Papers

Vision-language-action (VLA) policies achieve strong performance in robotic manipulation but remain vulnerable to runtime disturbances that break the temporal alignment among visual observations, robot states, and executed actions. We introduce ActFovea, a plug-and-play safeguarding framework that detects and mitigates such failures without retraining or modifying the underlying VLA policy. ActFovea uses robot kinematics, proprioceptive states, and recent actions to construct action-conditioned foveated regions that retain contact-relevant areas and predicted motion corridors while suppressing task-irrelevant visual content. It detects runtime risks by evaluating whether visual motion and observation freshness remain consistent with geometric, proprioceptive, and action transitions. For recoverable disturbances, ActFovea constructs disturbance-specific candidate observations and accepts a recovery only after verifying the resulting action chunk. When stale or replayed observations make reliable recovery impossible, it invokes a bounded safe-failure procedure. In closed-loop evaluations of $\pi_0$ across multiple LIBERO suites, ActFovea increases success under localized visual overlays from 49.3% to 90.3%, closing 93.7% of the gap to clean performance. It further improves success under action drift and visual delay by 7.0 and 9.8 percentage points, respectively, while preserving clean-task performance. Under frozen-observation replay, ActFovea triggers timely safe failure in all trials, with no unprotected failures. These results demonstrate that spatiotemporal visual-action consistency provides an effective basis for runtime safeguarding of VLA policies.

Long-term memory lets large language model(LLM) agents reuse prior preferences and work flows, but it also turns untrusted observations into persistent action context. We identify memory provenance laundering: during LLM-based memory consolidation, an external observation may be rewritten as apparent user history or workflow support, preserving an action trigger while erasing the low-trust source that should limit its authority. Existing prompt filters, content sanitizers, and tool guards do not enforce source-authority non-amplification after lossy memory consolidation. We formalize this boundary and instantiate it as Provenance-Preserving Memory Fire wall (PPMF), a lightweight memory middleware that preserves platform-maintained provenance and authorizes tool calls by matching action risk to the authority of action-relevant memories. In our schema-grounded evaluation with fixed risk policies, vulnerable consolidated memories reach up to 1.000 attack success rate(ASR); with intact platform-maintained provenance, confirmation, and risk labels, no evaluated unauthorized high-risk action passes the PPMF gate while confirmed benign actions and targeted low-risk memory use remain executable.

We introduce implicit machine learning force fields (I-MLFFs), which replace explicit stacks of neural network layers with self-consistent fixed-point equations. In molecular simulations, this formulation enables intermediate representations to be reused across successive timesteps, thereby warm-starting force evaluation. The resulting models effectively combine the computational footprint of a shallow, single-layer MLFF with the representational capacity and accuracy of a deep neural network. Our approach unlocks architecture-agnostic efficiency gains that are inaccessible when force prediction and trajectory integration are considered separately. We demonstrate this across three major classes of graph neural networks: invariant, equivariant Cartesian tensor, and SO(3)-equivariant spherical-tensor architectures. Each yields a two- to five-fold reduction in compute and memory footprint. Crucially, these gains are achieved while retaining full atomistic resolution and the original integration timestep, avoiding spatial or temporal coarse graining. Our contribution therefore advances the scaling frontier of quantum-mechanically faithful molecular simulation, enabling longer trajectories and larger atomistic systems within fixed GPU memory and compute budgets, and thereby opening access to new insights across biomolecular and material systems.

Synthetic face datasets are increasingly used to reduce privacy exposure and data access constraints in biometric recognition. Yet the generators that produce these datasets are trained on real faces, so synthetic data may still reveal their real source data. We study this risk through a dataset-level membership inference attack that first identifies the synthetic dataset used to train a face recognizer and then infers the real dataset used to train the generator. Across 11 face recognition models, 11 synthetic datasets, and 7 real datasets, the attack recovers the synthetic training dataset in 100% of cases and identifies the generator's source dataset in 54.5% of cases. These results show that synthetic data can retain dataset-level traces of real training data and that privacy-preserving deployment requires stronger leakage mitigation.

