Yuetong Liu, Zhehao Zhou, Yitong Zhang, Haorui Chen, Jia Li
Large language model (LLM) agents typically append raw tool outputs to their context. Over multi-step interactions, long and noisy observations incur substantial context overhead and dilute the agent's attention to task-relevant information. Existing token pruning and summarization methods remain constrained by discrete text representations, limiting the achievable compression ratios. In this work, we challenge the prevailing assumption that agents must consume tool observations as explicit text tokens. To this end, we propose Coca-Cola, a continuous latent context compression model for LLM agents. Coca-Cola compresses tool observations into latent tokens and embeds them into the input embedding space of the agent. We further introduce Action-Preserving Training, which aligns the agent's next-action distributions under full and compressed observations. This trains the compression model to learn action-sufficient latent tokens that retain decision-critical information. Experiments with web-search agents on four QA benchmarks and two model families show that Coca-Cola reduces the agent's token consumption by 70.01%-89.24% and end-to-end runtime by 24.45%-55.97%, while consistently improving task performance. These results suggest that Coca-Cola offers a promising approach for context compression in agents. Our code is available at https://github.com/THU-Agent/Coca-cola.
Zhehao Zhou, Tongrui Su, Shengyu Zhu, Da Li, Zeyu Liu, Jun Zhu, Xueqi Cheng
Latent Chain-of-Thought (CoT) reasoning improves efficiency by shifting intermediate computation into continuous hidden states. However, fully latent methods often struggle on precision-critical tasks such as mathematical reasoning, while existing hybrid approaches retain substantial token-generation overhead. We introduce VITAL, an importance-guided hybrid reasoning framework that selectively verbalizes informative tokens as reasoning anchors while keeping most computation latent. A lightweight gating head dynamically selects between explicit token generation and direct hidden-state feedback. VITAL constructs hybrid trajectories from explicit CoT using token- importance estimation and trains the model through curriculum-based supervised training, followed by reinforcement learning with a reward that balances correctness and conciseness. Experiments show that VITAL consistently out- performs existing fully latent reasoning methods, achieving relative average accuracy gains of 9.6% on GPT-2 and 5.1% on Llama-3.2-1B at comparable reasoning lengths. Compared with hybrid baselines, VITAL achieves higher reasoning accuracy with substantially shorter reasoning trajectories on Qwen3-1.7B.
Text-to-SVG systems are commonly evaluated by applying raster-image metrics to rasterised outputs, yet whether these scores capture fine-grained prompt faithfulness remains unclear. We introduce SVGFaith, a two-layer meta-evaluation framework in which the more faithful candidate of each pair is determined by construction or by human judgment. SVGFaith-Diagnostic contains 1,842 controlled contrastive pairs across eight capability axes, split evenly between two directions: two renderings compared against one prompt, and two descriptions compared against one rendering. SVGFaith-Arena contains 1,327 human-annotated comparisons of real generator outputs. On SVGFaith-Diagnostic, CLIP and three other image--text embedding metrics reach at least 0.789 pairwise accuracy on object identity but at most 0.672 on spatial relations and attribute binding, where chance is 0.5. We further introduce SVGFaith-Score, an interpretable pointwise metric. A domain-tuned decomposer turns each prompt into a dependency graph of atomic visual requirements, a visual question answering model scores whether each requirement is satisfied, and a requirement's score is discounted when its prerequisites score low. In addition to the overall score, the per-requirement scores show which requirements are judged unsatisfied and which inherit discounts from their prerequisites. SVGFaith-Score significantly outperforms all nine baselines, including a Qwen3.5-122B-A10B judge, on Arena and in both Diagnostic directions. Code and data: https://github.com/Yukinonooo/SVGFaith.
