Science & Mathematics

Despite achieving high visual fidelity, video models still suffer from physical hallucinations, limiting their reliability as world simulators. Existing methods leverage model-internal or external signals to improve physical plausibility, but their roles and limitations require a unified explanation. We propose a two-dimensional physical-hallucination risk framework that defines low-density sampling risk and high-density drifting risk. The first risk captures generations assigned low likelihood by the video model, for example, a ball suddenly teleporting despite the model favoring continuous motion, while the second risk captures physical inconsistencies in high-likelihood generations, for example, a ball bouncing progressively higher without additional energy. Our analysis explains how model-internal signals steer sampling toward higher-density regions of the learned distribution to address low-density sampling risk. However, higher model likelihood does not guarantee physical plausibility, motivating external guidance to address high-density drifting risk. Based on this framework, we develop Hallucination-Aware Sampler (HA-Sampler), which applies external physical guidance within latent refinement to generate physically plausible videos, and HA-Scorer, which derives a likelihood-based signal from the generator itself for evaluating generated videos using continuous normalizing flow (CNF), with coarse ODE integration and finite-difference divergence estimates. Experiments on Physics-IQ and VideoPhy benchmarks support our theoretical analysis and demonstrate that our methods outperform the compared sampling and scoring SOTA baselines.

Open resources for computer use agents (CUAs) -- environments, datasets, and models -- remain scattered: without a standard, these resources cannot be integrated and scaled; additionally, evaluation, RL, and SFT cannot share one infrastructure. We present CUA-Lite, an open platform for developing CUAs, built around three standardized abstractions: (1) Unified environment interface, under which we integrate 12 benchmarks across desktop, browser, and mobile, plus lightweight VM-free sandboxes that replicate OSWorld's desktop at much lower cost, providing over 30,000 verifiable tasks. (2) Unified supervised data format, into which we convert the major public CUA datasets, from GUI grounding and understanding to full rollouts, and record our fresh rollouts from frontier CUAs on these verifiable environments. (3) Model harnesses shared across evaluation, RL, and SFT, one per model and implemented for 14 model families. Each harness runs its model in these computer environments, producing the rollouts that evaluation and RL consume, or renders stored supervised data into the model's own training format for SFT. We release reference scores across the integrated benchmarks, and show that RL and SFT through the same harness improve open-weight models. We release our dataset, code, and models, and the platform is open to contributions.

Recently, Allamigeon, Dadush, Loho, Natura, and Végh (ADLNV, SICOMP 2025) introduced an interior point method for linear programs whose number of iterations is bounded by the so-called straight line complexity (SLC) of the instance. This yields a strongly polynomial bound on the number of iterations of the corresponding interior point algorithm for broad classes of linear programs. In particular, in the companion paper we give a strongly polynomial straight line complexity bound for two-variable-per-inequality linear programs. However, in order to obtain a strongly polynomial algorithm for these classes, one also needs to address two important technical issues: initialization and bit-complexity bounds. Standard IPM initialization techniques either require numerical estimates that cannot be obtained in the strongly polynomial model, or involve modifying the instance in ways that do not preserve the geometry of the original LP, potentially affecting the SLC. In this work, we present an initialization framework for interior point methods for linear programming, along with a rounding procedure that ensures the continued feasibility of the IPM. We develop a multistage scheme in which each stage can be initialized in strongly polynomial time given the output of the previous stage. Additionally, we address the challenge of controlling the bit complexity of each iterate throughout the algorithm. To this end, we show that it is possible to represent iterates as low-complexity convex combinations of vertices and extreme rays. Our combinatorial rounding step is independent of the algorithm in ADLNV 2025, and can be applied more broadly in the context of path-following methods.

This work presents a novel open-source toolkit designed for the evaluation of face deidentification techniques with available evaluation models. The framework provides a complete evaluation pipeline, including preprocessing, deidentification, and performance assessment, based on data utility, image quality, and identity protection metrics. It supports a variety of datasets, enabling researchers to easily benchmark new deidentification techniques. Results show the effectiveness of some techniques on maintaining privacy while preserving image utility. The toolkit is aimed at unifying future research by providing a standard evaluation framework for privacy-enhancing technologies in facial biometrics. Implementation is available at github.com/blazm/deid-toolkit.