Neural operators provide fast surrogates for time-dependent partial differential equations (PDEs) by applying a learned evolution operator recursively to its own predictions, but this autoregressive rollout feeds every prediction error back as input, so local errors accumulate. Existing rollout-training strategies reduce the mismatch between training inputs and self-generated states, yet their supervision still measures only the absolute discrepancy from the ground-truth trajectory. Such supervision is therefore uninformative about whether the operator has overcome the long-horizon failure behaviors it exhibited earlier during optimization. We propose history-enriched rollout training (HERO), which augments conventional absolute trajectory supervision with relative supervision derived from the model's optimization history. HERO ranks detached candidate rollouts from a periodically refreshed lagged operator, the current model, and a perturbed input by rollout error, spectral discrepancy, energy drift, and error growth, and selects the strongest failure trajectory as reference. This reference enters a margin-based objective as a fixed comparison baseline, inducing a bounded, sample-dependent reweighting of the ground-truth rollout gradient rather than an independent gradient direction, which we further analyze theoretically. Experiments on nine PDE benchmarks with spectral and attention-based backbones show that HERO consistently improves long-horizon accuracy, stable rollout length, and out-of-distribution robustness at no inference-time cost. These results indicate that history-enriched relative supervision is effective for stabilizing long-horizon autoregressive prediction.

Recent work suggests that relational database management systems (RDBMSs) can execute quantum circuit simulation by compiling the simulation into SQL workloads (primarily join-and-aggregate tensor contractions). While early results are promising, they largely focus on a narrow set of highly structured circuits and offer limited support for systematic database research, such as query optimization, physical design, and engine-level evaluation across a broad range of circuits. We present InferQ, a database-oriented benchmark for quantum circuit simulation. InferQ generates general, compositional circuits by assembling subcircuits from a set of circuit templates, emits each simulation task as an RDBMS-ready SQL workload, and extracts circuit and query features (static, graph, SQL, and dynamic) for workload characterization. InferQ also releases a large dataset of 202,975 circuits online, with a web-based viewer to support searching, filtering, and downloading circuits and feature records. In experiments across RDBMS engines (PostgreSQL, SQLite, DuckDB, and Umbra) and the widely used Qiskit Aer simulator, we find that RDBMSs achieve better peak memory usage than Qiskit Aer on more than 50% of the circuits generated by InferQ. Moreover, using InferQ features, lightweight machine learning models (linear and tree-based models) can accurately predict when SQL execution is preferable (with accuracy up to 95.6% for runtime and 97.4% for memory), enabling data-centric simulator selection and opening the door to principled optimization of SQL-based quantum circuit simulation.

Link prediction aims to identify potential or future connections within a given graph structure. Position information is essential for link prediction, as it distinguishes homogeneous nodes through their relative relationships, facilitating the accurate capture of structural patterns and implicit connections. Previous studies derive node positional information as distances to single-granularity landmarks, defined as the centers of homophilic regions, while neglecting the multi-granularity nature of homophilic structures and their hierarchical interrelations. We propose the Multi-Granularity Position Embedding of Graphs via Granular-Ball for Link Prediction (MGLP) method to obtain multi-granularity position embedding of graphs. Specifically, MGLP introduces an Adaptive Granular-Ball Graph Refinement mechanism to adaptively refine the graph into homophilic subdomains with optimal levels of granularity. The central nodes within subdomains are treated as landmarks, which form a Hierarchical Central Graph. Moreover, a novel Multi-granularity Hierarchical Distance encoding mechanism is proposed to capture both the homophilic structures within a graph and their hierarchical correlations, improving the discriminative power of nodes. Experimental results demonstrate that the multi-granularity position embedding generated by our method exhibits excellent performance and strong competitiveness compared to baseline algorithms for link prediction. Our codes are available in https://anonymous.4open.science/r/MGLP-D3C5/.

Audio-visual speech recognition (AVSR) relies on effective fusion of audio and visual modalities, yet existing approaches treat cross-modal interaction as a single-step operation without structured iterative refinement. We present DoubleHelix, a multimodal fusion framework that reformulates fusion as an iterative cross-modal interaction process with adaptive degradation-aware enhancement. The framework comprises three components including ReverseParallelHelix for multi-turn structured interaction with learned alignment constraints, QualitySensor for learning degradation-aware gating signals, and HelixReplication for consistency-guided conditional feature enhancement. Experiments on LRS3 demonstrate that DoubleHelix achieves 0.68% WER on clean audio, outperforming previous best results by 5.6% relative improvement under matched backbone settings. Comprehensive ablation studies validate each component contribution, including targeted analysis of design choices such as asymmetric pathway weighting. The framework shows improved robustness under evaluated babble-noise conditions, achieving 11.6% WER at SNR -5dB.

