C-Gap

Data agents aim to fulfill natural-language instructions over heterogeneous data, including tables, files, and databases. However, data agents face a challenging agent-data gap: heterogeneous data resides outside the agent, while the agent can access it (e.g., column names and file paths) only through generic tools. Existing approaches either let agents directly explore raw data sources or inject manually constructed semantic layers into prompts. However, neither scales well to large heterogeneous data sources nor adapts to different agent behaviors. In this paper, we introduce EvoOntology, a self-evolving ontology layer for data agents. EvoOntology encapsulates the ontology as an MCP server comprising a schema layer, a content layer, and a tool layer, enabling agents to actively query and interact with the ontology at runtime. To this end, we introduce a builder agent for autonomous ontology construction and a self-evolution loop that continuously refines the ontology through attribution-guided typed edits that are accepted only after a backbone-conditional paired evaluation. Experiments on three well-adopted data-agent benchmarks with four LLM backbones demonstrate that EvoOntology consistently outperforms strong baselines and existing semantic-layer approaches, effectively bridging the agent-data gap and enabling more effective interaction with heterogeneous data. Code: https://github.com/ruc-datalab/EvoOntology

Autoregressive generation of interleaved text and acoustic tokens is a common approach to spoken-response generation in speech large language models. Although this design enables streaming generation with explicit textual guidance, generated acoustic tokens become part of the context for subsequent text predictions. Given identical speech inputs, we observe markedly lower answer accuracy for the internal text generated in speech-to-text-and-speech (S2TS) mode than for speech-to-text (S2T) responses. We term this discrepancy the \emph{output-mode gap} (OMG). To reduce OMG, we propose \emph{Joint-Output On-Policy Distillation} (JO-OPD), which distills the model's stronger S2T policy into joint generation using student-generated S2TS trajectories. At each text position, the S2T teacher provides soft targets from a text-only projection of the student's preceding outputs, while the student predicts from the corresponding full interleaved history. A preservation objective further regularizes native non-text predictions. Experiments on Step-Audio-2-mini and Baichuan-Audio-Instruct reveal OMG across two interleaved generation architectures. On Step-Audio-2-mini, JO-OPD reduces OMG from 42.87 to 16.26 percentage points on Spoken-MQA and from 29.72 to 13.04 points on speech-rendered GSM8K, with little change in S2T accuracy and substantially larger reductions than matched SFT baselines. ASR-based evaluation further shows a 7.49-point improvement in spoken-answer accuracy on Spoken-MQA.

Vision encoders for Cell Painting are typically ranked by a single evaluation, commonly replicate mean average precision (mAP). We introduce CP-BG-Bench, a paired-view evaluation framework that holds the central cell fixed across four matched views (raw crop C, segmented S, and density-augmented variants CD and SD), ablating or augmenting surrounding pixels as a controlled intervention. Instantiating the framework on three datasets (JUMP-CP, RxRx1, RxRx3-core) and three encoders (DINOv3 ViT-B/16, OpenPhenom, SubCell) under four community-standard protocols (replicate mAP, scIB batch integration, CellProfiler feature prediction, cross-batch perturbation recall), we find that the four protocols rank the same encoders systematically differently, with disagreements decomposing along three axes: cell versus background, morphology versus context, and within-study versus across-batch. The largest effect: on RxRx3-core, SubCell with segmented inputs retains 94% of crop replicate mAP but only 32% of crop R@10, so the within-study signal preserved under segmentation is largely non-transferable; density augmentation recovers 84% of the within-study C-to-S gap but only 8% of the cross-batch gap. Segmented views predict CellProfiler features as well as or better than crops on two of three datasets, inverting the replicate-mAP ranking, and the C-to-S gap varies by an order of magnitude across datasets while remaining similar across encoders, indicating that background-driven gain is set by experimental design rather than by the encoder. Single-metric ranking of Cell Painting encoders is therefore sensitive to the protocol used, and protocol disagreements are interpretable as projections onto the three axes the paired-view design exposes. We will release the paired-view datasets, reconstruction pipelines, 36 trained checkpoints, aggregated embeddings, and the full evaluation suite.

Software development follows an implementation-verification loop in which developers or agents iteratively revise an implementation until an evaluator, such as a test suite, accepts it. The evaluator checks the implementation against a set of requirements under a model of the deployment environment. Yet even a formal proof that the implementation satisfies the requirements under the model cannot guarantee acceptable behavior after deployment. Requirements only approximate stakeholder intent, and the model only approximates the real deployment environment. We call these together - requirement gap and model gap - the two-gap framework, which unifies the main failure modes of agentic software engineer-ing: reward hacking exploits omissions in the requirements or model, while hallucination widens the gaps by fabricating requirements or environment assumptions. Because neither gap can generally be certified closed in an open, changing world, the goal shifts from closing them to continuously narrowing them. We therefore propose an assurance-revision loop that uses deployment evidence to revise the requirements, model, or evaluator when stakeholders reject the resulting behavior. We then cast assured agentic development as a resource-allocation problem over human judgment, agent capability, and compute. The two principal bottlenecks mirror the two gaps: human judgment for the requirement gap and faithful, costly evaluation for the model gap. Reality remains the final verifier: acceptable behavior under actual deployment conditions is the ultimate test, while predeployment evaluations remain proxies for it.

