Mixture-of-Experts (MoE) enables multimodal large language models (MLLMs) to scale capacity efficiently, but expert parameters dominate memory even though only a few experts are activated per token: they account for over 90% of model size in modern MoE-MLLMs (e.g., 54 GB out of about 58 GB in BF16 for InternVL3.5-30B-A3B and Qwen3-VL-30B-A3B). Expert pruning has emerged as an effective compression strategy, but existing methods aggregate importance uniformly across tokens, ignoring that the same expert may serve different roles for vision versus language processing. To address this, we propose MAEP (Modality-Aware Expert Pruning), a training-free framework that decomposes expert importance by modality in a single forward pass. MAEP introduces (1) Delta-H Output ($\Delta H$), a redistribution-aware metric measuring the MoE output change when an expert is removed, and (2) low-attention vision tokens as a vision-side pruning signal that is dataset-stable: its expert-importance ranking remains stable across different datasets. Text-side importance serves as a preservation signal while vision-side importance serves as a pruning signal, and experts are ranked globally across layers. Experiments on three MoE-based MLLMs across multimodal and text benchmarks show that MAEP outperforms modality-agnostic baselines on the majority of multimodal benchmarks while remaining competitive on text-only tasks, reducing the BF16 expert weight footprint by about 27 GB (about 47% of total memory) so that the full weight footprint of an approximately 58 GB model fits within a single 40 GB GPU. Code is available at https://github.com/nota-github/MAEP.
Quantifying glomerular inflammation in kidney-transplant biopsies is traditionally performed on single whole-slide sections using the Banff "most-severe section" rule, despite the inherently three-dimensional nature of lesion distribution. This 2D based assessment might lead to instability when inflammation varies across slices. This paper reports a feasibility study examining the technical feasibility and clinical relevance of extending Banff scoring into three dimensions. To this end, we propose a 3D Banff lesion-scoring framework that aligns serial sections, reconstructs glomerular identities across depth, and integrates inflammatory-cell counts in 3D under three scoring strategies: Compatible, Volumetric, and Deduplicated Volumetric. In the experiments, glomerulitis (g-scores) was used as an example Banff metric and applied to multi-section renal allograft biopsies. Our findings indicate that 3D Banff g-scores are more consistent across sections and, under the semi-automatic setting, the Volumetric strategy correlates more strongly with clinical biomarkers than traditional 2D scores. These results show that 3D volumetric quantification offers promising added value, underscoring the potential benefit of 3D-aware Banff scoring for kidney transplant pathology.
Sparse attention accelerates few-step video diffusion transformers by computing a few key blocks exactly and compensating for the rest. However, existing sparse attention still faces key challenges, including identification overhead from finding critical blocks at every denoising step, suboptimal budget allocation that sets sparsity budgets uniformly or by local metrics, and cross-step error accumulation as reused masks or clusters go stale. Observing that the errors left on skipped blocks cancel each other once every key is compensated, we propose DecoupAttn, a new paradigm of decoupled sparse attention: semantic clusters compensate every key, and exact attention corrects their error on a few blocks fixed once offline by position. To support this paradigm, we introduce Distortion Budget, a rate-distortion objective for budget allocation that places exact computation across timesteps, layers, and heads where it removes the most distortion that compensation leaves in the final latent. Built on this budget, we further propose Reassign Cache, in which each step reassigns the drifting keys to cached centroids and refreshes compensation while the budgeted blocks stay fixed. Moreover, DecoupAttn is training-free: a few calibration clips compile one plan offline, and inference runs standard block-sparse kernels with one fused compensation kernel. Our code will be released after submission.