Metasignal is an open-source Python package for signal detection theory (SDT) and metacognitive measurement. It implements the 17 metacognitive measures evaluated by Rahnev (2025), together with the reference variables d' (perceptual sensitivity), response criterion c (response bias), and mean confidence. The 17 measures comprise three meta-d' family estimates, meta-d', M-ratio, and M-difference; four nonparametric Type-2 measures, the Type-2 area under the receiver-operating-characteristic curve (AUC2), Gamma, Phi, and delta confidence, together with their eight SDT-normalized ratio and difference forms; and two model-based measures, meta-noise and meta-uncertainty. A single function computes the complete set from trial-level stimulus, response, and confidence arrays. `metasignal` currently supports binary (two-alternative) discrimination tasks, in which each trial's stimulus and response are coded with exactly two categories. The package also provides a command-line interface, group summaries, bootstrap confidence intervals, permutation tests, optional hierarchical Bayesian models, and information-theoretic measures. `metasignal` unifies these measures in a single platform to encourage broader metacognition research and adoption in decision-making studies.

Large language models (LLMs) are increasingly deployed in enterprise settings, yet individual models remain bounded by model-specific capability limitations. These heterogeneous boundaries pose a deployment challenge, but also create an opportunity: strategically coordinating multiple LLMs may unlock collective intelligence exceeding any single model. Existing approaches fix how models are combined in advance, overlooking the dynamic, state-dependent role of complementarity in complex problem solving. Drawing on the wisdom-of-crowds paradigm, we reconceptualize collective LLM intelligence as relay-style complementarity: a sequential process in which each successor model is selected to address the specific bottleneck identified in its predecessor's output. To operationalize this, we propose WILC (Wisdom Integration of LLM Crowds), a framework grounded in two design principles. First, iterative reflection-and-refinement establishes a state-preserving workflow through which models diagnose and refine prior outputs. Second, complementarity-driven model selection governs transitions via a dual-gate mechanism: prospective complementarity fit (PCF) identifies the worker most suited to the current bottleneck, while posterior complementarity gain (PCG) evaluates whether the selected transition improves the evolving solution. Experiments across four diverse benchmarks show that WILC outperforms existing approaches, including single-model self-refinement, ensemble methods, and query-routing methods. Under standardized pricing assumptions, WILC matches the average benchmark performance of GPT-5.2 at roughly 7 times lower estimated per-query cost, while facilitating data sovereignty through self-hosted deployment. This study extends wisdom-of-crowds theory from static aggregation to sequential AI complementarity and provides transferable design principles for multi-AI coordination.

Counterfactual explanations (CEs) enhance the interpretability of machine learning models by identifying the smallest change to an input required to obtain a desired output. Although CEs are conventionally formulated as a distance-minimization problem, the theoretical basis of this formulation has received limited attention. We show that a distance-minimization-based CE is mathematically equivalent to the maximum a posteriori (MAP) estimate of a Gibbs posterior within the generalized Bayes framework, specifically when a distance-based prior is used. We call this formulation the Distance-Prior Generalized Bayes CE (DP-GBCE). Building on this posterior perspective, we introduce two decision rules beyond MAP within a unified framework: a Bayes decision that minimizes expected decision loss and CVaR-CE, a risk-averse decision rule. We also propose an extension that uses Bayesian model weights to mix the posterior distributions of multiple models, thereby accounting for model multiplicity, where several models have comparable predictive performance. Finally, we define metrics for evaluating both individual CEs and the posterior distribution as a whole, and use experiments on simulated data and Google Trends data to quantify the trade-offs among the decision rules.