Block drafters propose several tokens in one forward pass, before earlier target tokens are realised. Their rejection mixes two losses: missing within-block path information and imperfect modelling of observable information. Accepted length cannot distinguish them. We separate the two with an information floor, the minimum expected rejection at a specified conditioning order; rejection above this floor is the model gap. Estimating both from target rollouts across four domains, four open-weight targets, and a frontier API target yields three findings. First, the all-parallel floor reaches $0.286$ at the final slot on Qwen3-4B, limiting even the best proposal to $71\%$ per-slot acceptance. Second, one realised token removes $86$--$100\%$ of this floor, a locality also recovered by an independent mutual-information analysis. Third, current drafters remain far above their floors: the final-slot model gap accounts for $43$--$64\%$ of DFlash rejection and $85$--$92\%$ of DSpark's oracle-conditioned rejection. These findings separate the value of short-range conditioning from proposal quality.

Existing guidance identifies excessive agency, excessive permission, weak task-bound authorization, and inadequate agent controls as important risks. Control frameworks also describe capabilities for constraining, authorizing, observing, validating, and responding to agent activity. Yet security programs still need a way to manage persistent deployed instances that span components and outlive any one event. We propose the agentic posture vulnerability (APV) as a task-conditioned vulnerability-management abstraction: a durable record for a composed agent-control exposure. One posture may produce different runtime manifestations across tasks; APV links those manifestations to the invariant posture and remains open until authority is narrowed, a missing control is added, risk is accepted, or closure is verified. APV is not proposed as a new root-cause class of risk; it operationalizes existing excessive-agency, authorization, and control-composition weaknesses. We distinguish APVs from CVE-addressable product defects, OWASP Excessive Agency, Agent Baseline control outcomes, and the runtime authorization-execution gap. We then provide a field vignette, a thresholded definition, six recurring APV patterns, a vulnerability lifecycle, a minimum record, a control-and-closure matrix, tooling implications, and a testable research agenda.

Long-horizon reasoning in recent LLMs demands that the model switch between distinct skills inside a reasoning chain, such as first doing a math derivation, then using the result to plan a schedule. We call such problems cross-skill long-horizon tasks: multi-step tasks whose steps require different reasoning skills and depend on earlier outputs. Existing benchmarks often evaluate individual skills, lacking a principled way to measure how well a model switches between skills. We address this gap from both the evaluation and training sides. We introduce Skill Entropy, a measure of the difficulty of switching from one skill to another. We then propose Skill^2-Bench, a benchmark of cross-skill long-horizon tasks built over 558 skills across 9 verifiable and open-ended domains. Each task is assigned a task-level skill-entropy score and grouped into three difficulty levels. Evaluating 8 frontier and 4 open-source models on Skill^2-Bench reveals a skill-switching gap: accuracy decreases on higher-entropy tasks. We then turn skill entropy from a benchmark scale into a training signal. We propose Skill-Entropy RL, an RL framework where the model predicts not only the answer at each step but also the skill used to produce it. The reward combines step-level correctness with a skill-entropy reward that measures the alignment between the model-predicted skill sequence and the gold skill sequence. On Qwen3-4B-Instruct and Qwen3-1.7B, Skill-Entropy RL improves the Skill^2-Bench score from 34.4% to 68.4% and from 14.6% to 40.1%, respectively, outperforming competitive baselines. The same pipeline can be applied to off-the-shelf training data such as OpenR1-Math, indicating that skill entropy is a reusable training signal. Code available at: https://github.com/Gen-Verse/Skill-Entropy-RL

A benchmark score means nothing without knowing what a trivial method achieves and what the best possible method could achieve. We construct both bounds for a task with a rare kind of ground truth: predicting which sentences a crowd of readers -- highlighting for their own purposes, unpaid, uninstructed, and blind to each other -- marked in 120 web documents. The floor is naive truncation (lead); the ceiling is a split-half oracle: half the crowd predicting the other half. The gap between them is +0.2028 AP [+0.1698, +0.2342, domain-clustered], and three findings structure it. First, the gap is semantic: position and length features recover 5% of it. Second, frontier language models reach 35-53% of it zero-shot -- far above classical baselines, far below the crowd; a state-of-the-art prompt compressor (LLMLingua-2) lands below the floor, indistinguishable from random selection. Third, an unweighted cross-vendor fusion of five frontier rankings plus a position prior reaches 60%, beating the best single model by +0.0159 [+0.0044, +0.0269; Holm p=0.019] -- a gain that survives ablation of its best member, split-half arm selection, prompt paraphrase, and label, gate, and seed perturbations, and was CONFIRMED by a pre-registered replication on 217 independent documents (+0.0179, Holm p=0.042). Finally, the bracket compresses: distilling the fusion into one open-weight 8B student that reads the whole document retains 90% of the fusion's edge and reaches statistical parity with the strongest single frontier model (+0.0070 [-0.0068, +0.0200]), where a local-context student retains only 63% -- the crowd's signal lives in document-level structure, and the cheapest known improvement is to ask several different models and average.

Multimodal fake news detectors often generalize poorly across domains because they learn to trust unreliable evidence: domain-specific shortcuts amplified by imbalanced data and semantically inconsistent text-image pairs that make cross-modal evidence unreliable. We propose Expert-Guided Mutual Distillation (EGMD), which learns what evidence to trust across the prediction pipeline. At the input level, input-level calibration encodes pair-level coherence as a shared gain before fusion. At the representation level, an expert-guided teacher aligns domain statistics and encourages domain-specific patterns to concentrate in specialized experts. At the decision level, prototype-anchored domain-specific students use mutual learning and dual-channel distillation to inherit the teacher's feature geometry and calibrated predictions while discouraging local domain priors. We further construct Weibo_Balanced, a domain-balanced benchmark that isolates the effect of imbalance on generalization. Across four datasets in two languages, EGMD achieves state-of-the-art accuracy while reducing domain bias by up to 57.3%.