We study the minimum-disagreement cluster editing problem that, from a complete edge-signed graph, where a positive edge represents two vertices agreeing to each other while a negative represents a disagreement, constructs disjoint cliques of upper bounded size such that the total number of negative edges inside cliques and positive edges outside cliques is minimized. When the sizes of the candidate cliques are unbounded, the problem is NP-hard and APX-hard, and many approximation algorithms with improving ratios have been proposed in the literature and the state-of-the-art approximation ratio of $\alpha = 1.3865 + \epsilon$. The capacitated variant, in which the candidate cliques have sizes no greater than a given $k$, is less studied and the best approximation ratio is $5.37$. In this paper, we study the latter problem. We first present a downsizing procedure that achieves an approximation ratio of $\alpha_k = 2 \alpha + 1$, based on any $\alpha$-approximation algorithm for the uncapacitated problem. Secondly, we propose a hybrid of cluster-LP sampling and pivoting and show that for every fractional capacitated cluster-LP solution of cost $L$, the hybrid using ordinary pivoting has an expected cost at most $\frac 95 L + \frac 1{10} \OPT_k$, and the hybrid using conditional pivoting has an expected cost at most $\frac {306}{203} L + \frac {50}{203} \OPT_k$. A cardinality-constrained adaptation of the approximate dual-separation framework of Garc\'ia-Soriano and Schohn supplies the required fractional solution and yields a randomized $(1.9 + \epsilon)$-approximation algorithm and a randomized $(1.7537 + \epsilon)$-approximation algorithm, respectively.
Saliency maps, offered as evidence that mammography classifiers use the cancer, are untested in dense breasts, which can hide cancers. We tested whether they locate cancers worse there, and why. On 1,345 malignant CBIS-DDSM views, we scored three CNNs' Grad-CAM, Integrated Gradients and occlusion maps by whether their peak hits the lesion and a chance-normalized lesion-to-background saliency ratio (bits; +1 doubles it). Unadjusted gaps were large: Grad-CAM peaks hit 44% of dense vs. 58% of non-dense lesions. Case mix, chiefly more calcifications in dense breasts, explained about half or more of each gap. Occlusion kept a significant average of -0.56 bit gap. Controls also implicate the classifiers: unadjusted gaps exceeded an image-only conspicuity map's, and models relied less on dense-breast lesions. We recommend reporting density gaps before and after case-mix adjustment, per classifier, with conspicuity and lesion-reliance controls.
Developers insert logging statements in source code to record its execution, and the resulting logs are essential for detecting and troubleshooting system anomalies. A key part of these statements is the log level (e.g., DEBUG, INFO, WARN, ERROR), which determines whether a message is recorded. Choosing it is a trade-off: levels set too low mask important information in production, while levels set too high create noisy logs and extra I/O and storage overhead. Recent work uses Large Language Models (LLMs) with In-Context Learning (ICL) to predict log levels, but typically augments the prompt with randomly sampled examples. We argue that this overlooks the socio-technical nature of logging, where conventions vary by component and development team. Off-the-shelf LLMs using random examples are reported to achieve median AUCs of only 0.56 to 0.76, leaving room for improvement without fine-tuning. In this paper, we propose a retrieval approach for LLM-powered log level prediction that prioritizes context relevance over random selection and basic similarity. Our method retrieves logging examples relevant along two socio-technical dimensions: (i) functionality (code performing similar operations) and (ii) ownership (code maintained by the same developers). Evaluating four LLMs across ten popular Java projects, we find that this retrieval improves AUC over random baselines by 0.07 (median), and combining both signals yields median AUCs of 0.83 to 0.97. Our ablation studies show that each signal improves on random retrieval, and that combining them outperforms the stronger of the two on all ten projects. Our results also significantly exceed state-of-the-art supervised predictors (e.g., DeepLV) by a median AUC of 0.12 to 0.19. Our findings highlight the importance of socio-technical signals such as code functionality and ownership for LLM-powered log level prediction.
