Agent skills today are static artifact: authored once -- by human curation or one-shot generation from parametric knowledge -- and then consumed unchanged, with no mechanism to improve from real use. We propose \textbf{SkillEvolver}, a lightweight, plug-and-play solution for online skill learning, in which a single meta-skill iteratively authors, deploys, and refines domain-specific skills. The learning target of SkillEvolver is the skill's prose and code, not model weights, so that the resulting artifact drops into any agent without retraining; and the meta-skill itself is just another skill, loaded through the same interface by any protocol-compliant CLI-agent. Unlike trace-distillation, the meta-skill refines only after deploying the learnt skill, such that the learning signal comes from failures another agent encounters while using it -- not from exploratory traces alone. Refinement iterations are governed by a fresh-agent overfit audit that catches possible leakage as well as deployed-skill-specific failures, including the silent-bypass mode in which a skill appears valid in content but is never invoked at runtime. On $83$ SkillsBench tasks spanning $15^{+}$ domains, SkillEvolver reaches $56.8\%$ accuracy versus $43.6\%$ for curated human skills and $29.9\%$ for the no-skill baseline; on three GPU kernel optimization tasks from KernelBench, it also raises mean speedup from $1.16$ to $1.51$ on average.

Long-tailed distributions in class-imbalanced data present a fundamental challenge for deep learning models, which tend to be biased toward majority classes. While recent methods for long-tailed recognition have mitigated this issue, they are largely restricted to single-modal inputs and cannot fully exploit complementary information from diverse data sources. In this work, we introduce a new framework for long-tailed recognition that explicitly handles multi-modal inputs. Our approach extends multi-expert architectures to the multi-modal setting by fusing heterogeneous data into a unified representation while leveraging modality-specific networks to estimate the informativeness of each modality. These confidence-guided weights dynamically modulate the fusion process, ensuring that more informative modalities contribute more strongly to the final decision. To further enhance performance, we design specialized training and test procedures that accommodate diverse modality combinations, including images and tabular data. Extensive experiments on benchmark and real-world datasets demonstrate that the proposed approach not only effectively integrates multi-modal information but also outperforms existing methods in handling long-tailed, class-imbalanced scenarios, highlighting its robustness and generalization capability.

Probing heads map the representations learned from audio by a machine learning model to downstream task labels and are a key component in evaluating representation learning. Most bioacoustic benchmarks use a fixed, low-capacity probe, such as a linear layer on the final encoder layer. While this standardization enables model comparisons, it may bias results by overlooking the interaction between encoder features and probe design. In this work, we systematically study different probing strategies across two bioacoustic benchmarks, BEANs and BirdSet. We evaluate last- and multi-layer probing, across linear and attention probes. We show that larger probe heads that leverage time information have superior performance. Our results suggest that current benchmarks may misrepresent encoder quality when relying on a last-layer probing setup. Multi-layer probing improves downstream task performance across all tested models, while attention probing has superior performance to linear probing for transformer models.

Agent-compiled knowledge bases provide persistent external knowledge for large language model (LLM) agents in open-ended, knowledge-intensive downstream tasks. Yet their quality is systematically limited by \emph{incompleteness}, \emph{incorrectness}, and \emph{redundancy}, manifested as missing evidence or cross-document links, low-confidence or imprecise claims, and ambiguous or coreference resolution issues. Such defects compound under iterative use, degrading retrieval fidelity and downstream task performance. We present \textbf{DeepRefine}, a general LLM-based reasoning model for \emph{agent-compiled knowledge refinement} that improves the quality of any pre-constructed knowledge bases with user queries to make it more suitable for the downstream tasks. DeepRefine performs multi-turn interactions with the knowledge base and conducts abductive diagnosis over interaction history, localizes likely defects, and executes targeted refinement actions for incremental knowledge base updates. To optimize refinement policies of DeepRefine without gold references, we introduce a Gain-Beyond-Draft (GBD) reward and train the reasoning process end-to-end via reinforcement learning. Extensive experiments demonstrate consistent downstream gains over strong baselines.