Video overlay removal aims to recover scenes obscured by subtitles, titles, and logos while preserving surrounding content. Existing methods use temporal propagation or source-conditioned generation, but can leave residual overlays and distort scene structure when useful source information is obscured. We propose DeOverlay, a source-anchored flow model that reconstructs the clean video from the source latent with fixed source guidance. Specifically, we train a shared velocity network along paired source-to-clean paths using region-weighted flow matching, enabling successive cor- rections from intermediate states. To select local source cues, we develop visibility-normalized temporal aggregation that excludes masked positions and averages the remaining observations. Global source context and edit-region embeddings complement these cues through additive condi- tioning, retaining a single video-token sequence. To evaluate scene recovery across diverse overlay conditions, we introduce DVTE-Bench, comprising 1,631 paired videos across eight text-overlay categories, with OCR-screened clean targets, frame-aligned pixel masks, and a common evaluation protocol. Experiments on the complete release show that DeOverlay outperforms four baselines on all five reconstruction and perceptual metrics, reducing Mask LPIPS and Mask VFID by 58.3% and 70.3%, respectively, relative to the strongest evaluated baseline. Additional qualitative examples demonstrate the removal of graphic logos and stickers. We will make the complete benchmark, evaluation code, and test outputs availabl. Project page:https://yahooo-m.github.io/DVTE-Bench/

Block-scaled quantization formats such as OCP Microscaling (MX) and NVFP4 have been widely adopted as efficient data formats for LLM inference. However, the design space of these formats spanned by shared scale, weight, and activation format combinations has not been systematically explored, and the joint accuracy-silicon tradeoff across this space remains uncharacterized, leaving the true Pareto-optimal configurations for NPU-based edge LLM deployment an open question. To bridge this gap, we conduct a joint hardware-software (HW-SW) co-design exploration of the above underexplored MX-like design space with three innovations. 1) We define and systematically characterize the full 8-bit scale format family within the MX-like design space, jointly with asymmetric weight-activation precision, identifying configurations that outperform the MX and NVFP4-like (without per-tensor scaling, PTS) baselines. 2) We build a per-format specialized MAC datapath template that enables fair hardware cost evaluation across the space. 3) We develop an automated co-design flow, integrated into the open-source SNAX NPU platform, that enables joint HW-SW analysis across the design space. Overall, our co-design exploration reveals that scale format matters only under aggressive FP4 quantization, yielding a case-oriented design guideline for both scale format and asymmetric weight-activation precision. We further identify UE5M3 as a favorable accuracy-hardware tradeoff for the FP4 regime, reducing perplexity degradation by up to 25.3% relative to the NVFP4-like baseline, at a cost of only +0.06% area and +0.49% power overhead.

EEG-eye movement emotion recognition can support mental-health monitoring and adaptive interfaces, but most fusion models treat the two signals as parallel features, classify each window alone and degrade when eye data are lost. We propose the Context-Aware Topological Distillation Graph Network (CTD-GNet), in which EEG features form the nodes of an electrode graph, and eye movements and preceding windows modulate its edges, nodes and readout, the edges through electrode-specific prompts. Bidirectional distillation of the graph, node relations and predictions couples this multimodal teacher with a single-window EEG student, improving the teacher and aligning the two graphs 35-fold. In leave-one-subject-out tests on SEED and SEED-IV against 17 baselines, CTD-GNet has the highest mean accuracy, 84.91% and 76.17%, significantly above 9 and 6 of 13 eye-based baselines after Holm correction, the smallest cross-subject spread among multimodal models and the highest worst-subject accuracy. It is the most accurate model at every rate of missing eye data on SEED-IV, keeps 66.1% on SEED without eye data, more than any EEG-only model, and stays 9.3 and 11.9 points above all single-window fusion models under eye-movement noise of one standard deviation. On film clips unseen in training, it leads on SEED with 77.1%.

Tibetan speech recognition is usually evaluated with character or syllable error rates because the writing system does not mark word boundaries. These metrics can rank systems differently and respond differently to adaptation failures. We combine character error rate, syllable error rate, and segmented word error rate in one evaluation framework. We use dictionary-based, neural, and language-model-based segmentation and compare their boundaries with expert annotations. We evaluate five ASR architectures on Lhasa Tibetan, then test zero-shot transfer and target-corpus adaptation on five additional corpus subsets. We also vary LoRA rank and training-data size to examine adaptation failure. The results show that a lower character error does not always lead to a lower syllable or segmented-word error. These ranking differences persist across corpora. Metric correlations also weaken in some unsuccessful adaptation settings. Higher LoRA ranks improve syllable and segmented-word scores but remain worse than full fine-tuning in the tested range. A standardized phonological representation provides another view of spelling errors. These findings support evaluation at several linguistic levels and motivate better segmentation and adaptation methods for low-resource ASR.