Forecasting emotion in conversation is increasingly framed as predicting the next emotional state's value. We show this targets a hard quantity in a state-observed testbed and propose a learnable, deployable reformulation instead. On valence-arousal-dominance trajectories from a dyadic corpus, repeating a speaker's own previous state sets a persistence baseline that no frozen-encoder correction meaningfully improves upon and a state-space filter exceeds; the floor requires the historical state to be observed, since an estimated anchor collapses persistence below a direct forecaster. What survives is the timing of a departure from persistence, learnable on three of four corpora and near chance on the highest-base-rate corpus, a negative result we report. We recast the task as selective departure forecasting: a lead-time-weighted metric with a leakage protocol, and commitment policies whose best choice switches with the departure rate. A dialogue-level conformal risk control bounds the expected loss of committed predictions, meeting the target in all twelve configurations tested, though the guarantee is bought through abstention, with coverage and alarm recall falling sharply on the hardest corpus. One tested language model's confidence fails as a ranking signal where another succeeds, a model-dependent failure not yet attributable to family versus scale. Code and result logs are available from the corresponding author upon request.
Comparisons in conversational emotion-shift forecasting are easily distorted when a learned model receives rich features but its transition baseline is given a noisy or unavailable current-emotion label, when hyperparameter budgets differ, or when target-turn information enters the representation. We present an information-matched, leakage-safe auditing protocol that closes these gaps with an explicit commitment point, executable causality tests, equal four-configuration validation searches, and seed-dialogue-aware inference. Applying it to six causal model families on four text corpora and two feature-source sensitivity configurations shows that a deployable train-only current-emotion classifier feeding the transition baseline yields a different verdict than an oracle diagnostic where both sides receive gold current emotion. Under the deployable condition, learned models improve over the predicted-label baseline on all four primary corpora; DailyDialog gains are large and survive a 72-comparison Holm correction (best ROC-AUC gain = +0.167, adjusted p=.0144). Under matched gold information, the baseline remains competitive on IEMOCAP and stronger on MELD, while every model retains a large DailyDialog advantage. We additionally report precision-recall AUC, calibration, a strictly utterance-local feature control, same-person self-shift forecasting, and a future-access dose-response test. An eight-item leakage checklist accompanies the protocol. Code, raw seed predictions, and reporting artifacts are available from the corresponding author upon request.
Identity-preserving video generation aims to synthesize high-quality and visually realistic videos from textual and visual conditions while faithfully maintaining the subject's identity across diverse motions. Prior methods have extensively introduced reference-guided control signals and identity-aware conditioning to improve facial alignment from reference images. However, improving static identity alignment alone creates a fundamental tension between identity fidelity and temporal controllability, which frequently causes identity drift or motion suppression under large pose variations and multi-segment sequential actions. To address this, we propose Chronos, a framework that unifies single-segment facial identity preservation and multi-segment sequential action generation under a common temporal formulation. Specifically, Chronos integrates Semantic Prompt Anchoring to adaptively refine the user prompt, Visual Frame Anchoring to transform conditions into visually grounded keyframe anchors for visual guidance, and Metric-Guided Candidate Routing to select the highest-quality videos from heterogeneous generators via multi-dimensional evaluation. Extensive evaluations demonstrate the superiority of our approach, which achieved first place in both Track 1 and Track 2 of the ACM MM 2026 Identity-Preserving Video Generation Challenge, demonstrating its effectiveness across both facial identity preservation and sequential action generation settings.
A learned cost model that predicts query runtimes is trained on measurements from one machine. When the database moves to other hardware, its plans stay the same but their runtimes change, each by its own factor. We ask how to adapt such a model from k queries measured on the new machine: which queries to run, how to use their runtimes, and how many are enough. On PostgreSQL we timed the same plans of 1,162 TPC-H and TPC-DS queries in six environments: a laptop and a server, each with all its processors or pinned to one, the server also pinned to two, and the laptop also with a small buffer pool. No single factor describes the slowdown between two environments: most of its variance lies between query templates, so a speed factor stops improving after a few queries. A correction that fits separate lines for serial and parallel plans and adds an offset for every measured template brings the moved model, with one measured query per template, to an error (geometric mean over templates of the median Q-error) of 1.10 on TPC-H and 1.19 on TPC-DS, close to the same correction on a model trained on the new machine; fine-tuning a DACE-style neural cost model on the same queries gains at most 2%. On templates never measured, no correction we tried is on average more than 2% better than the speed factor. We then fixed hypotheses and timed new data. On TPC-H at ten times the scale the errors follow these results, but two of five repeated tests miss. On the Join Order Benchmark, whose queries differ widely within a template, the model errs by 2.36 even when retrained on the new machine, no correction gains much, and only the test on new templates holds; boosted trees and a neural model fit these queries better, and correcting the neural model works as well as fine-tuning it. We report every test as registered.