Precise spatial reasoning is fundamental to robotic manipulation, yet the visual backbones of current vision-language-action (VLA) models are predominantly pretrained on 2D image data without explicit 3D geometric supervision, resulting in representations that lack accurate spatial awareness. Existing implicit spatial grounding methods partially address this by aligning VLA features with those of 3D-aware foundation models, but they rely on empirical layer search and perform alignment on LLM-level visual tokens where spatial structure has already been entangled with linguistic semantics, limiting both generalizability and geometric interpretability. We propose VEGA (Visual Encoder Grounding Alignment), a simple yet effective framework that directly aligns the output of the VLA's visual encoder with spatially-aware features from DINOv2-FiT3D, a DINOv2 model fine-tuned with multi-view consistent 3D Gaussian Splatting supervision. By performing alignment at the visual encoder output level, VEGA grounds spatial awareness before any linguistic entanglement occurs, offering a more interpretable and principled alignment target. The alignment is implemented via a lightweight projector trained with a cosine similarity loss alongside the standard action prediction objective, and is discarded at inference time, introducing no additional computational overhead. Extensive experiments on simulation benchmark and real-world manipulation tasks demonstrate that VEGA consistently outperforms existing implicit spatial grounding baselines, establishing a new state-of-the-art among implicit spatial grounding methods for VLA models.

Scene graph alignment establishes object correspondences between two 3D scene graphs constructed from partially overlapping observations. This enables efficient scene understanding and object-level relocalization when a robot revisits a place, as well as global map fusion across multiple agents. Such capabilities are essential for robots that require long-term memory for long-horizon tasks involving interactions with the environment. Existing approaches mainly focus on subscan-to-subscan (S2S) alignment and depend heavily on geometric point-cloud features, leaving frame-to-scan (F2S) alignment and open-set vision-language features underexplored. In addition, existing datasets for scene graph alignment remain small-scale with limited object diversity, constraining systematic training and evaluation. We present a unified and efficient scene graph alignment framework that predicts object correspondences by fusing vision-language, textual, and geometric features with spatial context. The framework comprises modules such as a distance-gated spatial attention encoder, a minimum-cost-flow-based allocator, and a global scene embedding generator to achieve accurate alignment even under large coordinate discrepancies. We further introduce ScanNet-SG, a large-scale dataset generated via an automated annotation pipeline with over 700k samples, covering 509 object categories from ScanNet labels and over 3k categories from GPT-4o-based tagging. Experiments show that our method achieves the best overall performance on both F2S and S2S tasks, substantially outperforming existing scene graph alignment methods. Our code and dataset are released at: https://autonomousrobots.nl/paper_websites/opensga.

Over the years, research in system identification has provided a rich set of methods for learning dynamical models, together with well-established theoretical guarantees. In practice, however, the choice of model class, training algorithm, and hyperparameter tuning is still largely left to empirical trial-and-error, requiring substantial expert time and domain experience. Motivated by recent advances in agentic artificial intelligence, we present ASIA, a framework that delegates this iterative search to a large language model acting as an autonomous coding agent. Building on existing agentic platforms, ASIA closes the loop between hypothesis, implementation, and evaluation without human intervention, requiring only a plain-English description of the identification problem. We conduct an empirical study of ASIA on two system identification benchmarks and analyse the agent's search behaviour, the architectures and training strategies it discovers, and the quality of the resulting models. We also discuss the potential of the approach and its current limitations, including implicit test leakage, reduced methodological transparency, and reproducibility concerns.

Analog neural networks are gaining attention due to their efficiency in terms of power consumption and processing speed. However, since analog neural networks are implemented as physical circuits, they are highly sensitive to manufacturing process variations, which can cause large deviations from the nominal model. We present a polynomial-based model that resembles the performance of the neuron circuit under process variations. Then, we formally verify the behavior of the circuit-level model using reachability analysis with polynomial zonotopes, thus, avoiding conventional, time-consuming Monte Carlo simulations. We evaluate our proposed verification approach on three different datasets, verifying both fully-connected and convolutional analog neural networks. Our experimental results confirm the effectiveness of our verification approach by reducing the verification time from days to seconds while enclosing 99% of the variation samples.

