting workflows by capturing the dependencies among analytical tasks from threat attribution to incident response. Next, each task is addressed through a set of operational modules tailored to its specific analytical requirements. This transforms threat hunting into a structured sequence of reasoning steps, with each step grounded in a discrete operation and ordered according to task-specific dependencies. Guided by this framework, LLMs are directed to perform threat-hunting tasks through modularized steps. Overall, CyberTeam integrates 30 tasks and 9 operational modules to guide LLMs through standardized threat analysis. We evaluate both leading LLMs and state-of-the-art cybersecurity agents, comparing CyberTeam against open-ended reasoning strategies. Our results highlight the improvements enabled by standardized design, while also revealing the limitations of open-ended reasoning in real-world threat hunting.
arXiv:2605.30294v1 Announce Type: new
Abstract: We present RaFI, a CUDA and MPI based software framework that simplifies the task of building GPU-enabled data-parallel software where rays or similar work items need to migrate between different GPUs. RaFI provides a simple interface for CUDA kernels to forward such work items to other GPUs, while under the hood managing all the CUDA and MPI related work required to make this happen. We describe RaFI's motivation and implementation, and show its potential in several example applications.
arXiv:2510.09831v2 Announce Type: replace
Abstract: Permanent magnet electron optics offer many advantages over electromagnets, and are being increasingly used in high-energy (GeV) electron accelerator designs. Here, we identify the advantages of permanent magnet electron optics for low-energy (keV) electron accelerators. We explore the applications of a class of designs based on axially magnetized permanent magnets, which offer a variety of advantages such as short focal lengths (few mm), while also preventing apparent emittance growth resulting from starting particles in a magnetic field. The proposed design philosophy is applied to an accelerator based on the ultracold electron source. The design is shown to be `emittance preserving' even for very short focal lengths (~5 mm) at an emittance level better than 1 nm-rad, while the short beamline (12 cm) limits space-charge effects. Two remedies for the mitigation of typical manufacturing and alignment challenges are considered. The performance of the design (related to parasitic aberrations) is enhanced by the proposed techniques. Applications of this design philosophy can improve the performance of ultrafast electron diffraction setups with minimal manufacturing effort.
arXiv:2605.30308v1 Announce Type: new
Abstract: Modern table formats such as Apache Iceberg compute and store metadata-commit timestamps, record counts, and column-level statistics such as null counts and value bounds at write time as part of file writing. These statistics serve query planning, yet they overlap substantially with data quality (DQ) monitoring needs. We describe a metadata-first approach that repurposes write-time statistics for continuous DQ observability: anomaly detection, drift monitoring, null-rate tracking; without scanning any data. Deployed at LinkedIn across 200,000+ Iceberg tables (800+ PB), this approach satisfies approximately 60% of user-defined DQ rules at zero marginal compute cost and reduces profiling resource consumption by around 50%. Extending manifest statistics with lightweight counters (sum, zero-value counts, boolean counts) and incrementally mergeable sketches; Theta sketches for distinct counts, KLL sketches for quantiles; can further raise metadata-satisfiable coverage to close to 90% of production DQ rules. We validate sketch accuracy, mergeability, and storage overhead on production data and propose that table formats should store per-file sketches in Puffin sidecar files, following the same store-then-aggregate pattern used for existing manifest statistics.
arXiv:2511.04934v3 Announce Type: replace
Abstract: Unlearning in large language models (LLMs) is critical for regulatory compliance and for building ethical generative AI systems that avoid producing private, toxic, illegal, or copyrighted content. Despite rapid progress, in this work, we show that \textit{almost all} existing unlearning methods fail to achieve true forgetting in practice. Specifically, while evaluations of these `unlearned' models under deterministic (greedy) decoding often suggest successful knowledge removal using standard benchmarks, we show that sensitive information reliably resurfaces when models are sampled with standard probabilistic decoding. To rigorously capture this vulnerability, we introduce \texttt{leak@$k$}, a new meta-evaluation metric that quantifies the likelihood of forgotten knowledge reappearing when generating $k$ samples from the model under realistic decoding strategies. Using three widely adopted benchmarks, TOFU, MUSE, and WMDP, we conduct the first large-scale, systematic study of unlearning reliability using \texttt{leak@$k$} metric. Our findings demonstrate that knowledge leakage persists across methods and tasks, underscoring that current state-of-the-art (SOTA) unlearning techniques provide only limited forgetting. We propose an algorithm, termed Robust Unlearning under LEak@$k$ metric (\texttt{RULE}) to address this concern. We demonstrate that \texttt{RULE} provides an unlearned model for TOFU benchmark with no information leakage for a large number of generation samples. On the MUSE benchmark, \texttt{RULE} outperforms SOTA unlearning methods under the \texttt{leak@$k$} metric across most sampling budgets $k$. Codes are available at https://github.com/OptimAI-Lab/Leak-k.
arXiv:2605.30312v1 Announce Type: new
Abstract: Differentially private (DP) image synthesis generates images that preserve the statistical characteristics of a sensitive dataset, enabling sensitive data analysis and usage while providing rigorous guarantees of privacy leakage. Existing methods fine-tune public models using DP Stochastic Gradient Descent (DP-SGD) on sensitive images to generate synthetic images. But full fine-tuning public models on sensitive images is computationally expensive, because current public models typically contain a large number of parameters. Recent work proposes heuristically using Low-Rank Adaptation (LoRA) on all attention-layer parameters of public models to reduce the number of trainable parameters. However, we argue that exhaustive LoRA coverage across all attention-layer parameters is suboptimal in a DP setting, as it leads to noise accumulation and collapse during private training.