Closing the gap between benchmark performance and reliable real-world operation remains a central challenge for Vision-Language-Action (VLA) humanoid robots, which must handle execution errors, distribution shifts, and environmental variability. This paper presents DEED (Data-Efficient Post-Training and Experience-Driven Learning), a systems-level approach evaluated on a supermarket chip-restocking task using a Unitree G1-Edu humanoid robot and the GR00T N1.6 foundation model. DEED comprises three key components: (1) a data-efficient post-training pipeline with control-frequency alignment, data curation, task-relevant visual highlighting, and reduced VLA dependence; (2) a real-world study of experience-driven refinement, adapted from RECAP via a text-based advantage prefix and a vision-language value function; and (3) a latent-space analysis tool for studying in- and out-of-distribution behavior. Our results suggest that bridging the lab-to-store gap is primarily a systems integration challenge rather than an architectural one: careful data design and targeted post-training can transform a policy that fails under naive fine-tuning into a competent real-world system using only a single GPU.

Attention-based KV cache eviction (H2O and its descendants) compresses the memory-constrained state of a long-context model by ranking tokens on accumulated attention mass, treated here as signal energy, and keeping the heaviest. On schema-dense input streams such as nested JSON, this score acts as a non-stationary filter that disproportionately retains noise: a non-content sink role (delimiters or whitespace) carries an order of magnitude more energy than any content role, and structural KEY tokens are over-retained at roughly 1.8x the rate of the answer-carrying VALUE tokens, collapsing exact-match accuracy from 88% to 0% at a 5% budget as the signal-to-noise ratio of the retained state degrades. A counterfactual experiment establishes that suppressing KEY tokens is the best deployable filter. Our retraining-free, role-conditional allocation over SnapKV's windowed score, governed by a single tuned hyperparameter, closes 63-98% of the H2O gap at sub-20% budgets and, at higher budgets, modestly matches or exceeds full-cache accuracy -- a small, seed-sensitive denoising effect (borderline significant at B=0.50; not distinguishable from zero at B=0.30 over four seeds). A 15 MB linear role probe supplies these labels at negligible inference cost, though matching parser-level downstream accuracy remains open.

When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.

Safety-critical perception systems must reliably detect rare object classes within small label spaces, a setting that long-tailed detection methods, designed for hundreds of classes with dense annotation, fundamentally do not address. Open-vocabulary detectors offer a promising alternative, as they use natural language queries at inference time, making prompt quality a first-class lever for detection performance. We exploit this property to address class imbalance: rather than retraining models or collecting additional annotations, we ask whether iteratively refining the language prompts, fed to frozen detectors, can improve minority class detection. We introduce C-GAP Caption-Guided Augmentation and Prompting), a detector-agnostic, annotation-free framework that operates in two phases. First, we establish a composite caption baseline combining per-image scene descriptions with class-quantity context, which we show outperforms scene-description only or class-quantity-only prompts across multiple open-vocabulary architectures and benchmarks. Second, an LLM iteratively refines each image's caption individually, with trials triaged into accept, tentative, or regenerate buckets based on minority-class AP@0.5 against a dynamic threshold derived from the composite baseline. Refinement terminates early once sufficient AP@0.5 gain is achieved. No detector weights are updated at any stage. Our experiments shows that C-GAP improves minority-class average precision up to 53% over the baselines. On COCO, C-GAP improves minority-class AP@0.5 by ~81% relative over the composite baseline (17.69 -> 32.09). Experiments confirm that composite captions provide the critical foundation for effective refinement: using scene-description-only or class-quantity-only prompts as the refinement starting point yields diminishing returns, supporting both stages of C-GAP as necessary contributions.

Close-up rendering, zooming into a scene well beyond any training camera, is important for virtual production and interactive 3D content, yet remains an open challenge. 3D Gaussian splatting (3DGS) enables high-fidelity, real-time novel view synthesis, but its rendering quality degrades at close range. Recent diffusion-based methods that enhance the rendering by conditioning on reference images from the training set produce significant artifacts in this setting. We analyze this failure and identify its root cause: the scale gap between the close-up and reference views. We show that the features in reference-conditioned enhancement models are not scale-invariant, causing cross-view attention to retrieve incorrect correspondences when the same content appears at different scales, and that this mismatch cannot be corrected in latent space because the VAE encoder is not scale-equivariant. Building on this analysis we introduce MACRO, Multi-plane Attention for Closeup Render Optimization, a training-free method for high-quality close-up novel view synthesis from 3DGS. MACRO resolves the scale gap by leveraging the scene's known 3D structure: it decomposes the close-up into depth planes, crops and resizes references in image space to match the scale of each plane before encoding, and applies a depth-aware attention mask so each token attends only to scale-matched references. The method requires no architectural changes or additional training. We further contribute two new close-up novel view synthesis benchmarks, the first standardized evaluation protocol for this setting, and demonstrate state-of-the-art results on both, outperforming existing 3DGS and diffusion-based methods on both reconstruction and perceptual metrics. Project page: https://nitzanhod.github.io/MACRO

We audit fourteen mainstream large language models (LLMs) for hiring discrimination using the paired-resume methodology of Kline, Rose, and Walters (2022). The sole 2023-vintage model reproduces the pro-White callback gap documented in field experiments on labor market discrimination ($+2.12$ pp, significant at the 1% level). Every model released in 2024 or after shows either a null gap or a significant pro-Black reversal (up to $-3.01$ pp). The same pattern holds on the gender axis. Based on 24,024 paired postings per model across 14 models, our results document a reversal in the direction of algorithmic hiring bias across model generations.