Super-resolution (SR) is a severely ill-posed problem with inherent ambiguity, as widely recognized in both empirical and theoretical studies. Although recent semantic-guided and multi-modal SR methods exploit large models or external priors to enhance semantic alignment, the fusion of heterogeneous modalities remains insufficiently understood in practice and theory. In this work, we provide the first theoretical modeling of multi-modal SR, revealing that prior methods are bottlenecked by sub-optimal modality utilization. Our analysis shows that the generalization risk bound can be improved by strengthening the alignment between modality weights and their effective contributions, while reducing representation complexity. This theoretical insight inspires us to propose the novel Multi-Modal Mixture-of-Experts Super-Resolution framework (M$^3$ESR) that employs generalization-oriented dynamic modality fusion for accurate risk control and modality contribution optimization. In detail, we propose a novel spatially dynamic modality weighting module and a temporally adaptive modality temperature scheduling mechanism, enabling flexible and adaptive spatial-temporal modality weighting for effective risk control. Extensive experiments demonstrate that our M$^3$ESR significantly boosts generalization and semantic consistency performances, which confirms our superiority.

Muon has emerged as an efficient alternative to Adam for pretraining, yet remains underused for fine-tuning. A key obstacle is that most open models are pretrained with Adam, and naively switching to Muon for fine-tuning leads to degraded performance due to an optimizer mismatch. We investigate this mismatch through controlled experiments and relate it to the distinct implicit biases of Adam and Muon. We provide evidence that the mismatch disrupts pretrained knowledge, and that this disruption scales with update strength. This leads us to hypothesize that constraining updates should mitigate the mismatch. We validate this with LoRA: across language and vision tasks, LoRA reduces the performance gap between Adam and Muon observed under full fine-tuning. Studies on LoRA rank, catastrophic forgetting, and LoRA variants further confirm that mismatch severity correlates with update strength. These results shed light on how optimizer mismatch affects fine-tuning and how it can be mitigated. Our code is available at https://github.com/XingyuQu/muon-finetune.

Large language models (LLMs) exhibit two striking and ostensibly unrelated behaviours: in-context learning (ICL) and repetitive generation. In both, the model behaves as though it had summarised the context into a population-level statistic and discarded token-level detail. We ask whether this ``summarisation and forgetting'' can be derived from the attention mechanism itself, and answer in the affirmative. Under stationary, ergodic and elliptical inputs, the softmax attention output converges almost surely to $Θ_VΣΘ_K^{\top}Θ_Q x_t$, where $Σ$ is the input covariance; the long-context limit is therefore a linear readout of the input's second-order statistics. Two consequences follow. (i) For in-context linear regression, a single softmax head can implement one step of population gradient descent. Stacking such heads with residual connections iterates this update and implements multiple gradient descent steps. (ii) Propagated across an $L$-layer transformer, this readout drives the terminal hidden state at the parametric $1/t$ rate to a deterministic function of the current token alone, so that autoregressive generation collapses asymptotically to a first-order Markov chain whose attracting orbits furnish a structural account of repetition and mode collapse. The two phenomena thus emerge as facets of a single covariance-readout principle.

Zebrafish embryos are a valuable model for drug discovery due to their optical transparency and genetic similarity to humans. However, current evaluations rely on manual inspection, which is costly and labor-intensive. While machine learning offers automation potential, progress is limited by the lack of comprehensive datasets. To address this, we introduce a large-scale dataset of high-resolution microscopic image sequences capturing zebrafish embryonic development under both control conditions and exposure to compounds (3,4-dichloroaniline). This dataset, with expert annotations at fine-grained temporal levels, supports two benchmarking tasks: (1) fertility classification, assessing zebrafish egg viability (130,368 images), and (2) toxicity assessment, detecting malformations induced by toxic exposure over time (55,296 images). Alongside the dataset, we present the first transformer-based baseline model that integrates spatiotemporal features to predict developmental abnormalities at early stages. Experimental results present the model's effectiveness, achieving 98% accuracy in fertility classification and 92% in toxicity assessment. These findings underscore the potential of automated approaches to enhance zebrafish-based toxicity analysis.