We study the component-oracle complexity of finding approximate local minima of nonconvex finite sums under mean-squared averaged smoothness with parameter $L$ and a $ρ$-Lipschitz average Hessian. Each query returns the value, gradient, and one Hessian-vector product of a component. We show that the complexity in the high-accuracy regime $\varepsilon\le c_0L^2/(nρ)$ is $\tildeΘ(n^{3/4}L^{1/2}ρ^{1/4}Δ\varepsilon^{-7/4})$ for a sufficiently large initial gap $Δ$. We attain the upper bound by adapting the inverse solver and eigenvector routine of FastCubic to averaged smoothness. Our lower bound assigns the bridge edges of a nested chain to random components; a replicated spectral construction gives the same bound for unrestricted randomized algorithms. Under individual smoothness, a factor of $n^{1/4}$ remains for $\varepsilon\le cL^2/(n^2ρ)$, up to logarithmic factors. In the complementary low-accuracy regime, we propose PathVR, which uses $\tilde O(n+\sqrt{n}LΔ\varepsilon^{-2})$ queries and matches a lower bound from dense weak hiding.
We introduce LLM4Robot-Bench, a hierarchical, paired Sim-Real benchmark for evaluating physical interaction capabilities in contact-rich manipulation. Its 16 household tasks span three cumulative levels: contact selection and geometric alignment (L1), force regulation within sustained contact (L2), and coordinated contact reconfiguration and force regulation (L3). Each task is instantiated in physically and visually matched simulated and real-world environments, with shared task definitions and success criteria. We evaluate GPT-6 Astra zero-shot with three trials per task in each environment. Real-world success rates decline from 80.0% at L1 to 47.6% at L2 and 0.0% at L3, with corresponding simulation rates of 66.7%, 47.6%, and 0.0%. Ten of the 16 tasks have identical success counts across environments; the remaining six differ by one success out of three trials. Across both environments, recurring difficulties in spatial reasoning, contact selection, force regulation, and coordination often prevent local physical progress from accumulating into task completion. These findings characterize Astra's observed capability boundary across increasing interaction demands and, for this model, support matched simulation as a repeatable, scalable testbed for studying the sustained, coordinated physical interactions that limit reliable manipulation.
Diffusion-based recommenders learn user preferences by denoising interaction histories, but the added noise can distort the original preference signals. Flow-based collaborative filtering (CF) instead learns a discrete path from a sampled source to the user's observed training interactions (the target history), enabling inference in a few steps. Existing flow-based CF methods sample sources using global item frequencies to reflect the sparsity and popularity of observed interactions. Each source is sampled independently of its target history, leaving the predictor without user-specific source cues for learning to reconstruct that history. We propose SourceFlow, which brings collaborative structure into the source distribution itself rather than adding further conditioning to the predictor. Like inference from observed histories, training uses collaborative evidence to predict missing interactions. Specifically, Neighborhood Coupling constructs displaced binary sources by replacing historical items with unobserved collaborative alternatives. Collaborative Consensus further refines replacement probabilities using evidence from other interactions of users who share the observed item. Stretched Restoration retains these replacements at late flow times during training, providing more opportunities to recover the original interactions. Experiments on five recommendation datasets show consistent improvements over competitive baselines, including a mean relative Recall@10 gain of 15.7% over the strongest flow-based baseline. Controlled experiments confirm that these gains depend on correctly pairing collaborative sources with their target histories, as reassigning the same source vectors across users reverses the gains. Code will be publicly released upon acceptance.
Predicting stock returns requires modeling persistent company relations and time-varying market signals. Graph Neural Network (GNN) approaches combine static priors with dynamic relational graphs, but treat every extracted edge as a trustworthy binary fact, ignoring heterogeneous reliability across relational sources. We propose LENS (LLM-Extracted priors and News-driven Semantics), a static--dynamic graph framework for stock return prediction and ranking that models relational reliability as a first-class signal across the pipeline. On the static side, we extract a multi-relational company graph from a Large Language Model and aggregate it over multiple runs with confidence weighting to downweight low-consistency edges. On the dynamic side, we build a news-driven semantic graph with exponential-decay temporal aggregation to capture changing market context under uneven coverage. A context-dependent gated fusion module rebalances the two graphs at the per-stock, per-day level. On S&P 500, LENS improves AR@5 by 16.4% and SR@5 by 22.6% over the strongest baseline, and on NASDAQ 100 it achieves comparable improvements (15.5% and 20.9% respectively). Ablations isolate each component's contribution, and a human-validation study of sampled relations supports that the aggregation is grounded, providing evidence that reliability-aware LLM relations and news-based semantic graphs carry complementary information for graph-based financial prediction.