Federated learning of foundation models faces a fundamental resource-asymmetry challenge: the institutions holding the most valuable domain-specific data cannot host billion-parameter models. Existing heterogeneous federated approaches attempt to bridge this gap through parameter-efficient tuning, model pruning, or knowledge distillation, yet each trades away a critical property, whether full-model memory reduction, architectural self-containedness, or representational fidelity, leaving the core tension unresolved. We propose FedSLM, a parameter-centric framework for federated fine-tuning with heterogeneous compressed clients. FedSLM uses SVD-based decomposition to produce self-contained client models, whose low-rank subspaces form nested manifolds that are structurally compatible for aggregation. It then applies a two-stage protocol that synchronizes lightweight adapters within compression groups and fuses full-rank reconstructions across groups via structural alignment. Finally, a weak-to-strong elicitation step with auxiliary confidence loss transfers the aggregated knowledge to the full-scale server, while an explicit bias–variance trade-off mitigates compression artifacts. We provide theoretical guarantees for adapter-level aggregation, subspace-alignment bounds for cross-group fusion, and a characterization of how the confidence loss mitigates weak-supervision noise. Experiments on natural language and vision–language benchmarks show that FedSLM outperforms existing federated baselines under both IID and non-IID partitions, while client models operate at roughly 50% of the GPU memory required by the full model.

Deception detection has critical implications for legal proceedings, law enforcement, and online security. Although human judgment is limited in accuracy and scalability, Natural Language Processing (NLP) offers a data-driven alternative. We present a survey and comparative analysis of NLP-based Automatic Deception Detection (ADD) focusing on the legal domain, reviewing the evolution from feature-based machine learning to Large Language Model (LLM) approaches. We conduct a unified empirical evaluation across seven datasets (two legal, five general-domain), comparing six fine-tuned transformer models and seven LLMs under four prompting strategies. The results show strong domain sensitivity, with fine-tuned models excelling in data-rich general domains and few-shot LLMs remaining competitive in low-resource legal settings. Chain-of-Thought prompting often underperforms direct classification. These findings highlight the need for domain adaptation and interpretable systems in high-stakes legal contexts.

Police crash narratives contain information that may supplement structured crash databases, but manual review is labor-intensive and it remains unclear how well large language models (LLMs) reproduce official crash coding. This study benchmarked six frontier LLMs by comparing narrative-derived crash attribute codes with corresponding fields in the Arkansas fatal-crash database. The analysis linked 5,587 fatal-crash narratives with 5,889 structured crash records from Arkansas (2015-2025), yielding 4,194 matched crashes. Six LLMs were evaluated using an identical zero-shot prompt to code crash manner, non-motorist relation, intersection type, work-zone relation, roadway surface condition, and light condition. Performance was evaluated using agreement, macro-averaged F1 score, Cohen's kappa, coverage, selective agreement, and comparisons with always-majority, always-Unknown, and keyword-rule baselines. Repeated-measures analyses and a generalized estimating equations model assessed differences among models and attributes. GPT-5.5 High achieved the highest agreement among the evaluated LLMs, but the always-majority baseline produced higher raw agreement and the keyword-rule baseline achieved macro-averaged F1 score and Cohen's kappa comparable to the best-performing LLM. Agreement was highest for non-motorist relation and crash manner and lowest for light condition, roadway surface condition, and work-zone relation. Differences across crash attributes exceeded differences across models. These results provide a benchmark for evaluating LLM-based crash coding and show that deployment should be evaluated on an attribute-specific basis using transparent baselines and human review.

Model capabilities have improved in large part due to scaling chain of thought. This has been a promising development for AI safety–where models verbalize their reasoning, it is possible to monitor it. However, in some cases, models do not verbalize important steps in their reasoning process. For example, models prompted with a cue suggesting the incorrect answer may fail to acknowledge that cue, even when it appears instrumental to their conclusion. When chain of thought (CoT) fails to disclose instrumental reasoning steps, we describe it as unfaithful. Prior work has shown that activation steering can be a useful method to improve faithfulness in CoT. We extend this line of work by studying how well steering for faithfulness generalizes across cue types, datasets, and methods of constructing the steering vector for three models (Gemma-3 4B, Qwen-3.5 9B, Gemma-3 12B) in a cued question-answering setting. While steering reliably increases cue acknowledgment for only the largest model (Gemma-3 12B), we find that when steering is effective, its effect generalizes broadly across cue types and datasets–in cross-cue and cross-dataset analyses, effect size is determined primarily by the evaluation setting, rather than the vector's train setting. How the vector is built also matters little–four construction methods, including one whose optimization target mentions no specific cue, yield similar effect sizes. Finally, we consider the possibility that steering promotes the salience of the cue and causes greater cue use, rather than targeting verbalization behaviors. However, we find no evidence for this–steering leaves the rate of cue use roughly unchanged while reducing hidden cue use, i.e., cue use that is not acknowledged.