Formalisation is the process of writing system requirements in a formal language. These requirements mostly originate in Natural Language. In the field of Formal Methods, formalisation is often identified as one of the most delicate and complicated steps in the verification process. Not seldomly, formalisation tools and environments choose various levels of requirement descriptions: Natural Language, Technical Language, Diagram Representations and Formal Language, to mention a few. In the literature, there are various maxims and principles of good practice to guide the process of requirement formalisation. In this paper we propose a new guideline: Coherency through Formalisations. The guideline states that the different levels of formalisation mentioned above should roughly follow the same logical structure. The principle seems particularly relevant in the setting where LLMs are prompted to perform reasoning tasks that can be checked by formal tools using Structured Natural Language to act as an intermediate layer bridging both paradigms. In the light of coherency, we analyze NASA's Formal Requirement Elicitation Tool FRET and propose an alternative automated translation of the Controlled Natural Language FRETish to the formal language of MTL. We compare our translation to the original translation and prove equivalence using model checking. Some statistics are performed which seem to favor the new translation. As expected, the translation process yielded interesting reflections and revealed inconsistencies which we present and discuss.

Atomic properties such as partial charges or multipoles encode chemically meaningful information that can inform downstream molecular property prediction, but their evaluation as machine learning targets has been complicated by the absence of a principled out-of-distribution evaluation protocol at the atomic level. In this work, we propose a held-out evaluation protocol that clusters atomic environments by SOAP descriptors and computes metrics accounting only for cluster labels unseen during training. Following this procedure, we use 5$\times$5 cross-validation and Tukey's HSD to run a statistically rigorous comparison of E(3)-equivariant against non-equivariant, rotationally augmented models for predicting electron populations and multipoles of H, C, N, and O atoms. Building on our results, we introduce the Quantum Topological Neural Network (QT-Net), a rotationally augmented, non-equivariant graph neural network. We show that QT-Net can be used to infer properties of atoms in molecules from QM9 outside our training set, and that these inferred properties can yield improvement when used as input features for downstream molecular property prediction. To further validate the framework, molecular dipole moments computed from QT-Net's per-atom outputs recover the ground-truth values reported in QM9. We release all code and data, including a JAX implementation of QT-Net, to support the broader use of learned QTA properties as inductive biases for atomic-scale molecular machine learning.

Point clouds are a fundamental representation for robotic perception tasks such as localization, mapping, and object pose estimation. However, LiDAR-acquired point clouds are inherently sparse and non-uniform, providing incomplete observations of the underlying scene geometry. This makes reliable geometric reasoning challenging and degrades downstream perception performance. Existing approaches attempt to compensate for these limitations by estimating local geometry, but often rely on hand-crafted statistics or end-to-end supervised learning, which can suffer from limited scalability or require large amounts of accurately labeled data. To address these challenges, we explicitly model point cloud geometry under a principled mathematical formulation. We represent local geometry as a statistical manifold induced by a family of Gaussian distributions, where each point is associated with a Gaussian capturing its local geometric structure. Based on this formulation, we introduce Point-to-Ellipsoid (POLI), a deep neural estimator that predicts per-point Gaussian geometry. POLI learns a mapping from point cloud observations to their underlying geometry in a self-supervised manner, removing the need for labeled data while preserving strong geometric inductive biases. The resulting representation integrates seamlessly into existing robotic perception pipelines without architectural modifications. Extensive experiments show that POLI enables accurate and robust geometry estimation and consistently improves performance across diverse robotic perception tasks.