To address this issue, we propose DP-SAPF, which uses a saliency-aware strategy to identify specific target parameters for LoRA training under DP. DP-SAPF is inspired by the fact that larger gradients signify higher saliency, indicating that these parameters are most critical for the DP learning. Specifically, we feed the sensitive images into public models, compute gradients, and add noise to the gradients to satisfy DP. Then, DP-SAPF identifies the most salient parameters, those exhibiting high gradient magnitudes on sensitive images, for DP fine-tuning. Experiments on four sensitive image datasets show that DP-SAPF improves the utility and fidelity of synthetic images while requiring fewer computational resources than fine-tuning methods without parameter selection.
arXiv:2605.30314v1 Announce Type: new
Abstract: Software engineering (SWE) agents are transitioning from code generation to full software development lifecycle automation. A critical phase in this lifecycle is specification design: transforming initial proposals into carefully considered requirements through expert review. Existing benchmarks such as SWE-Bench are implementation-focused by measuring the agent's ability to generate code given fixed, precise design requirements. This formulation assumes specifications are correct and complete. In real-world complex and critical software systems, initial specifications are often incomplete and flawed, requiring extensive expert reviews and revisions before being accepted for implementation. To fill this gap, we introduce SpecBench to evaluate specification-level reasoning: the ability to generate complete, unambiguous, consistent, and correct system specifications. SpecBench tasks are derived from the Request for Comments (RFC) process used by mature open-source projects. For each task, an agent is given an initial design proposal, the project codebase, and all past project RFC discussions. The agent is tasked with identifying specification deficiencies: omissions, ambiguities, inconsistencies, or incorrect assumptions in the initial proposal. We evaluate predictions against critiques raised by expert maintainers during historical RFC reviews. SpecBench contains tasks from 5 diverse repositories: Kubernetes, React, Rust, TVM, and vLLM. We evaluate state-of-the-art SWE agents on SpecBench, analyzing their capacity to reason about system design without execution feedback. The best performing agent, GPT-5.4, achieves 44.4% accuracy.
arXiv:2502.10330v4 Announce Type: replace
Abstract: Recent advances in diffusion models show promising potential to accelerate nonconvex problem solving by leveraging their multimodality. However, most existing diffusion-based optimization approaches rely on supervised learning and lack a mechanism to enforce constraint satisfaction, which is required in real-world applications. In that case, we investigate and theoretically analyze the inherent problem of supervised diffusion solvers and identify the distributional misalignment problem, i.e., the generated solution distribution often exhibits low probability mass on the feasible region. To resolve this issue, we propose DiOpt, a new diffusion-based learning framework for constrained nonconvex optimization, which effectively learns the mapping from noise to the constraint region. Specifically, this framework operates in two distinct phases: an initial warm-start phase, implemented via supervised learning, followed by a bootstrapping training phase. This dual-phase architecture is designed to iteratively refine solutions, thereby improving the objective function with high constraint satisfaction. Finally, we also employ a solution selection technique in inference for better optimality. Notably, DiOpt is the first successful integration of the diffusion solver in constrained nonconvex optimization. Evaluations on diverse nonconvex tasks demonstrate the superiority of DiOpt in both optimality and constraint satisfaction. Our official page is released at https://dingsht.tech/diopt-webpage.
arXiv:2502.12385v4 Announce Type: replace
Abstract: The rising complexity of photonic applications, ranging from quantum computing to neuromorphic processing, has driven the demand for highly programmable and scalable photonic integrated circuits. While mesh-based architectures built from Mach-Zehnder interferometers have enabled significant advances, their reliance on beam splitting and light bending introduces optical loss, fabrication challenges, and scalability bottlenecks. Continuously lateral-coupled integrated waveguide arrays (WAs), by contrast, offer compact systems with no direct free-space analogs, but their static nature has limited their utility. Recently, programmable waveguide arrays (PWAs) have emerged as a promising alternative, combining the Hamiltonian richness of WAs with tunable control. This perspective outlines the conceptual foundations, recent progress, and future potential of PWAs across quantum simulation, photonic neural networks, topological photonics, and nonlinear optics. We examine the theoretical and practical challenges of modeling, fabrication, and control, and propose PWAs as a next-generation architecture for compact, reconfigurable photonic processors.