Medical Vision-Language Models (Med-VLMs) achieve strong expert-level performance, yet their ability to generate patient-accessible descriptions remains underexplored. With the 21st Century Cures Act now mandating immediate patient access to diagnostic imaging results, evaluating whether Med-VLMs can bridge this Expert-Lay Gap is both urgent and clinically consequential for patient education and shared decision-making. To this end, we introduce MedLayXPlain, the first large-scale multimodal benchmark and evaluation framework for Medical Lay Language Generation (MLLG). MedLayXPlain-122K provides 122,789 region-grounded samples across 8 imaging modalities from 12 publicly available source datasets, each comprising a medical image with paired expert and lay captions anchored in a three-level Unified Medical Language System (UMLS) ontology hierarchy spanning 7 semantic groups, 43 semantic types, and 2,411 medical concepts. Lay captions are constructed via Hierarchical Ontology-Verified Refinement (HOVER), a three-step pipeline combining patient-centric vocabulary mapping, LLM-based constrained rewriting, and cross-model visual verification to enforce semantic equivalence while preventing hallucination. We further introduce MedLayEval, a lightweight 3B evaluator distilled from a 27B verifier that scores expert-lay alignment across five clinically grounded attributes, addressing the poor correlation between standard NLG metrics and clinical judgment. Benchmarking 33 VLMs on MedLayXPlain-122K reveals a systematic Expert-Lay Gap: medical VLMs achieve strong expert captioning but suffer significant lay-register degradation, while general-purpose VLMs produce more accessible language yet lack clinical precision, confirming that neither current paradigm adequately serves patient-facing communication.

Patient contexts span hundreds of heterogeneous documents and thousands of structured data points, yet the document-level metadata that AI systems need for retrieval and triage is absent or incomplete. Standard retrieval-augmented generation fails on this data, mishandling temporal reasoning, cross-document dependencies, and missing metadata. We deploy ACIE (Agentic Clinical Information Extraction) at University Medicine Essen: an on-premise agentic RAG pipeline that reasons over complete patient contexts and grounds every answer in source passages for clinician verification. We quantify the metadata gap, trace the architectural decisions it shaped, and evaluate extraction alongside an independent retrospective lymphoma registry study, in which nuclear-medicine physicians verify every extracted value against its cited sources. Across 7,326 judgments, clinicians accepted 96.5% of extractions, with per-type acceptance ranging from 80% to 99%.

Interactive world models aim to simulate environment dynamics under real-time user actions. However, their action vocabulary is largely confined to navigation: most actions correspond to motion (e.g., walk, turn, look around), while interaction with objects in the scene (e.g., pick up plates, open doors, or trigger physical responses) is either absent, restricted to game domains, or relegated to prompt-to-full-video scenarios. The resulting worlds are visually explorable but not truly actionable. In this work, we present ActWorld, an interactive world model that extends prior navigation-centric generators to support mid-rollout object interaction within a chunk-autoregressive framework. We argue that the navigation-interaction gap stems from two bottlenecks. First, a data bottleneck: the lack of human-object interaction data with accurate, dense labels. Second, a memory bottleneck: recency-biased history compression in existing world models discards the event-transition frames that causally determine subsequent object states, leading to an action-forgetting pathology. On the data side, we construct a 100K interaction video dataset, each annotated with per-chunk captions via chain-of-thought reasoning. On the model side, we introduce a hierarchical action-aware memory design that routes history compression by interaction importance, complemented by a persistent memory bank that maintains event-update and object-identity tokens across long rollouts. Experiments show that ActWorld supports both flexible navigation and rich object interaction within a single model, substantially improving interaction fidelity over navigation-only baselines without sacrificing viewpoint control. Project page is available at https://interactwm.github.io/ActWorld.

KV cache eviction methods typically use a single retention-rule family throughout a model, making eviction-method identity a model-level design choice. Yet Transformer layers differ substantially in their attention behavior, representations, and sensitivity to compression, suggesting that a uniform rule may overlook useful layer-wise structure. This raises a basic question: should eviction methods themselves vary across layers? We investigate this question by composing existing eviction methods across Transformer layers and systematically exploring the resulting layer-wise design space. Using simple offline profiles, we construct fixed routes and study how their quality varies with method placement and cache budget. On LongBench, heterogeneous routing improves performance on a majority of tasks over homogeneous policies at the same cache budget. Even when method counts are held fixed, the profile-guided placement ranks second among 100 evaluated assignments, demonstrating that routing quality depends strongly on where methods are placed. Moreover, the same fixed route outperforms the best of nine standalone baselines across all five tested cache budgets. Together, these results establish layer-wise method composition as an exploitable, placement-sensitive design dimension for KV cache compression.