Short-term ocean forecast skill depends strongly on the three-dimensional ocean structure of the upper ocean, which governs stratification, subsurface heat storage, and the response of the ocean to atmospheric forcing. However, AI ocean forecasting models often fail to preserve this vertical structure, resulting in over-smoothed subsurface features and weak physical consistency under strong forcing. Here, we present AxiomOcean, a global AI ocean forecasting model that explicitly represents vertical hierarchy and cross-layer dependence within the water column. By combining a fully three-dimensional encoder-backbone-decoder architecture with surface atmospheric forcing, AxiomOcean jointly predicts upper-ocean temperature, salinity, and three-dimensional currents at global 1/12° resolution down to 643 m depth. In 10-day forecasts, AxiomOcean outperforms an advanced AI comparison model across variables and lead times, reducing day-1 RMSE by approximately 20 to 35% while maintaining higher anomaly correlation. The gain is not achieved through excessive smoothing: AxiomOcean better preserves eddy kinetic energy, temperature and salinity variance. Its advantage also extends through the water column and remains evident across the equatorial Pacific, Kuroshio Extension, and Southern Ocean, yielding a more realistic reconstruction of upper-ocean heat content. These results show that explicitly preserving upper-ocean three-dimensional structure can improve both forecast accuracy and physical fidelity in AI ocean prediction.

Speculative decoding speeds up autoregressive generation in Large Language Models (LLMs) through a two-step procedure, where a lightweight draft model proposes tokens which the target model then verifies in a single forward pass. Although the drafter network is small in modern architectures, its LM-head still performs projection to a large vocabulary, becoming one of the major computational bottlenecks. In prior work this issue has been predominantly addressed via static or dynamic vocabulary truncation. Yet mitigating the bottleneck, these methods bring in extra complexity, such as special vocabulary curation, sophisticated inference-time logic or modifications of the training setup. In this paper, we propose SlimSpec, a low-rank parameterization of the drafter's LM-head that compresses the inner representation rather than the output, preserving full vocabulary support. We evaluate our method with EAGLE-3 drafter across three target models and diverse benchmarks in both latency- and throughput-bound inference regimes. SlimSpec achieves $4\text{-}5\times$ acceleration over the standard LM-head architecture while maintaining a competitive acceptance length, surpassing existing methods by up to $8\text{-}9\%$ of the end-to-end speedup. Our method requires minimal adjustments of training and inference pipelines. Combined with the aforementioned speedup improvements, it makes SlimSpec a strong alternative across wide variety of draft LM-head architectures.

Spectral neural operators achieve strong performance for PDE learning, but rely on fixed global bases that limit their ability to represent spatially heterogeneous and multiscale dynamics. We propose Adaptive Basis Learning (ABLE), a framework that learns data-dependent spectral representations instead of relying on predefined bases. ABLE constructs a spatially adaptive Parseval frame via a learned ancillary density, enabling the operator to act in a lifted spectral space while preserving invertibility and maintaining $O(N\log N)$ complexity through FFT-based implementation. This shifts the source of expressivity from spectral coefficients to the representation itself, allowing the model to capture localized structures and non-translation-invariant interactions more efficiently. ABLE integrates seamlessly into existing neural operator architectures as a drop-in replacement for spectral layers. Across a range of benchmarks ABLE improves accuracy over strong baselines, with the largest gains in regimes characterized by sharp gradients and multiscale behavior. Moreover, augmenting existing models (e.g., U-FNO, HPM) with ABLE further enhances their performance, demonstrating its role as a general and complementary spectral refinement. Our results highlight that the data-driven choice of representation, rather than operator complexity alone, is a key bottleneck in neural operator design. By learning the basis itself, ABLE provides a principled and efficient framework for improving spectral methods in PDE learning.

Quantifying fish community structure is essential for understanding biodiversity and ecosystem responses in a changing environment, yet existing survey methods provide limited high-frequency, quantitative observations. Conventional approaches, including catch-based methods, underwater visual censuses, and environmental DNA metabarcoding, either require intensive labor or lack reliable estimates of abundance and biomass. Here, we develop an automated framework for quantifying fish communities from underwater video using computer vision. Using videos acquired with a custom-made stereo camera system, the framework integrates deep learning-based fish identification, multi-object tracking, and 3D reconstruction to estimate species-level abundance and biomass. We applied the approach to a reef fish community over a 20-day period with hourly daytime observations, revealing dynamic fluctuations in species richness, abundance, and biomass associated with changes in species composition. By comparing fish communities estimated from visual census and environmental DNA surveys, we demonstrate that our method provides complementary strengths for continuous, non-invasive, and quantitative monitoring of consistently observed species. This approach provides a scalable foundation for long-term monitoring and advances the capacity to resolve fine-scale temporal dynamics in fish communities.