Surrogate modeling of transport requires operators that capture the combined influence of spatially distributed sources and Dirichlet and Neumann boundary conditions, together with training datasets that can be generated at low computational cost. This work presents MIGONet, a multiple-input graph neural operator, and a discrete method of manufactured solutions (MMS) that trains it without solving a partial differential equation, using steady heat conduction as the model problem. In the discrete MMS procedure, temperature fields are prescribed a priori, and the source terms and Dirichlet or Neumann boundary data are evaluated with the same discrete operators as the reference finite-difference method (FDM), so every training sample satisfies the discrete governing equations by construction. Because the boundary data are derived from each field, a Dirichlet or Neumann condition can be assigned to every boundary segment of every sample, and the boundary specification becomes an operator input. MIGONet supplies representations of the complete source field and boundary data to every node before local message passing. On the rectangular domain, it is compared with a version without the complete-source encoder (MIGONet-local) and with the multiple-input operator network MIONet, using the same training data. MIGONet achieves mean relative temperature errors of 10.31% and 10.64% for curved-wire and Gaussian-spot heating, compared with 45.97% and 48.34% for MIONet and 72.25% and 67.82% for MIGONet-local. For 10,000 samples, discrete MMS reduces data-generation time from 486.65 s to 3.15 s relative to matched iterative FDM labeling. Separately trained MIGONet models are also evaluated on L-shaped, disc-shaped, and ring-shaped domains.
Humans can easily recognize the same objects across different viewpoints of a 3D scene, whereas teaching models such multi-view-consistent perception typically requires videos or calibrated multi-view captures, which are considerably more expensive and less diverse than large-scale monocular data.~In this paper, we introduce Feature Puzzles, a novel framework for learning multi-view-consistent segmentation using only single-image semantic segmentation supervision. Specifically, we propose a simple yet effective pipeline that constructs multi-view-consistent segmentation pseudo-labels through image cropping and geometric transformations. We further introduce a lightweight, plug-and-play network that leverages the multi-view-consistent image features learned by existing pretrained 3D foundation models and the generated pseudo-labels to perform multi-view-consistent segmentation. Importantly, our flexible framework supports various types of prompts provided from one or multiple views and can be readily extended to open-vocabulary 3D segmentation. Under this single-image supervision, our feed-forward model achieves the best mIoU on NVOS, remains competitive on SPIn-NeRF and ScanNet++ against baselines using per-scene optimization or in-domain multi-view training, and demonstrates zero-shot transfer to open-vocabulary 3D segmentation. Visual comparisons are available on our project page: https://featurepuzzles.github.io/
Temporal-difference (TD) learning underlies many prediction and control algorithms, but can diverge when combined with off-policy learning, bootstrapping, and function approximation. TD with Regularized Corrections (TDRC) is theoretically sound and practically simple, but existing convergence theory covers only the one-step case under i.i.d. sampling, while the Huber Bellman-error formulation has not been extended to TDRC or eligibility traces. In this work, we establish convergence of constrained TDRC($\lambda$) with diminishing stepsizes and non-i.i.d. off-policy data, for state-dependent and history-dependent $\lambda$-parameters. We also propose TDRC($\lambda$)-Huber, which preserves the TDRC($\lambda$) fixed point while reducing sensitivity to large Bellman errors. For both algorithms, we apply an ODE-based weak-convergence method from stochastic approximation theory. For the nonlinear mean dynamics of TDRC($\lambda$)-Huber, we give an integral small-gain condition that guarantees uniqueness and global asymptotic stability of the target equilibrium. We then establish convergence of the constrained algorithm, derive computable sufficient conditions for convergence, and identify a regularization-based convergence mechanism shared by linear TDRC($\lambda$) and TDRC($\lambda$)-Huber. When the mean TD matrix is positive stable, both constrained algorithms converge under sufficiently strong regularization without requiring positive definiteness.