Embodied VLMs can execute increasingly complex manipulation tasks, yet execution errors can leave the environment in states that require replanning for recovery. Existing recovery methods either rely on inference-time mechanisms with fixed policies or learn from failure examples and reference corrections rather than verified action outcomes. To enable recovery learning through verifiable interaction, we introduce the Embodied Replanning for Recovery Environment (ERR-Env), which provides feedback on task progress and execution errors. We further construct ERR-Bench, a benchmark with held-out tasks spanning nominal, intermediate, and error states for evaluating recovery capability and execution reliability. We then propose Action-Grouped Turn-GSPO (AGT-GSPO), an online reinforcement learning method combining turn-level rewards for task progress and error states with action-grouped advantage estimation, and use it to train ERR-VLM, a compact 2B embodied VLM. Evaluation on ERR-Bench reveals a mismatch between task progress and execution reliability across diverse VLMs. AGT-GSPO improves ERR-VLM by 10.25 percentage points in Strict Success Rate for error-free completion and 22.95 percentage points in Recovery Rate, while ablations and zero-shot transfer further validate the effectiveness of our method.

Autonomous clinical-modeling agents can optimize predictive performance while violating conditions that make a model usable. We introduce LeakBench, an executable, guideline-derived audit suite that tests agent decisions for prediction-time leakage, calibration, site-level evaluation, support awareness, and decision-curve net benefit. LeakBench combines a synthetic task with known failure structure and two real-data settings: MIMIC-IV (430k admissions) and governed multi-hospital inpatient record front-page (MHFP) data. Our primary analysis uses reward hacking descriptively, without attributing intent: an agent selects prediction-time-invalid inputs because they improve its validation reward. A deterministic optimizer and every evaluated unconstrained local-model configuration select a fatal leaking feature. In fully logged September 2026 A/B runs, DeepSeek Flash and GPT-6-astra each leak in 10/10 unconstrained runs and 0/10 runs with an explicit admission-time constraint. This contrast is version-sensitive: a blinded Claude/Fable-5.1 chat run does not leak in either condition, while explicit-constraint violations appear only in a historical August 2026 slice lacking raw trajectories. Replacing AUROC with AUPRC does not remove leakage selection. Renaming or disguising the leaking inputs changes neither selection nor provenance-based detection; a name-based filter fails. A deterministic provenance verifier-in-loop removes the leak, making the LLM repair loop defense-in-depth. Specific verifier feedback clears fatal leakage in a hosted agent, and a three-seed ablation finds principle-only feedback clears decision-level defects, though these mitigation results are limited. LeakBench treats trustworthy clinical prediction as a property of an agent's modeling process, not merely its score; it is an audit testbed with known-failure controls, not yet an expert-validated benchmark.

Bounded model checking unrolls a program into a single formula, so a function holding a loop costs the solver a fresh copy of that loop at every call site. Function contracts break that coupling. Verify the body once against its contract, then replace each call by a stub derived from it. CBMC supports this through dynamic frame condition checking, applied at industrial scale, but with no formal account of why the replacement is sound and no characterisation of what the frame check costs. We present the function-contract framework for ESBMC. The clause syntax, both phases, and the first assigns enforcement the tool has had are all introduced here. Its technical core is an observational frame rule: snapshot the locations the contract does not permit the function to write, then compare them at exit. The encoding emits a number of verification conditions set by how many globals the program has, not by how many writes the body performs, while dynamic frame condition checking grows with the write count. That is a property of the encoding rather than a timing, so it moves with neither the machine nor the solver. The rule itself is the classical one; what we add is its bounded form inside a BMC encoding, with the pipeline's algebraic skeleton machine-checked in Lean 4 and the limits of that model stated. We evaluate on synthetic sweeps, SV-COMP tasks, an ANSSI X.509 parser, the FreeRTOS heap_1 allocator, and the ML-KEM reference implementation whose CBMC contracts we reuse unchanged. At the coefficient count FIPS-203 mandates, neither tool finishes on the monolithic route inside ten minutes, and both finish on the modular one. The framework is merged into ESBMC master. Contracts are what is left when unrolling runs out of room.