Interactive agent benchmarks map an agent run to a binary outcome through outcome checks. When these checks rely on surface level signals or fail to capture the agent's actual action path, they cannot reliably determine whether the run succeeded. For example, a benchmark task may ask whether Alice's shipping address was changed, while the outcome check only verifies that the agent clicked "Save." This does not guarantee that the intended state change occurred, since the agent may have modified the wrong record. Treating such a run as successful therefore makes the reported score misleading. Benchmark quality thus depends not only on task design, but also on the reliability of outcome detection. We address this problem by introducing an outcome evidence reporting layer for existing benchmarks, without modifying their tasks, agents, or evaluators. The layer performs three functions. First, before scoring, it specifies which stored artifacts are required to verify the claimed outcome for each case. Second, it applies a locked checklist to each completed run and assigns one of three evidence labels: Evidence Pass, Evidence Fail, or Unknown. Third, it reports evidence supported score bounds that quantify uncertainty arising from Unknown cases. Rather than silently counting, discarding, or hiding uncertain cases inside a single aggregate success rate, the framework keeps them explicitly visible. We evaluate the outcome evidence layer on five public benchmarks: ANDROIDWORLD, AGENTDOJO, APPWORLD, tau3 bench retail, and MINIWOB. The resulting reports separate several empirically distinct failure modes.

Unified Multimodal Models (UMMs) excel in general tasks but struggle to bridge the gap between personalized understanding and generation. Prior works largely rely on implicit token-level alignment via supervised fine-tuning, which fails to fully capture the potential synergy between comprehension and creation. In this work, we propose Sync-R1, an end-to-end reinforcement learning framework that jointly optimizes personalized understanding and generation within a single, explicit reasoning loop. Through this unified feedback process, Sync-R1 enables personalized comprehension to guide content creation, while the resulting generation quality reciprocally refines understanding within an integrated reward landscape. To efficiently orchestrate this dual-task synergy, we introduce Sync-GRPO, a reinforcement learning method utilizing an ensemble reward system. Furthermore, we propose Dynamic Group Scaling (DGS), which adaptively filters low-potential trajectories to reduce gradient variance and accelerate convergence. To better reflect real-world complexity, we introduce UnifyBench++, featuring denser textual descriptions and richer user contexts. Experimental results demonstrate that Sync-R1 achieves state-of-the-art performance, showcasing superior cross-task reasoning and robust personalization without requiring complex cold-start procedures. The code and the UnifyBench++ dataset will be released at: https://github.com/arctanxarc/UniCTokens.

Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing diffusion models, enabling users to inject new visual concepts or styles through lightweight parameter updates. However, LoRAs can memorize training images, causing generated outputs to reproduce copyrighted or sensitive content. This risk is particularly concerning in LoRA-sharing ecosystems, where users distribute trained LoRAs without releasing the underlying training data. Existing approaches for mitigating memorization rely on access to the training pipeline, training data, or control over the inference process, making them difficult to apply when only the released LoRA weights are available. We propose \textbf{Base-Anchored Filtering (BAF)}, a training-free and data-free framework for post-hoc memorization mitigation in diffusion LoRAs. BAF decomposes LoRA updates into spectral channels and measures their alignment with the principal subspace of the pretrained backbone. Channels strongly aligned with this subspace are retained as generalizable adaptations, while weakly aligned channels are suppressed as potential carriers of memorized content. Experiments on multiple datasets and diffusion backbones demonstrate that BAF consistently reduces memorization while preserving or even improving generation quality. Our code is available in the supplementary material.