Large language model (LLM) services face rapidly rising serving costs. To sustain free or affordable access for a broad user base, alternative monetization mechanisms are needed. Native advertising offers a promising approach, as it creates economic value by aligning user intent with merchant offerings. Yet how to integrate sponsored content naturally into LLM responses remains largely underexplored. Existing approaches primarily rely on prompt-level injection, multi-pass rewriting, or fine-tuning, which can compromise response quality or introduce additional computational costs. We propose AdSteer, a lightweight decoding-time advertising framework that adaptively adjusts token probabilities to integrate contextually appropriate advertisement into responses. AdSteer calibrates the strength and duration of advertising influence, making sponsored content a natural part of the response. Experiments on GEM-Bench across three application scenarios show that AdSteer achieves an average improvement of 14.1% in overall score over the strongest baselines and 22.5% fewer total tokens on average than rewrite-based methods. These results demonstrate that AdSteer provides an effective and practical solution for native advertising in LLMs.
This paper studies retrieval-augmented generation (RAG), which leverages external knowledge to improve the performance of language models. Despite the remarkable progress, current RAG approaches are still far from satisfactory due to inadequate retrieval and potential hallucinations. Towards this end, we propose a novel approach named Task Refinement with Individual Optimization (TRIO) for RAG. The core idea of our TRIO is to equip three well-designed agents with an iterative refinement paradigm by aligning their individual optimization objectives. Specifically, we first expand ambiguous queries with a query extension agent and quantify the reward based on the improved answer quality, which can be utilized to refine our agent for higher expansion efficiency. To enhance the retrieval quality, we include a unanimity voting to annotate the expanded query as insufficient or sufficient, and their generative outcomes are utilized as ground truth to optimize the judge agent. To further improve evidence utilization and answer accuracy, an answer agent is optimized with dynamic matching-based rewards with curriculum learning for final outputs. Extensive experiments across multiple benchmark datasets validate the effectiveness of the proposed TRIO in comparison with state-of-the-art approaches. Our code is available at https://github.com/yingtaoren/TRIO.
Ziyang Su, Yao Zhou, Kelin Liu, Guanjie Zheng, Linghe Kong
Large-scale super-apps generate heterogeneous transaction logs in which users interact with multiple businesses, item identifiers are only partially available, and future value targets are sparse, heavy-tailed, and cumulative over time. We study multi-business, multi-horizon value prediction as a data-intensive analytics problem over such logs, rather than as a single-table task. (i) a batch-induced user-item graph that provides collaborative signals without materializing a global graph and remains applicable when item identifiers are missing through reserved unknown-item handling, (ii) business-aware sequence encoders that preserve recency, inter-event gaps, payment context, and goods-table availability, and (iii) cross-view contrastive alignment that transfers user-level intent across businesses while retaining business-specific temporal patterns. To handle zero-inflated and long-tailed cumulative targets, CoGSeq-LTV adopts a zero-inflated mixture log-normal head with MAE-aligned point supervision and validation-calibrated zero-aware inference. Experiments on the BiBiSheng transaction-log dataset show that CoGSeq-LTV improves over representative LTV baselines across businesses and horizons, including under a later out-of-time evaluation window.
Transforming raw neuroimage archives into analysis-ready derivatives relies on three brittle stages: data standardization, modality-specific preprocessing, and quality control (QC). While individual neuroimaging tools are well developed, their orchestration requires project-specific scripts, environment-adaptive tuning, and labor-intensive manual QC. To address this, we introduce NeuroPilot, a multi-agent system that digitalizes the expertise of neuroimage processing, QC, and data management into three LLM-invocable skills: dcm2bids-skill, neuroimage-pre-skill, and qc-agent-skill. The LLM-driven agent autonomously orchestrates workflows, generalizing various infrastructure settings into a single configuration to achieve the highest scalability. Demonstrating the system's generalizability, we deployed NeuroPilot across 17 cohorts (>123,000 subjects) spanning infant to aging populations and multiple MRI modalities (structural, diffusion, functional). In practice, after standardizing data via the dcm2bids-skill, the agent dynamically routes datasets to the optimal neuroimage-pre-skill based on available modalities and cohort traits (e.g., dispatching T1w and fMRI data to fMRIPrep, or selecting specialized pipelines for infant cohorts). The qc-agent-skill then drives an evidence-based, semi-automated QC via a 3-D browser dashboard, utilizing a multi-tiered verification system to optimize failed cases and escalate complex issues for supervisor inspection. Quantitatively, our QC agent screened 558 production subjects, validating its automated flags against FreeSurfer's topology-defect metrics. The infant processing pipeline achieved a 100% (201/201) completion rate on QC-validated inputs. Importantly, NeuroPilot compresses the traditional 2--3 month timeline for training staff and processing complete datasets into a single week. NeuroPilot is deployed in https://wanda-cyberbench.com/.