Large language model (LLM) shopping assistants now recommend products in multi-turn conversations. However, reliable evaluation faces Data scarcity and Difficulty Standardizing Criteria. To address these obstacles, we introduce CARTBench for CART (Complex and Ambiguous Requests with Timelines). For Data scarcity, we curate cases from real commercial shopping needs across 17 retail domains, pairing timestamped dialogue with 1,603 product records. For Difficulty Standardizing Criteria, we develop case-specific rubrics guided by Ideal Response Principles, distilled from excellent replies manually selected from successfully completed orders. These principles also guide Principle-Guided CART Prompting, our structured recommendation strategy. The experimental benchmark contains 1,328 expert-annotated criteria across Core, Excellence, and Safety. To answer the CART question, we evaluate 7 flagship commercial API models, including GPT-6.1-Sol and Claude-Opus-5, and 35B+ parameters open-weight models. In the first full-suite evaluation under the frozen rubrics, 3 of 10 models exceed 90% on Core, while none exceeds 45% on Excellence. Therefore, we identify an understanding-to-execution gap between required behavior and high-quality purchase advice. To examine these limitations, we diagnose shared failures in product evidence and unresolved needs, linked to changing requirements and decisive unknowns. Code and data are available at https://github.com/Da1yuqin/CARTBench.
Various land-use models characterize land-use patterns and facilitate scenario planning. However, existing models are largely predictive, forecasting transitions based on numeric data without synthesizing new land-use maps from simple text prompts. Meanwhile, generative AI in urban systems remains unimodal, weakly controllable, and evaluated mostly by generic image quality rather than domain-specific metrics. To address these gaps, we introduce DualControlNet, a generative land-use model that jointly synthesizes high-resolution land-use maps and satellite imagery, addressing multimodality, controllability, and verification challenges. Trained on over 40,000 paired tiles from Orlando and Chicago, DualControlNet is evaluated using land-use, spatial consistency, and image quality metrics. Relative to unimodal baselines, DualControlNet improves land-use diversity consistency by 62.8% and building coverage ratio consistency by 16.3%, more than doubles cross-modal spatial consistency, and triples visual quality. Scenario-based experiments demonstrate that DualControlNet supports sustainable development through multidimensional controls, increasing density and diversity, and responding to text prompts like "high-density development." Overall, this work advances land-use modeling via a multimodal, controllable, and verifiable generative framework, lowering technical barriers and enabling intuitive exploration of urban futures.
Foundation-model predictors of perturbation response are evaluated primarily by aggregate accuracy. Using their predictions to prioritize experiments requires assessing each prediction's reliability. We audit the state-transition (ST) decoder used in STATE across three encoders (SE-600M, scGPT, and Tahoe-x1) and four cell lines under CRISPRi. The decoder's first attention layer has a low-rank output. Projecting this output onto 4-32 fixed directions estimated from training activations preserves mean per-perturbation Pearson delta correlation (PDC). This low-rank attention bottleneck bounds the decoder's reliability by distance to training responses: PDC declines as the cosine distance from a perturbation's observed response to its nearest training response increases. We propose predicted-NN-distance, the cosine distance from a predicted response to its nearest training response, as a per-prediction confidence score. The score has a mean absolute Spearman correlation of 0.81 with PDC, close to the 0.88 obtained with the observed response. Thresholds calibrated on this score transfer across encoders and cell lines, and the score remains predictive of PDC for drug perturbations.