Professional-document review is a constraint-checking problem in which decisions depend on relations among text, geometry, pages, and document revisions. We present an evidence-grounded pipeline that normalizes extracted facts, executes four-state rules deterministically, retains source spans, and escalates unresolved cases. We evaluate its PDF evidence allocator on 160 reference-based tasks from 29 construction projects using a repeated four-system test and a disjoint two-system breadth extension. In the repeated test, reallocating a four-image budget from retrieved page overviews to one overview and three overlapping tiles improves project-family standardized decision accuracy by 10.6 percentage points (95% project-cluster bootstrap CI: 4.3 to 18.0; exact p = 0.031). This effect does not persist in the broader block: Region-RAG changes accuracy by -4.1 points (95% CI: -10.2 to 1.9; exact p = 0.209), while an equal-image sensitivity favors page breadth. Exact finding-set recovery remains low, false passes remain common, and repeated-run agreement is poorly calibrated. The results identify a resolution-breadth trade-off rather than a universal advantage for region-focused evidence, motivating rule-aware evidence routing and expert review.

Multi-agent systems (MAS) are increasingly deployed to solve complex tasks. In case of incorrect or unsatisfactory outputs, users have to manually locate agent mistakes by inspecting agent trajectories (i.e., failure attribution) and provide feedback to refine the outputs (i.e., repair). Despite some recent work in MAS failure attribution, automated mechanisms to recover from such mistakes remain largely unexplored. To bridge this gap, we propose MARS, a search-based framework that formulates MAS repair as a Monte Carlo Tree Search (MCTS) process and navigates the vast space of potential repairs via diagnosis-guided expansion with taxonomy-augmented evaluation. Unlike standard MCTS, which evaluates a complete simulation via full rollout, MARS evaluates the agent trajectory using partial rollout to reduce token consumption. Furthermore, we introduce StateMAS, a large-scale MAS repair benchmark with 1,310 replayable multi-agent failure trajectories spanning four types of agent architectures and four LLM backbones. Experiments on StateMAS demonstrate that MARS consistently outperforms state-of-the-art methods, achieving an absolute improvement from 3.0% to 12.1% across all settings, while maintaining a comparable token consumption cost. The ablation study further confirms that taxonomy-augmented evaluation and diagnosis-guided expansion are critical to achieving these performance gains.

State-of-the-art neural network verifiers use the branch-and-bound procedure as their core solving mechanism. We introduce an inprocessing framework for neural network verification driven by the lookahead procedure. Under this framework, lookahead derives new lemmas over the phases of unstable ReLUs, which are collected into an implication graph that is used to prune the search space and vivify boolean cuts. We instantiate the framework in two state-of-the-art verifiers, Marabou and $\alpha$-$\beta$-CROWN, and demonstrate that it improves performance in both, proving up to 34% more instances unsatisfiable.

Single-cell RNA sequencing (scRNA-seq) has become an essential tool in modern cellular biology, and generating accurate synthetic scRNA-seq data is becoming increasingly important. Although diffusion models have achieved promising results in conditional scRNA-seq generation, existing guidance strategies, including classifier guidance and classifier-free guidance (CFG), rely on an unconditional branch trained to approximate the true marginal distribution, which may retain substantial gene-specific structure and limit guidance effectiveness. Inspired by recent work showing that diffusion models can be effectively guided using intentionally degraded references, we propose a sparsity-biased classifier-free guidance (SB-CFG) strategy for scRNA-seq generation. Rather than approximating the assumed "neutral" marginal distribution, SB-CFG introduces a deliberately under-informative sparse reference for the unconditional branch, removing gene identity while preserving only coarse sparsity statistics. This "bad" reference amplifies the contrast between conditional and unconditional predictions, leading to stronger and more effective guidance during sampling. We evaluated SB-CFG as a training-free sampling modification on five publicly available scRNA-seq datasets. Experimental results demonstrate consistent improvements over standard CFG-based sampling in terms of marker gene expression fidelity, cell-type consistency, and sparsity preservation, indicating that SB-CFG better captures biologically meaningful gene expression patterns.