AI 速览：作者提出基于大语言模型的智能体 ClinGeX，通过检索与依赖感知编排 105 个可执行技能来辅助临床基因组解读，并在基准与人类-AI 协作研究中报告了优于对照的表现。
Clinical genomic analysis requires experts to integrate evolving guidelines, databases, and computational evidence. As sequencing has become increasingly accessible, this complex and labor-intensive process has become a major bottleneck to the timely use of genomic information in clinical decision-making. To fill this gap, we developed ClinGeX, a large-language-model agent that composes 105 curated executable skills through task-specific retrieval and dependency-aware orchestration. On 223 temporally held-out benchmark cases, ClinGeX achieved 71.7\% macro-averaged task-level concordance with expert reference classifications, compared with 45.7\% for the strongest comparator. In a three-condition, counterbalanced human-AI collaboration study, 45 clinical genomic professionals assessed 135 clinically derived cases under routine (no AI assistance), DeepSeek-assisted and ClinGeX-assisted conditions. ClinGeX assistance increased raw case-level concordance across criterion-level variant analysis, genomic interpretation in rare disease and molecular oncology decision-making (51.1\%, 57.8\%, 64.4\% versus 15.6\%, 31.1\%, 20.0\% in comparator conditions) and reduced reported interpretation time in all scenarios. During an exploratory one-month real-world non-interventional assessment, 11 experienced professionals submitted feedback on 96 self-selected analysis episodes and reported frequent conclusion adoption and time savings. These findings support executable domain skills as a promising approach to augment specialist genomic interpretation.
Before AI can be trusted to discover new science, it should be able to reproduce what science already knows. Reproduction provides a rigorous checkpoint for an AI scientist: the target claim is known, yet recovering it still requires translating an incomplete scientific description into an executable experiment and producing evidence that can be independently verified. Existing research-reproduction benchmarks largely focus on AI and machine learning, where experiments share relatively homogeneous software stacks. Extending this capability across scientific disciplines introduces a different challenge: agents must operate domain-specific scientific software, make physical and numerical choices, diagnose long-running computations, and distinguish valid evidence from merely plausible outputs. We introduce PAPERBENCHX, a benchmark of 93 expert-constructed paper-reproduction tasks spanning 12 research areas across electromagnetics, photonics, chemistry, materials science, biology, and robotics. Each task provides the source paper, a public task specification, and a domain-native scientific environment involving specialized scientific software. Agents must reconstruct a key experiment, execute and diagnose it, and deliver a runnable workflow that supports the target scientific claim. To evaluate scientific validity rather than surface-level agreement, we replay every submission from scratch and score only regenerated evidence against expert-validated criteria. Across 10 frontier agent configurations, the best configuration fully reproduces only 14% tasks under the 90% threshold. Meanwhile, mean rubric scores range from 27.53% to 59.08%, indicating that substantial partial progress rarely translates into complete scientific reproduction. These results reveal a large gap between paper understanding and reliably producing reproducible scientific evidence.
Reinforcement learning has advanced large language models on tasks with reliable verifiers, where checking an outcome provides a clear basis for reward. Extending this approach to open-ended tasks requires environments that can both evaluate complex, partially successful outputs and turn these judgments into effective learning guidance. We propose a reinforcement learning framework built around a training environment that integrates human domain knowledge with model reasoning. Human domain knowledge, encoded in rubrics, anchors evaluation to task intent, while model reasoning combined with hierarchical rewards distinguishes complex open-ended outputs and links diagnosed failures to likely responsible output regions. Furthermore, these diagnostic attributions guide token-level weights in the GRPO objective, thereby translating the ambiguous notion of success in open-ended tasks into concrete guidance for policy optimization. We instantiate this framework in interactive web application generation, a representative open-ended generation task. The resulting model achieves an average pass rate of 41.25\% on MiniAppBench and 76.19\% on ArtifactsBench. Ablation studies show the contributions of individual components, while additional out-of-distribution evaluations support transferable gains without degrading performance on other evaluated tasks.