Current 3D tokenizers largely treat representation as spatial compression: compact codes reconstruct surface geometry, but leave component ownership and attachment validity implicit. In open-world assets with intersecting components, noisy topology, and weak canonical structure, this creates a representation mismatch: local shape, component identity, and assembly relations become entangled in a latent stream and are not natively addressable during decoding. We formulate an alternative view, interface-centric generative states, in which tokenization constructs an operational state rather than a passive compressed code. The state exposes local geometry, component ownership, and attachment validity as variables that can be queried, constrained, and repaired during decoding. We instantiate this formulation with Component-Conditioned Canonical Local Tokens (C2LT-3D), factorizing representation into canonical local geometry, partition-conditioned context, and relational seam variables. Each factor targets a distinct failure mode of compression-centric tokens: pose leakage, cross-component interference, or invalid local attachment. This exposed state supports attachment validation, latent structural repair, targeted intervention, and constrained serialization without a separate post-hoc structure recovery module. Trained on single-object CAD models and evaluated zero-shot on open-world multi-component assets, C2LT-3D improves structural robustness and shows that its latent variables remain actionable under adversarial attachment settings. These results suggest that open-world 3D generative representations should be evaluated not only by reconstruction fidelity, but by whether their discrete states remain operational for assembly-level structural reasoning.

Commercial video generation systems such as Seedance2.0 and Veo3.1 have rapidly improved, strengthening the view that video generators may be evolving into "world simulators." Yet the community still lacks a benchmark that directly tests whether a model can reason about how an observed world should evolve over time. We introduce WorldReasonBench, which reframes video generation evaluation as world-state prediction: given an initial state and an action, can a model generate a future video whose state evolution remains physically, socially, logically, and informationally consistent? WorldReasonBench contains 436 curated test cases with structured ground-truth QA annotations spanning four reasoning dimensions and 22 subcategories. We evaluate generated videos with a human-aligned two-part methodology: Process-aware Reasoning Verification uses structured QA and reasoning-phase diagnostics to detect temporal and causal failures, while Multi-dimensional Quality Assessment scores reasoning quality, temporal consistency, and visual aesthetics for ranking and reward modeling. We further introduce WorldRewardBench, a preference benchmark with approximately 6K expert-annotated pairs over 1.4K videos, supporting pair-wise and point-wise reward-model evaluation. Across modern video generators, our results expose a persistent gap between visual plausibility and world reasoning: videos can look convincing while failing dynamics, causality, or information preservation. We will release our benchmarks and evaluation toolkit to support community research on genuinely world-aware video generation at https://github.com/UniX-AI-Lab/WorldReasonBench/.

This paper investigates the effectiveness of large language models (LLMs) in answering questions over datasets. We examine their performance in two scenarios: (a) directly answering questions given a dataset file as input, and (b) generating SQL queries to answer questions given the schema of a relational database. We also evaluate the impact of different prompting strategies on model performance. The study includes both state-of-the-art LLMs and smaller language models that require fewer resources and operate at lower computational and financial cost. Experiments are conducted on two datasets containing questions of varying difficulty. The results demonstrate the strong performance of large LLMs, while highlighting the limitations of smaller, more cost-efficient models. These findings contribute to a better understanding of how LLMs can be utilized in data analytics tasks and their associated limitations.

Rotary Positional Encoding (RoPE) is widely used in modern large language models. However, when sequences are extended beyond the range seen during training, rotary phases can enter out-of-distribution regimes, leading to spurious long-range alignments, diffuse attention, and degraded retrieval. Existing remedies only partially address these failures, as they often trade local positional resolution for long-context stability. We propose GAPE (Gated Adaptive Positional Encoding), a drop-in augmentation for positional encodings that introduces a content-aware bias directly into the attention logits while preserving the rotary geometry. GAPE decouples distance-based suppression from token importance through a query-dependent gate that contracts irrelevant context and a key-dependent gate that preserves salient distant tokens. We prove that protected tokens remain accessible, while the attention mass assigned to unprotected distant tokens decays as a function of the query gate. We further show that GAPE can be implemented within standard scaled dot-product attention. We validate these properties empirically, finding that GAPE consistently yields sharper attention and improved long-context robustness over rotary baselines across both synthetic retrieval and long-context benchmarks.