Existing autonomous-driving world models typically perform dense prediction of future videos, occupancy states, BEV representations, or agent motion. We argue that planning need not reconstruct the complete future world, but only focus on scene features that affect future ego action. Based on this perspective, we propose Auto-JEPA, an action-oriented latent world model that learns continuous future driving intent through joint-embedding prediction. Given visual observations, egomotion history, and navigation commands, Auto-JEPA predicts an intent embedding aligned with the latent representation of the future ego trajectory. The predicted intent retrieves executable trajectories from a fixed trajectory memory, which are then ranked by a scene-conditioned candidate selection module. Auto-JEPA keeps the visual encoder frozen, requires no explicit perception annotations, and uses no learned trajectory generator. By optimizing only task-specific modules for trajectory representation, intent prediction, and candidate selection, Auto-JEPA achieves 91.3 PDMS on NAVSIM v1 and 89.1 EPDMS on NAVSIM v2. Semantic occlusion experiments show that masking dynamic-agent regions induces an average intent change 2.97x that of equal-area random masking. Moreover, occluding vehicles that affect future driving substantially changes the predicted intent and selected trajectory, whereas both remain essentially unchanged when non-influential vehicles are occluded. These results show that future-intent prediction encourages the model to focus on planning-relevant visual features and supports high-quality planning without dense future-world modeling.

Online shoppers increasingly turn to AI shopping assistants, using images and multi-turn dialogue to express and refine product needs that are difficult to articulate in text alone. However, existing benchmarks largely rely on text-only or synthetic requests, underrepresenting complex real-world shopping requirements jointly expressed through images and language. We introduce MMShopBench, the first real-log benchmark for multimodal, multi-turn shopping agents. Built from carefully cleaned and manually annotated shopping logs, MMShopBench provides ground-truth annotations of each request's purchase intent and mandatory product requirements. Agents must infer these requirements jointly from user images and multi-turn dialogue, retrieve candidate products through image and text search, and verify that each candidate satisfies all requirements using its product images and structured attributes. We evaluate representative open-source and proprietary models using an evidence-grounded multimodal protocol and construct a companion training set for fine-tuning an open-source model. To ensure reproducible experimentation, we build an offline shopping sandbox, where fine-tuning substantially narrows the performance gap between our open-source model and leading proprietary models, demonstrating the effectiveness of our training data.

High-fidelity radio fields are typically simulated for every scene–transmitter configuration or fitted separately to each scene, failing to exploit propagation structures shared across environments. We present Point2Radio, a foundation model that learns a transferable propagation prior from multiple environments. Given a material-aware point cloud and a transmitter (TX) setting, a common encoder produces a TX-conditioned scene representation that can be queried at arbitrary receiver (RX) locations. Task-specific query decoders map this representation to different radio quantities, e.g., three-dimensional (3D) path-gain (PG) fields and power angular spectra (PAS). At inference for a new scene, the model uses only a material-aware point cloud and transceiver queries, running in milliseconds on a single GPU without meshes or explicit path tracing. We evaluate PG prediction on a scene-disjoint split of a 337-scene corpus containing 86,272 TX-conditioned fields. Point2Radio achieves 0.871 dB mean absolute error (MAE), reducing error by 76.7% relative to a same-split UNet-style baseline. The same encoder also supports PAS prediction via a task-specific decoder. Experiments further show that light target-scene fine-tuning improves adaptation to a specific environment.

Large language model agents have shown promising capabilities in data-driven scientific discovery tasks, where an agent interacts with an execution environment and produces a statistical claim. Long-horizon scientific analysis remains constrained by the lack of process supervised environments over real-world scientific data. This paper introduces SciDisco, a scalable framework for training Scientific Discovery agents in process-verifiable environments. SciThèque compiles hypotheses, datasets, hidden evidence graphs, and verifiers into task environments where analytical progress can be checked during interaction. DAG-grounded trajectory synthesis uses these environments to construct verifier-filtered multi-turn demonstrations. DiscoPO then uses the environment as the source of training signal, assigning turn-level credit to actions that produce verifiable analytical evidence. Experiments show that SciDisco-14B reaches state-of-the-art on hypothesis-driven scientific data analysis benchmarks.