Mingming Ding, Ruilin Chen, Yuzhen Huang, Hang Qi, Menglu Yu, San Tan, Damian Reeves, Boris Sarana 等 32 位作者
Lifecycle overhead silently consumes accelerator capacity across large-scale recommendation training fleets. Our largest recommendation workloads process tens of billions train- ing examples per day on thousands of GPUs. Before this work, only 50-60% of their end-to-end wall time advanced training on new data. We present a fleet-scale study of this lifecycle overhead and a set of optimizations spanning the full training stack. We use Effective Training Time (ETT%) as an operational framework to instrument lost time, localize it to independently owned infrastructure components, and expose work repeated across job restarts. This analysis guides optimizations like communication elimination and pipeline overlap during trainer initialization; dynamic-shape handling, autotuning pruning, and reusable Py- Torch 2 compilation caches; asynchronous checkpointing; stan- dalone model publishing; and reductions in recovery cost. We evaluate the optimizations on representative models and measure their impacts in our training fleet. ETT% improves on every benchmark, by 15.5% on average, and reaches 85% on our largest workload. Fleet-wide ETT% rose from about 80% to above 90% after deployment.
Recent vision-language navigation (VLN) systems increasingly adapt pretrained vision-language models (VLMs) into vision-language-action (VLA) policies that map egocentric observations and language instructions directly to navigation actions. Although semantically capable, such action-centric training does not explicitly model how the agent's visual observations evolve under its predicted motion. Generative world-action models (WAMs) jointly predict future observations and actions, yet existing WAMs for continuous VLN do not condition joint future-view and action generation on geometry-aware representations inferred from the observation history. We present WNM-3D, a generative World Navigation Model with 3D scene conditioning for continuous VLN. A frozen feed-forward geometry encoder extracts cross-view representations from monocular RGB history, and a trainable 3D Scene-to-Token Adapter converts them into a fixed-length scene prefix for the world-action Diffusion Transformer. Through block-causal attention, this prefix remains visible to every future video-action block throughout joint denoising, providing persistent geometric context. We train WNM-3D with supervised world-action learning, DAgger-style adaptation on policy-visited states, and Counterfactual DanceGRPO refinement. Across GN-Bench and VLN-CE, WNM-3D consistently outperforms the backbone-matched WNM-2D counterpart, demonstrating gains from 3D scene conditioning across distinct visual domains. WNM-3D further outperforms strong VLN baselines on GN-Bench and demonstrates closed-loop deployment on a physical robot. Component and stage-wise ablations support the Adapter design and the complementary roles of DAgger-SFT and subsequent reward-guided refinement.
Han Wang, Jingyuan Bao, Juntao Wu, Saiao Zhou, Yuman Nie
Event cameras, spiking cochleas and similar sensors emit asynchronous events, yet most learning systems first bin them into frames or packets, so the output depends on how the stream was sliced. Networks whose state is updated exactly at every event and evolves in closed form between events are invariant to this slicing. We ask what more makes the output (i) independent of the processing order of simultaneous events and (ii) continuous in the time stamps, even when two events cross in time, while (iii) staying sensitive to temporal order. For arbitrary nonlinear event models we prove that (i) and (ii) are each equivalent to pairwise commutation of the event updates, and that once updates commute, all order dependence comes from their non-commutation with the inter-event dynamics and vanishes linearly in the time lag. This identifies a class of time-mediated non-commutative (TM) models and yields a resolution-robustness bound for any model. We construct the minimal TM layer, a learnable, provably Lipschitz generalization of the Hassenstein-Reichardt correlator; prove that no single additive layer can uniformly represent ordered pair interactions, whereas a three-dimensional TM state can; show that each TM layer adds two levels of temporal nesting versus one for additive layers; and give a single-layer higher-order construction (TM-k) and a stacking rule preserving all three properties. We instantiate these results as the Clock-Free Event (CFE) layer, linear in the number of events, give a topology-free variant for temporal point processes with closed-form likelihoods, and specify a preregistered clock-robustness protocol with a physical-time threat model for spike-retiming attacks. In preliminary development-set experiments, the TM state beats an otherwise identical additive state by 3.5-9.1 points on DVS128 Gesture and by 24.5 and 30.6 points on SHD and Speech Commands.