Autonomous vehicles must detect objects faster than a frame camera delivers images. Event cameras record brightness changes with microsecond resolution, so RGB--event detectors can decide between frames, at an operating frequency, the rate of decisions, far above the frame rate. As this rate rises, each decision receives a sparser event window, and the reliability of the events relative to the frame varies over the scene and over time. Existing detectors weight the sensors per sample and encode the frame at the reduced event resolution. Reliability-supervised routing assigns a fusion weight to every location and supervises it with targets from the losses of two single-modality classifiers used only in training, moving loss-based supervision of fusion weights from whole samples to individual locations. Asymmetric-resolution fusion lets the frame carry the decision where the events fail: it encodes the frame at native resolution and pools every feature level onto the event grid, reusing all pretrained weights. On the DSEC-Det dataset, routing halves the accuracy loss from 20 to 180 Hz, from 6.3 to 3.1 mAP, and lowers it in every seed. The resulting detector, ERA-Det, sets a new state of the art at 63.2 mAP, 3.7 mAP above FlexEvent, leads at every operating frequency, and at 180 Hz is more accurate than FlexEvent at 20 Hz. On two DAVIS346 datasets, the router lowers the frame weight at night and in low light and raises it when the events vanish. ERA-Det upgrades released FlexEvent weights in 5,000 iterations on one GPU.

Disentangled real-image manipulation requires factor-specific semantic control while preserving image fidelity and non-target content. StyleSpace $S$ provides interpretable controls but is limited by GAN inversion, whereas diffusion models offer stronger image realization without an explicit factorized control basis. We introduce GeoNaviDiff, an unsupervised cross-space framework that transfers disentangled StyleSpace interventions to local Riemannian trajectories in diffusion $h$-space. GeoNaviDiff constructs reconstruction-decoupled supervision by extracting a spherical Log-map tangent from a matched StyleGAN reconstruction-edit pair and parallel-transporting it to the real-image diffusion state. A shared, factor-conditioned Navigator predicts this tangent at inference, while norm-preserving geodesic realization and source-anchored rollout stabilize the reverse process without edited references or factor-specific models. Experiments across multiple datasets show that GeoNaviDiff outperforms the compared methods in fidelity, factor selectivity, edit locality, and trajectory smoothness.

Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) achieve high information transfer rates but often require sufficient subject-specific calibration data for reliable decoding. Analytical spatial filters are effective with limited data, but their linear formulations constrain representation capacity. Deep models learn more flexible representations but are prone to overfitting when calibration data are scarce. We propose KDHNet, which combines knowledge-driven analytical processing with data-driven representation learning. After filter-bank preprocessing, cascaded task-discriminant component analysis (TDCA) and task-related component analysis (TRCA) extract discriminative and trial-reproducible components. Learnable sub-band weighting then fuses spectral information, and an SSVEPformer decoder learns temporal and spectral representations for classification. For inter-subject decoding, a target-guided least-squares transformation aligns source trials using one labeled target trial per class. Extensive experiments were performed on three public datasets, covering intra- and inter-subject protocols, five analysis windows, and multiple calibration ratios. In intra-subject evaluation, KDHNet achieved the highest mean accuracy and information transfer rate in 14 of 15 window settings and all ten calibration-ratio settings. In inter-subject evaluation, KDHNet obtained the best or second-best performance, and the gains were generally larger with less calibration data. Ablation studies indicated that the estimated TDCA projection performed better when fixed than when optimized from analytical or random initialization. Combining fixed knowledge priors with trainable neural modules is effective for low-calibration SSVEP decoding. This framework supports more practical, robust, and plug-and-play BCIs. Code is available at https://github.com/2665023991/KDHNet.

Radiomic survival studies in glioblastoma report heterogeneous prognostic features, but how much of this stems from the methodology itself is rarely examined. We analysed the association of 463 radiomic, anatomical and clinical features from pre-treatment MRI with overall survival in 705 IDH-wildtype glioblastoma patients from three public cohorts, using three techniques: a log-rank test after median split, univariate Cox regression (both Benjamini-Hochberg corrected) and an XGBoost regressor interpreted with SHAP. The log-rank test selected 99 features, Cox regression 72 and SHAP its top 20; five were selected by all three: age, a peritumoural T2 texture feature, the number of connected peritumoural mask components and two image-level intensity maxima. The two survival tests shared 51 features, about half of the SHAP top 20 were also selected by either test, and where selections overlapped, effect directions agreed. Many selections were PyRadiomics diagnostic descriptors that are not restricted to tumour voxels. Analytical pipeline choices, including ranking method, pre-filtering and descriptor handling, can substantially influence which features appear prognostic.