LLMs are increasingly deployed as agents that interact with external environments and observe feedback such as execution results, error messages, and tool outputs. A well-functioning agent should be able to leverage this feedback to accurately assess its own performance. Yet we find a persistent reflection gap: LLM agents tend to mis-assess their own outputs after observing concrete environment feedback -- even for questions they correctly answered -- and standard RL barely helps due to a credit-assignment mismatch. To close this gap, we propose RefGRPO, a simple yet effective fix that augments standard RL algorithms with two key ingredients: a free calibration bonus computed by contrasting the agent's own reflection with the actual outcome (requiring no additional reward model, LLM judge, or external annotation), and a dynamic schedule on its coefficient. Compared to standard RL baselines, our method simultaneously improves reflection calibration (e.g., reduces underconfidence rate $44.4\% \to 7.7\%$) and task accuracy (e.g., $75.1\% \to 76.5\%$) on text-to-SQL across five benchmarks. The resulting calibrated reflection turns the agent into its own verifier grounded in environment feedback, which further enables (i) better self-improvement that uses reflections as pseudo-rewards without outcome supervision, and (ii) more effective test-time selective prediction by committing only to rollouts flagged as correct.

Forward-Forward (FF) learning [Hinton, 2022] replaces backpropagation with strictly layer-local goodness updates. Recent FF-CNN work has narrowed the gap to BP on 32x32 benchmarks, raising the question of whether layer-local training is becoming a viable alternative at realistic scale. To probe this rigorously, we develop DTG-FF -- dynamic temperature goodness, decoupled normalization, and multi-layer fusion -- as an instrument that sets FF-family state of the art across nine real-data benchmarks (91.8% CIFAR-10 and the first FF baseline at ImageNet-100 224x224), and use it to audit how far layer-local training actually scales. (1) Real-data scaling. Under identical recipe and backbone, an architecture-matched BP-DeepSup baseline beats DTG-FF by 2.40/5.93 pp on CIFAR-10/CIFAR-100, and the gap widens with class count. At 224x224 the same instrument reaches only 49.4% -- the first FF baseline at this scale, versus typical BP above 75% [Tian et al., 2020] -- exposing a real-data ceiling invisible at 32x32. (2) Synthetic vs. real K-conflict. DTG-FF increasingly outperforms BP as class count K grows on synthetic teacher-student tasks, yet on real images the FF-BP gap reverses sign and widens with K. A within-dataset CIFAR-100 coarse vs. fine probe isolates label-hierarchy from image distribution: synthetic K-sweeps confound output dimensionality with fine-grained discrimination difficulty and thereby overstate FF transferability. (3) Systems audit. FF can be implemented without storing depth-wide activations, but on commodity 8 GB hardware standard BP+gradient-accumulation reaches 4.18 GB / 157 imgs/s versus DTG-FF's 7.90 GB / 138 imgs/s, so a memory-based justification for FF at this scale is not supported under fair baselines.

Multimodal LLMs are increasingly used to assist scientific peer review, where a core requirement is verifying whether claims in a paper are supported by its evidence. Prior work has shown that models perform substantially better at this task when the evidence is a table than when it is a chart of the same underlying data. This raises the question of whether models fail to extract information from charts, or do they extract it but fail to use it when forming their prediction? We study this question through layer-wise linear probing and attention analysis on three open-weight VLMs over table and chart evidence, representing the same underlying data. We find consistent evidence for the latter. Chart information is encoded in the models' intermediate representations but does not reach the prediction position, a gap that is absent for tables and holds across all conditions tested. Attention analysis further reveals that this disconnect takes two architecturally distinct forms across model families. These findings reframe the table-chart gap as a failure of how encoded visual information is routed at prediction time, rather than a failure of encoding itself.

Autoregressive decoding in Transformer-based language models relies on the KV cache, whose memory footprint grows linearly with sequence length and becomes the primary bottleneck for long-context inference. KV cache eviction addresses this by retaining a fixed-size subset of key-value pairs and discarding the rest. We identify that a primary source of output degradation is not the residual attention mass on evicted tokens, which existing methods already minimize, but a directional mismatch between the retained and evicted token sets. Specifically, the evicted tokens in practice are often near-orthogonal to the retained ones. Thus, even a small evicted mass could have an oversized impact on the resulting direction distribution and amplify into substantial output error. This reveals a fundamental limit in existing strategies. To address this, we propose MomentKV, which maintains compact, small-size moment statistics over the evicted token set, including a count, key mean, value mean, and value-key covariance. During eviction, the moment statistics is leveraged to identify tokens already well aligned with and captured by the accumulated summary, keeping the evicted set geometrically regular. During inference, they yield a closed-form first-order approximation of the evicted attention output, forming a mutually reinforcing loop between selective eviction and accurate correction. On LongBench and RULER with LLaMA-3.1-8B-Instruct and Qwen3-4B-Instruct, MomentKV outperforms all baselines at every cache budget, with the largest gains under aggressive compression.