Joint-Embedding Predictive Architectures (JEPAs) support planning by comparing predicted latent states with an encoded goal. For this comparison to guide temporal progress, states closer in time to the goal should receive lower costs. However, training representations for prediction and distribution regularization does not directly enforce this ordering. We examine this mismatch in flat LeWorldModel (LeWM) by analyzing how its Euclidean goal cost changes with temporal separation. Our analysis identifies conditions under which states at different long offsets receive similar average costs, weakening their temporal distinction. To learn the missing ordering explicitly, we introduce Time-to-Goal (T2G), a lightweight metric trained on latent triples from offline trajectories. A margin ranking loss encourages smaller distances for temporally nearer states, requiring no rewards or policy labels. The learned metric then replaces the Euclidean planning cost while the JEPA encoder and predictor remain frozen. Diagnostics on AntMaze and FourRoom show that T2G preserves stronger temporal contrast and improves near--far ranking at long offsets. Across four navigation and manipulation benchmarks, T2G improves the average planning success rate from 32.7\% (flat LeWM) to 63.0\% and approaches or even surpasses the performance of the hierarchical H-JEPA baselines. These results show that learning temporal ordering for the goal cost can substantially improve latent planning even with a flat world model.
User preferences shift as needs, context, and tastes evolve. This matters for personalized recommendation, LLM response generation, and RLHF-trained reward models, which can become miscalibrated as preferences drift. Existing non-stationary bandit methods forget uniformly across all dimensions, but real preference shifts are typically partial: some dimensions go stale while others remain valid. Neural contextual bandits accumulate confidence monotonically via the Gram matrix, with no mechanism to direct uncertainty where predictions have gone wrong. We propose PEARL (Prediction Error-Adaptive Regularized Learning), which reverses the Gram update along feature directions where predictions are surprisingly wrong, while leaving confidence intact elsewhere. We prove that (i) any non-adaptive policy suffers linear regret after a preference shift, (ii) no fixed exploration rate achieves low regret in both stationary and non-stationary regimes, and (iii) PEARL's cumulative regret scales with the number of shifted users N_s, not the total population N. On PRISM, a personalized LLM response selection benchmark (1,500 users, 21 LLMs), PEARL beats greedy by +4.0 points and the best neural bandit baseline by +2.5 points (p<0.001), while performing on par with greedy when preferences are stable. It beats sliding-window and discounted approaches by 3 to 10 points, showing that selective forgetting outperforms uniform forgetting. The selection policy is a ~100K-parameter MLP that runs in milliseconds on CPU, leaves LLM weights frozen, and works with any API-based model.
Firms using large language models under fixed budgets must decide which model to call and whether to retry a failed task. Because a call's charge is revealed only after completion, one long response can exhaust the remaining budget and leave later tasks unserved. Our objective is to maximize the number of tasks solved before service stops. We develop a contextual bandit policy, ADP-HT, which learns each model's task-specific success probability and constructs conservative cost estimates from coarsened response-length labels on a quantile grid with $2^K$ intervals. The policy weighs predicted success against these costs, adjusting the cost penalty to realized spending and the budget available per remaining task. Under calibration, local regularity, and grid-coverage conditions, we bound the excess allowance in the policy's conservative cost estimates. At a fixed sample size, increasing $K$ reduces the discretization term but increases the estimation allowance. At the tightest tested budgets, ADP-HT solves more than twice as many MATH500 and GPQA tasks as the best implemented baseline. It achieves the highest mean cumulative score at ten of fourteen RouterBench budgets. On LiveCodeBench, it outperforms a deployed commercial router in both default and cost-saving modes at the three lowest test budgets.
Decision models choose among predefined options in tasks such as routing and action selection. In multi-agent collaboration, a decision determines how members divide work, exchange resources, and act in parallel. We introduce TandemBench, a benchmark of 1,200 questions in multi-agent cooking scenarios where coordination choices affect goal completion. Models receive complete global information and candidate team plans, with reference answers established through execution and exact search. Models select an optimal team plan and, in paired variants that retain or remove it, judge whether better arrangements exist beyond the supplied options. We evaluate 40 native decision models and compare them with 10 general-purpose language models. The best native decision model achieves 62.50% decision accuracy. Native decision models remain close to chance in optimality verification. Paired tests reveal that some report better alternatives regardless of whether an optimal plan is already present. These findings show that team-plan decision accuracy alone can conceal systematic errors in judging whether the supplied options contain an optimal arrangement. The dataset and evaluation code are publicly available at https://github.com/lililibaohang/TandemBench.