Zero-shot image captioning (ZIC) describes images without paired image-caption supervision during captioner training, relying on text-only corpora and frozen pretrained image-text scorers. Existing retrieval-augmented methods score image-text alignment once, at retrieval, then commit the captioner's autoregressive beam under language-model probability alone, leaving the decoder without further visual grounding feedback. Progress has stalled, with no method improving on the strict-regime best since 2024. We propose Adjudicated Captioning, an inference-time multi-agent framework that restores grounding feedback at multiple checkpoints over an unchanged IFCap captioner. First, we install a stronger frozen Retrieval Encoder at the input. Second, between retrieval and decoding we insert a frozen Cross-Attention Verifier that re-ranks the top-9 retrievals to top-5. Third, at the output beam we attach a learned Reranker pairing TriFuse, a multilayer perceptron, with MemAttend, a memory-attended transformer, the pipeline's only learned components; both are trained self-supervised by Borda-consensus distillation across the three frozen scorers, using no paired image-caption labels and no reference captions. Under the inductive headline protocol, with rerankers fit on the disjoint COCO Karpathy validation beam and applied frozen to test, the framework reaches CIDEr 117.6 and SPICE 21.9 on COCO Karpathy, up from 108.0 and 20.3 for IFCap, a +9.6 CIDEr gain, and +7.7 above NES, the strongest synthetic-image-augmented method at 109.9, without retraining the captioner. A training-free fixed-fusion baseline reaches 115.8 CIDEr, so +7.8 of the +9.6 gain comes from the non-learned architectural intervention and the remaining +1.8 from the learned rerankers. The same recipe transfers off-COCO without captioner retraining: +8.1 CIDEr on Flickr30k Karpathy and +5.7 on NoCaps overall.

Recent language models achieve strong performance across a variety of tasks, but conventional adaptation applies updates uniformly across training samples regardless of their local update benefit. We propose PARALLEL, a prefrontal-aligned reinforcement inspired approach for language-model learning. Inspired by the complementary roles of goal-related and uncertainty-related control, PARALLEL represents these forms of information as separate controller signals and combines them with the current model representation. A reinforcement-inspired controller assigns sample-dependent update intensity using immediate utility-cost feedback. PARALLEL therefore learns when and how strongly to adapt to each sample, prioritizing beneficial updates while limiting unnecessary parameter changes. PARALLEL uses available updates more efficiently than selective baselines while retaining 94.1–99.2% of Full-adaptation performance. Beyond multiple-choice reasoning, experiments on XSum and CNN/DailyMail show that PARALLEL retains 96.9–98.6% of the ROUGE-1 and ROUGE-2 scores achieved by Full adaptation and 98.8–98.9% of the corresponding ROUGE-L scores. When compared at the same cumulative adaptation time or GPU energy, PARALLEL achieves higher ARC accuracy and exhibits a more stable late-stage adaptation trajectory than Full adaptation in the representative run. These results show that learning when and how strongly to update each sample supports stable and efficient post-deployment stream adaptation while avoiding unnecessary updates.

The rapid advancement of image generation models has made it increasingly difficult for people to distinguish AI-generated images from real ones. To prevent the potential risks associated with the misuse of fake images, AI-generated image detection has gained significant attention. Existing methods neglect the inherent differences between real and fake images, thus lacking robustness and generalization ability. In this work, we innovatively investigate AI-generated image detection using bit-planes, and introduce the bit-reversed image. We propose a simple yet effective pipeline consisting of construction of bit-reversed images, gradient-based patch selection and a convolutional classifier. Besides, we provide a theoretical analysis from the mathematical perspective to demonstrate the validity of our approach. We also introduce two challenging datasets for AI-generated image detection. Extensive experiments verify the effectiveness of our approach across different settings, including cross-generator generalization, cross-dataset generalization and zero-shot performance. Without bells and whistles, our approach outperforms existing methods on over 40 benchmarks, and is nearly 100 times faster than counterparts. The code is at https://github.com/renxi-seu/RAID.