Large language models can score well on named game-theory benchmarks while failing on the same strategic computation once semantic cues are removed. We show this gap with procedurally generated zero-sum matrix games: a model that recognizes familiar games drops to 34%, 18%, and 2% success on anonymous $2{\times}2$, $3{\times}3$, and $5{\times}5$ payoff matrices. The benchmark separates semantic recall, learned approximate Nash computation, and an output-interface bottleneck that limits scale. Training only on $2{\times}2$ and $3{\times}3$ games, supervised fine-tuning raises unseen $5{\times}5$--$7{\times}7$ success from 2% to 61%, while exploitability-reward training averages 37% with high seed variance. We prove that the exploitability residual is $2$-Lipschitz in payoff perturbations, unlike discontinuous vertex-returning LP equilibrium selectors, explaining why residual training can transfer under payoff shifts even when formatting instability limits mean performance. A dominated-action padding experiment provides causal evidence: trained models solve $3{\times}3$ games embedded in much larger matrices, while random-padded controls fail and dense $12{\times}12$ games remain near failure. Procedural evaluation is therefore necessary for measuring strategic reasoning, and residual rewards expose a real but format-limited route to approximate equilibrium computation.

Existing image simplification techniques often rely on Non-Photorealistic Rendering (NPR), transforming photographs into stylized sketches, cartoons, or paintings. While effective at reducing visual complexity, such approaches typically sacrifice photographic realism. In this work, we explore a complementary direction: simplifying images while preserving their photorealistic appearance. We introduce progressive semantic image simplification, a framework that iteratively reduces scene complexity by removing and inpainting elements in a controlled manner. At each step, the resulting image remains a plausible natural photograph. Our method combines semantic understanding with generative editing, leveraging Vision-Language Models (VLMs) to identify and prioritize elements for removal, and a learned verifier to ensure photorealism and coherence throughout the process. This is implemented via an iterative Select-Remove-Verify pipeline that produces high-quality simplification trajectories. To improve efficiency, we further distill this process into an image-to-video generation model that directly predicts coherent simplification sequences from a single input image. Beyond generating cleaner and more focused compositions, our approach enables applications such as content-aware decluttering, semantic layer decomposition, and interactive editing. More broadly, our work suggests that simplification through structured content removal can serve as a practical mechanism for guiding visual interpretation within the photorealistic domain, complementing traditional abstraction methods.

Modern LLM APIs often reveal only top-$K$ logit scores and censor the remaining vocabulary. We study the per-position distribution-recovery limits of this access model. For censoring threshold $τ$, the compatible teacher distributions form an identified set whose total-variation diameter is exactly $U_K=(V-K)\exp(τ)/(Z_A+(V-K)\exp(τ))$, where $Z_A$ is the observed partition function. For KL recovery, we give a computable binary-endpoint lower bound and an asymptotically matching small-ambiguity upper bound, with an extension to reference-aware attackers. Experiments on a Qwen3 math-reasoning teacher reveal a layered extraction hierarchy: on-task top-$K$ distillation recovers 12% of private capability, full-logit distillation recovers 56% despite 99% KL closure, and generation-based extraction recovers 96%. Top-$K$ censoring therefore limits per-position distribution recovery but does not by itself prevent capability extraction, separating fidelity from transfer in prompt-only logit distillation.

Selecting the best large language model (LLM) for a fixed benchmark is often expensive, since exhaustive evaluation requires running every model on every example. Multi-armed bandit (MAB) algorithms can reduce the number of LLM calls by sequentially selecting the next model-example pair to evaluate, thereby avoiding wasted evaluations on clearly underperforming models. Further savings can be achieved by predicting model scores from the partially observed model-example score matrix using low-rank factorization. However, such predictions are not ground truth: they can be biased and may therefore lead to incorrect identification of the best model. In this work, we propose a principled framework that combines MAB with cheap predicted scores without compromising statistical validity. Specifically, we derive doubly robust estimators of each model's performance that use the low-rank predictions to reduce variance. This enables the construction of valid finite-sample confidence intervals in our setting, where models are selected adaptively and examples are sampled without replacement. Empirical results on real-world benchmarks show that our approach reduces the number of required evaluations, yielding meaningful savings in compute and cost while accurately identifying the best-performing model.