arXiv:2506.06254v2 Announce Type: replace
Abstract: Large Language Model (LLM) empowered agents have recently emerged as advanced paradigms that exhibit impressive capabilities in a wide range of domains and tasks. Despite their potential, current LLM agents often adopt a one-size-fits-all approach, lacking the flexibility to respond to users' varying needs and preferences. This limitation motivates us to develop PersonaAgent, the first personalized LLM agent framework designed to address versatile personalization tasks. Specifically, PersonaAgent integrates two complementary components - a personalized memory module that includes episodic and semantic memory mechanisms; a personalized action module that enables the agent to perform tool actions tailored to the user. At the core, the persona (defined as unique system prompt for each user) functions as an intermediary: it leverages insights from personalized memory to control agent actions, while the outcomes of these actions in turn refine the memory. Based on the framework, we propose a test-time user-preference alignment strategy that simulate the latest n interactions to optimize the persona prompt, ensuring real-time user preference alignment through textual loss feedback between simulated and ground-truth responses. Experimental evaluations demonstrate that PersonaAgent significantly outperforms other baseline methods by not only personalizing the action space effectively but also scaling during test-time real-world applications. These results underscore the feasibility and potential of our approach in delivering tailored, dynamic user experiences.
arXiv:2605.27959v2 Announce Type: replace
Abstract: Multimodal Large Language Models (MLLMs) have increasingly localized and interleaved visual evidence for deliberative reasoning. Grounding-based approaches typically focus on regions of interest (RoIs) by injecting cropped image patches or RoI-specific features into the reasoning context. However, such designs can weaken holistic scene understanding and inter-object relations, while incurring decoding costs that scale with the number and size of RoIs. Alternatively, adaptive visual feature selection often requires fine-grained supervision or complex heuristics. To address these limitations, we propose ROVER (Routing Object-centric Visual Evidence for grounded multi-image Reasoning), a lightweight, learnable plugin for efficient global visual evidence routing. Upon each object grounding prediction, ROVER injects a step-specific token triplet to synergistically: (i) aggregate the ongoing reasoning context, (ii) distill intra-image cues into a visual working space via object-centric differential attention, and (iii) route and integrate history-aware evidence across objects and images within this space for subsequent reasoning. We integrate ROVER into Qwen2.5-VL-7B and develop an interleaved SFT-to-GRPO training pipeline. Strictly adhering to the original datasets and evaluation protocols, our method achieves the best performance on MM-GCoT (+4.8% answer accuracy, +14.6% grounding accuracy) and VideoEspresso (+8.6% answer accuracy). The VideoEspresso-trained model demonstrates strong transferability, outperforming the base model by +4.7% on average across diverse benchmarks.
arXiv:2512.10401v3 Announce Type: replace-cross
Abstract: This paper is concerned with differentiable resampling in the context of sequential Monte Carlo (e.g., particle filtering). Drawing on reparametrisation, we propose a new resampling method that is informative and instantly differentiable, based on a training-free diffusion model surrogate. We theoretically prove that our diffusion resampling method provides a consistent resampling distribution, and we show empirically that it outperforms the state-of-the-art differentiable resampling methods on multiple filtering and parameter estimation benchmarks. Finally, we show that it achieves competitive end-to-end performance when used in learning a complex dynamics-decoder model with high-dimensional image observations.
arXiv:2510.14064v3 Announce Type: replace
Abstract: Vacuum magnetic birefringence (VMB) is an effect predicted by quantum electrodynamics, in which the vacuum behaves as a non-linear optical medium and exhibit a birefringence in the presence of a magnetic field. In this work, an interferometric scheme is introduced to measure this effect for the first time by sensing the changes in the frequencies of fields stabilized to the resonances of an optical cavity. Results are presented from a prototype setup with this novel sensing technique implemented on a 19 m test cavity without a magnetic field. We propose using this scheme for a measurement of VMB with a 245 m long optical cavity and a string of 24 superconducting magnets, arranged for the ALPS II experiment. In this manuscript we examine potential sources of noise in the prototype, and project these results in terms of the sensitivity of the full-scale experiment.
arXiv:2510.20743v2 Announce Type: replace
Abstract: We present Empathic Prompting, a novel framework for multimodal human-AI interaction that enriches Large Language Model (LLM) conversations with implicit non-verbal context. The system integrates a commercial facial expression recognition service to capture users' emotional cues and embeds them as contextual signals during prompting. Unlike traditional multimodal interfaces, empathic prompting requires no explicit user control; instead, it unobtrusively augments textual input with affective information for conversational and smoothness alignment. The architecture is modular and scalable, allowing integration of additional non-verbal modules. We describe the system design, implemented through a locally deployed DeepSeek instance, and report a preliminary service and usability evaluation (N=5). Results show consistent integration of non-verbal input into coherent LLM outputs, with participants highlighting conversational fluidity. Beyond this proof of concept, empathic prompting points to applications in chatbot-mediated communication, particularly in domains like healthcare or education, where users' emotional signals are critical yet often opaque in verbal exchanges.