Renault, Bergeaud, and Bosquet (hereafter RBB) argue that dating LLM adoption as the first month in which an author's abstract is flagged induces a stopping-time selection that can produce a positive event-study path even when there is no causal effect. Although this mechanism is mathematically possible, it does not constitute proof of a null effect. Recalibrating RBB's own random placebo to the detector's realized flag rate, we show that the measured association stays well above this benchmark, so the artifact is too small to explain the productivity changes. We further re-estimate the association between LLM adoption and productivity with a series of complementary designs in which the timing artifact cannot bias the estimate: a before-and-after comparison that dates adoption in one year and measures output in another, a conservative control group for difference-in-differences, an intensity-based specification that never defines an adoption date, and a rank-based measurement holding the flag rate fixed. A positive productivity association persists across all of these estimates, while the same tests run on pre-ChatGPT placebo data return null effects. The artifact RBB identify is real but bounded, and it does not account for the pattern we report.

Adversarial training is one of the most effective defenses against adversarial attacks, yet the computational cost remains prohibitive at modern scales, especially for large language models (LLMs). While existing mitigation strategies, e.g., latent adversarial training (LAT), have been developed, they still incur a high computational cost. In this work, focusing on LLM-based classifiers, we investigate computation-efficient strategies to speed up LAT from two complementary perspectives: (1) Defense-side optimization: We explore representation fine-tuning (ReFT) within LAT, and reveal a potential issue if there is a mismatch on which tokens to apply ReFT and the attack. (2) Attack-side optimization: When computing adversarial attacks in each LAT iteration, we extract only the relevant circuits from the LLM to construct a lightweight surrogate model, avoiding the computation in the forward-backward passes through the full model during the attack generation. For both perspectives, we provide theoretical justifications and numerical evidence to illustrate the effectiveness of the proposed strategies. Compared to standard LAT with full fine-tuning, our method on average reduces per-step adversarial-training FLOPs by 29.6% while requiring only 0.0118% trainable parameters, at a moderate cost in robustness.

Large language model agents are increasingly evaluated as autonomous tool users, yet most benchmarks focus on bounded tasks with immediate success criteria. Real-world deployments often require Long-Term Coherence, the capacity to preserve purposeful behavior across extended horizons while adapting decisions to accumulated evidence. Evaluating this capacity requires a persistent environment in which actions constrain future choices, feedback arrives at heterogeneous delays, and incoherent behavior produces measurable cumulative effects. Seller-side e-commerce provides a suitable setting for this evaluation through recurrent and interdependent decisions over Product Sourcing, Listing and Pricing Control, Cash-Flow Management, and Mixed-Latency Feedback Adaptation. We introduce MerchantBench, a 365-day order-level simulation grounded in 98,843 real e-commerce product records and equipped with 26 tools for agent interaction. MerchantBench couples promptly observable Upstream Supplier Events with delayed Downstream Order Outcomes, requiring agents to follow individual order lifecycles and revisit earlier decisions. We evaluate eight LLMs under two agent frameworks in 48 runs, each spanning 365 simulated days. Our results reveal a substantial gap between even the latest LLMs and human participants, with the best LLM configuration attaining only 27.3% of the mean final net assets achieved by human participants.

AI-generated videos are becoming increasingly realistic and difficult to distinguish from authentic ones, which facilitates malicious misuse and poses growing threats to cybersecurity and social governance. Attributing AI-generated videos to their specific generative sources is therefore of critical importance for forensic investigation and legal regulation. However, most existing visual attribution methods focus on images and particularly rely on the image generation model, thereby lacking the ability to generalize to large-scale AI-generated video data. To address these limitations, we introduce an training-free AI-generated video attribution paradigm. Specifically, we formulates AI-generated video attribution as an instance retrieval task, and design a generative fingerprint-based pipeline. This pipeline consists of an adapted orthogonal color transformation, multi-scale quantized residual generation, and temporal-semantic aggregation, progressively capturing and integrating artifacts introduced by generative models across video frames. Extensive experiments on the GenVidBench benchmark demonstrate that our method achieves strong performance in both AI-generated video detection and attribution, outperforming existing state-of-the-art methods with a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6%. The code is at https://github.com/renxi-seu/Video_Attribution.