We describe ConvMemory, a small 3.6M-parameter learned reranker for conversational long-term memory retrieval, trained with cross-encoder teacher supervision over fused dense and lexical features. On the LongMemEval memory family, ConvMemory operates above the BGE-large cross-encoder in Recall@10 at 12-47x lower latency, remains within 0.025 Recall@10 of mxbai-rerank-large-v1 on Clean500 while running 28x cheaper; under Stress1000 distractors the Recall@10 gap widens to 0.081 but ConvMemory still operates at 117x lower latency; these LongMemEval numbers are single-run or single-seed and are reported as indicative cost-frontier evidence, not benchmark-grade. We then publish a rigorous negative attribution result on a previously claimed mechanism: a five-seed retrained ablation with paired bootstrap shows that ConvMemory's learned temporal window is statistically significant on aggregate but not temporally specific, with the largest effects on hard non-temporal controls and no significant effect on multi-hop temporal queries. The honest description of the mechanism is cheap cross-encoder distillation in a fused dense+lexical feature space, not temporal-structure exploitation. We additionally release CCGE-LA, a low-amplitude conflict-aware candidate-set editor over ConvMemory, as a research preview with modest but consistent gains on supersession and stale/rescue slices on LoCoMo. All results are retrieval-stage; ConvMemory does not match mxbai-rerank-large-v1 in absolute LoCoMo MRR, and the report is single-author and not yet independently audited.

We study stochastic density control between Gaussian-mixture endpoint distributions under Brownian prior dynamics. Since the direct Schrödinger bridge between Gaussian mixtures is generally not available in closed form, we introduce a lifted path-space construction in which each trajectory is augmented with a source--target component label. Consequently, the problem decomposes into Gaussian component-to-component Schrödinger bridges with explicit marginal, drift, and cost formulas, while the mixture-level assignment reduces to a finite-dimensional entropic coupling problem with a Sinkhorn scaling form. We then analyze the projection obtained by discarding or forgetting the label. By construction, the projected law satisfies the original Gaussian-mixture endpoint constraints, but its relative entropy generally differs from the lifted relative entropy by a nonnegative conditional label-information gap. This gap reveals a path-space obstruction: the lifted optimizer cannot, in general, be identified with the direct unlabeled Schrödinger bridge after projection. We also derive the posterior-averaged Markov drift associated with the projected marginal flow, prove a kinetic-energy upper bound, and identify a common path-potential condition under which the projection gap vanishes. Several numerical illustrations showing density and shape control are recorded for a self-contained exposition.

This work investigates the ``small-vs-large gap'', where repeating on fewer samples can lead to compute saving during training compared to using a larger dataset. This is observed across algorithmic tasks, architectures and optimizers and cannot be explained using prior theory. We argue that the speedup comes from appropriate layer-wise growth enabled by sampling biases, which is more pronounced when the dataset size is smaller. We provide both theoretical analysis and empirical evidence from various interventions. Our results suggest that using a smaller dataset with more repetitions is not just a fallback strategy under data scarcity, but can be proactively leveraged as a favorable inductive biases for optimization, particularly in reasoning tasks.

Object counting is a foundational vision task with over a decade of dedicated research, yet state-of-the-art models still fail systematically in the mixed-object setting that dominates real-world applications such as industrial inspection and product sorting. We show that this gap is strongly driven by limitations in existing training and evaluation data: real counting datasets are prohibitively expensive to annotate and suffer from labeling noise, while existing synthetic alternatives lack diversity and realism. We address this with MixCount, a dataset and benchmark for mixed-object counting designed to target the failure modes of current counting models. To overcome the high cost of constructing and labeling such data, we develop an automatic generation pipeline that synthesizes images, fine-grained textual descriptions, and pixel-perfect counting annotations at scale, eliminating the labeling ambiguity that plagues prior datasets. Evaluating state-of-the-art counting models on MixCount exposes severe degradation in the mixed-object setting. More importantly, training these models on our synthesized data yields substantial gains on real-world benchmarks, reducing MAE by 20.14% on FSC-147 and by 18.3% on PairTally. These results establish MixCount as both a benchmark and a training dataset for fine-grained counting, and demonstrate that our pipeline, which produces effectively unlimited labeled data, helps address a long-standing bottleneck in counting models.

When researchers ask whether two transformer layers are "equivalent" for compression, they often conflate distinct tests. Replacement asks whether one layer's map can substitute for another's in place; interchange asks whether two layers approximately commute when their positions are swapped. Both are output-grounded swap-KL probes, but they need not agree: on pretrained transformers the protocol gap can change which layers look safe to prune by several-fold under the same evaluator, especially when replacement distances are high. We measure both protocols across checkpoints and architectures. On a Pythia training trajectory (410M and 1.4B), the replacement-interchange gap grows from initialization to convergence. Under one matched WikiText-2 contract at 8B scale, Qwen3-8B enters a divergent regime: interchange-guided removal is several-fold safer than replacement-guided at the same layer budgets, while Llama-3.1-8B ties the two protocols for pruning cost even though interchange KL is lower, showing metric gaps need not map one-to-one to removal. Before layer removal or merging, score both swap-KLs on the target checkpoint; the diagnostic requires only unlabeled forward passes.

Geometric differences between cross-view images, such as drone and satellite views, significantly increase the challenge of Cross-View Geo-Localization (CVGL), which aims to acquire the geolocation of images by image retrieval. To further enhance the CVGL performance, this paper proposes a parameter-efficient adaptation framework for bridging the geometric gap across images based on the vision foundation model (VFM) (e.g., DINOv3), termed BGG. BGG not only effectively leverages the general visual representations of VFM and captures the robust and consistent features from cross-view images, but also utilizes the generalization capabilities of the VFM, significantly improving the CVGL performance. It mainly contains a Multi-granularity Feature Enhancement Adapter (MFEA) and a Frequency-Aware Structural Aggregation (FASA) module. Specifically, MFEA enhances the scale adaptability and viewpoint robustness of features by multi-level dilated convolutions, effectively bridging the cross-view geometric gap with small training costs. Additionally, considering the [CLS] token lacks spatial details for precise image retrieval and localization, the FASA module modulates patch tokens in the frequency domain and performs adaptive aggregation for local structural feature enhancement. Finally, BGG fuses the enhanced local features with the [CLS] token for more accurate CVGL. Extensive experiments on University-1652 and SUES-200 datasets demonstrate that BGG has significant advantages over other methods and achieves state-of-the-art localization performance with low training costs.

Large vision models (LVMs) have recently proven to be surprisingly effective time series forecasters, simply by rendering temporal data as images. This success, how ever, rests on a largely unexamined premise: the rendered time series images are sufficiently close to natural images for knowledge in pre-trained models to transfer effectively. We argue that two gaps still remain, i.e., spectral and structural gaps, fundamentally limiting the potential of LVMs for time series forecasting. Spectrally, we systematically reveal that rendered time series images exhibit a markedly shallower power spectrum than the natural images LVMs are pre-trained to recognize. Structurally, reshaping 1D temporal sequences into 2D grids fabricates spurious spatial adjacencies while severing genuine temporal continuities, misleading the spatial inductive biases of pre-trained LVMs. To bridge these gaps, we propose SSDA, a dual-branch network that spectrally and structurally adapts to unlock the full potential of LVMs for time series forecasting. At the data level, a Spectral Magnitude Aligner (SMA) applies 2D FFT to selectively enhance the magnitude spectrum toward natural-image statistics while preserving phase. At the model level, a Structural-Guided Low-Rank Adaptation (SG-LoRA) injects position-aware temporal encodings into patch embeddings and adapts at tention via low-rank updates. The two branches are further adaptively fused to produce the final forecast. Extensive experiments on seven real-world benchmarks demonstrate that SSDA consistently outperforms strong LVM- and LLM-based baselines under both full-shot and few-shot settings. Code is publicly available at https://anonymous.4open.science/r/SSDA-8C5B.

Diffusion language models (DLMs) offer a structural alternative to autoregressive generation: denoising can update tokens in arbitrary orders or in parallel rather than along a fixed left-to-right chain. In practice, fast DLM decoding remains strongly order-sensitive and often drifts toward autoregressive-like trajectories. We trace this tension to compatibility. At each reverse-time step, a DLM provides local denoising conditionals over the unresolved tokens. Arbitrary-order denoising becomes well defined when these local conditionals compose into order-invariant pseudo-joints. We formalize this view by defining order-induced pseudo-joints and a local denoising circulation: the log-ratio between the two pseudo-joints obtained by swapping a pair of unresolved positions. This circulation is zero under compatible conditionals, and global order gaps decompose into sums of local circulations along adjacent swaps. We further separate incompatibility-driven path dependence from conditional-dependence error in parallel updates and from order-specific estimation error. The resulting framework provides inference-only diagnostics for testing when DLM decoding is genuinely order-free.

Autonomous vehicles (AVs) are widely known to follow conservative, rule-based motion policies that surrounding drivers can learn to anticipate. A direct consequence is that human drivers may accept shorter longitudinal gaps when cutting in front of an AV than when targeting another human-driven vehicle (HDV). We test this hypothesis using the Waymo Open Motion Dataset (WOMD), which provides 25,906 real-world highway scenarios at 10 hertz. An eight-criterion lane-change detector extracts 706 HDV-to-AV and 3,172 HDV-to-HDV cut-in events from the same traffic environment. The median accepted gap in front of the Waymo AV is 7.58 meters versus 9.57 meters for HDV targets, a 1.99 meter reduction that is statistically significant (p equals 5.76 times 10 to the negative eighth power, d equals negative 0.224) and persists under speed-matched resampling. Cut-in speeds toward the AV are 37 percent higher (51.7 versus 37.7 kilometers per hour, d equals 0.502), and 68.0 percent of AV-targeted cut-ins occur below the 10 meter gap boundary versus 51.8 percent of HDV-targeted events (chi-squared equals 60.5, p is less than 10 to the negative thirteenth power). These results reveal a systematic and safety-relevant asymmetry in human gap-acceptance behavior that warrants AV-specific calibration of both motion-planning safety envelopes and traffic simulation models.

While it is generally understood that zeroth-order (ZO) algorithms have an extra dependency on their number of iterations for any choice of parameters, compared to their first-order (FO) counterparts, in this work, we show that under several conditions, in expectation, ZO methods do not suffer from extra dimension dependencies in their convergence rates with respect to their FO counterparts. We look at optimisation algorithms from the dynamical systems perspective and analyse the conditions under which one can formulate the average of a ZO algorithm as the average of its FO counterpart with bounded perturbations with values dependent on design parameters. Then, using input-to-state stability properties, we show ZO methods follow the same decay rate as their FO counterparts and converge to a neighbourhood of the fixed point of FO methods, where its radius depends on the bound of the norm of the perturbations, which can be made arbitrarily small. The theoretical findings are illustrated via numerical examples.

Recent work has framed intelligence in verifiable tasks as reducing time-to-solution through learned structure and test-time search, while systems work has explored learned runtimes in which computation, memory and I/O migrate into model state. These perspectives do not explain why capable models remain difficult to deploy in open institutions. We propose intent compilation: the transformation of partially specified human purpose into inspectable artifacts that bind execution. The relevant deployment distinction is closed-world solver versus open-world agent. In closed worlds, a checker is largely given; in open worlds, verification is distributed across semantic, evidentiary, procedural and institutional dimensions. Weformalize this residual openness as a closure-gap vector, define delegation envelopes as pre-authorized regions of action space, distinguish misclosure from undersearch, and outline benchmark metrics for testing when closure interventions outperform additional inference-time search.

Hypernetwork-based methods such as Doc-to-LoRA internalize a document into an LLM's weights in a single forward pass, but they fail systematically on conflicts: when the document contradicts pretraining knowledge, accuracy collapses to 46.4% on the deepest facts. We show the failure is a magnitude problem rather than a representational one. The hypernetwork already targets the right layers, but its adapter margin is approximately constant across documents while the pretrained margin grows with training frequency, so deep conflicts lose by construction. The account predicts that failure should track prior strength: sorting 194 conflicts by the base model's log-probability on the contradicted fact, baseline accuracy falls from 68% on weak-prior questions to 16% on strong-prior ones, a 52 percentage-point gap. The cure is amplitude. Selective Layer Boosting scales the adapter at its top-norm layers, and Conflict-Aware Internalization triggers boosting only when the base model is confident. Both are training-free; together they raise deep-conflict accuracy from 46.4% to 71.0% on Gemma-2B and from 53.6% to 72.5% on Mistral-7B while preserving novel-knowledge recall, and beat vanilla retrieval-augmented generation on medium conflicts by 18 percentage points despite operating entirely in parameter space. We release KID-Bench, a 489-question benchmark that separates novel recall, cross-knowledge combination, and prior-graded conflicts.

Chromatic Correlation Clustering (CCC) extends Correlation Clustering by assigning semantic colors to edges and requiring each cluster to receive a single color label. Unlike standard CC, whose LP relaxation has integrality gap 2 on complete graphs and admits a 2.06-approximation, the analogous LP for CCC has a strict lower bound of 2.11, and the best known LP-rounding algorithm achieves 2.15. We explain this gap by isolating the source of difficulty: cross-edge chromatic interference. Neutral edges, whose color does not match the candidate cluster color, create an irreducible cost absent from standard CC and force any color-independent rounding scheme to pay an additional mismatch penalty. We make four contributions. First, we prove a Global Integrality Gap Decomposition Theorem showing that the gap of any color-independent CCC rounding algorithm equals the standard CC gap plus an irreducible chromatic penalty Delta(L) > 0. Second, we solve the associated min-max problem and derive the staircase formula Delta(L) = ((L-1)/L) Delta_infinity, where Delta_infinity is approximately 0.0734. In particular, the two-color gap is 2.0967, separating CCC from standard CC already at L = 2. Third, we introduce Color-Coupled Correlation Clustering (C4). Adding the valid global constraint sum_c x_uv^c >= L-1 and a correlated interval-packing rounding scheme makes neutral edges behave like classical negative edges, recovering the optimal 2.06 approximation and bypassing the 2.11 lower bound for the uncoupled LP. Fourth, experiments on extremal instances, real multi-relational networks, and fairness benchmarks validate the theory: empirical LP gaps follow the predicted staircase, and C4 matches the unconstrained approximation ratio under fairness constraints.

Global greenhouse gas emissions estimates are essential for monitoring and mitigation planning. Existing emissions datasets provide critical foundations for understanding emissions patterns across sectors, geographies, and time scales. Through a structured assessment of recent emissions datasets, we identified opportunities to further increase the actionability of emissions data through more comprehensive source-level coverage, finer spatial and temporal resolution, and more frequent updates. Building on existing resources to address these opportunities, we present the Climate TRACE framework and resulting dataset, which is available on an open-access platform (climatetrace.org). The Climate TRACE framework synthesizes existing emissions data, prioritizing accuracy, coverage, and resolution, and fills remaining gaps using sector-specific estimation approaches. The resulting dataset is the first to provide global emissions estimates for individual sources (e.g., individual power plants) for most anthropogenic emitting sectors. The dataset spans January 1, 2021, to the present, with a two-month reporting lag and monthly updates. This dataset and open-access platform provides access to detailed emissions estimates for most subnational governments worldwide. By combining source-level spatial detail, monthly updates, and broad sectoral coverage, the dataset is designed to support analyses relevant to emissions monitoring and mitigation planning.

Reinforcement learning (RL) is often credited with improving language model reasoning at the expense of knowledge. We challenge this narrative by showing that reasoning models consistently outperform their instruction-tuned versions on pure knowledge recall tasks. These gains do not reflect newly acquired information, but rather an improved procedural skill in navigating and searching existing knowledge hierarchies within the model parameters. Structured prompting, which explicitly guides models through hierarchical traversal -- recovers most of the instruct-reasoning gap across five model families. A controlled RL experiment on unseen, non-extractable facts improves recall of held-out frequent but previously inaccessible facts, ruling out simple data exposure. On depth-stratified retrieval tasks, reasoning models exhibit superior traversal as retrieval depth grows. Layerwise activation analysis further shows that while factual representations maintain high cosine similarity between instruct and reasoning models, query representations diverge noticeably, indicating that reasoning primarily reshapes how models traverse knowledge rather than the knowledge representation itself. Finally, we find that distilled models often fail to match reasoning models on knowledge recall because they imitate self-correction without acquiring the exploratory behavior needed for hierarchical navigation. Together, these findings suggest that improving factual recall in LLMs depends not only on expanding what models know but also on teaching them to navigate it -- motivating future post-training methods that optimize traversal.