Reinforcement learning (RL) has become a dominant paradigm for enhancing LLMs' reasoning capabilities. Yet long-horizon autoregressive rollouts introduce a fundamental limitation: local policy mismatch can compound into severe state distribution shift over trajectories. Existing LLM RL methods, such as GRPO and GSPO, control policy mismatch at the token or sequence level, but overlook this trajectory-induced shift, causing training instability and eventual collapse as the horizon grows. To close the state distribution gap, we propose Trajectory-Aware Policy Optimization (TAPO), a principled framework that theoretically derives the state-distribution correction from the off-policy policy gradient and incorporates it into the RL objective. TAPO further employs a geometric-mean estimator for variance reduction, together with a length-aware dynamic clipping that prevents overly deviated updates across the trajectory. Extensive experiments demonstrate that TAPO significantly surpasses existing LLM RL algorithms across eight competition-level benchmarks, achieving up to 48% and 12% improvement over GRPO on mathematical reasoning and algorithmic coding, with substantially enhanced training stability.
Web systems increasingly predict users' decisions from what users say, yet when interests partly conflict, users may promise to cooperate to win the other side's cooperation and then not do so. Sales or service agents that rely on such promises may grant concessions to users who then do not keep their promises, at a cost to the party they represent. We test how LLMs combine users' statements with decision records to predict users' decisions, using 1,260 rounds in which 42 users played a repeated prisoner's dilemma-style game with five agents, each prompted to express a dominant Big Five trait, and could talk with the agent before every decision. Across eight open-weight LLMs, predictions from decision records separated cooperation from defection well (AUC = .80), and adding dialogue brought no detectable gain. With dialogue, predictions agreed with users' statements more often, and errors concentrated on rounds in which users broke their statements: accuracy on rounds with a statement barely changed (.73 to .74), while it rose on kept statements (.76 to .95) and fell sharply on broken ones (.65 to .17). Predictions weighted statements more heavily than users' behavior warranted: when users stated a decision they had not yet made in that game, the predicted probability of keeping it rose from .13 to .69, whereas users kept .40 of such statements. Users' decisions also varied with the agent they faced, but predictions hardly reflected this: users cooperated more with the Agreeableness agent than with the Extraversion agent (70.6% vs. 54.5%), whereas predicted cooperation was nearly identical (64.5% vs. 63.1%). Users also cooperated more with an agent when they themselves scored higher on that agent's trait, whereas predicted cooperation barely changed with these scores. Dialogue-based predictions should therefore weigh users' statements by their earlier decisions and the agent they face.
Tuo An, Chuhao Zhou, Jiahua Dong, Jingliang Li, Jiaqi Bai, Jindou Jia, Jianfei Yang
Flow-based policies for language-conditioned multi-task manipulation commonly share a single $\textit{Gaussian}$ source prior and a common linear transport path across tasks. However, reliably executing multiple tasks with distinct motion requirements remains challenging for a shared policy. We introduce $\textbf{TOPS}$ ($\textbf{T}$ask-$\textbf{O}$riented $\textbf{P}$riors and $\textbf{S}$chedules), which adapts the flow source prior and interpolation schedule to each task without modifying the network architecture. TOPS first constructs multimodal task sources from reverse-mapped expert action sequences, allowing language instructions to select where generation begins. Guided by source-target covariance mismatch, it then slows the transition toward expert actions early in denoising and accelerates it later along directions requiring expansion. Under a $\textit{Gaussian}$ endpoint model, our analysis shows that this schedule adjustment can lower an upper bound on velocity-prediction risk. Experiments on LIBERO, LIBERO-Plus, and CALVIN demonstrate improved performance across in-distribution, perturbation and cross-environment settings, alongside better data efficiency. Using one integration step, TOPS raises LIBERO-10 success from $53.44\%$ to $69.67\%$ compared with ten-step Standard Flow Matching (FM), while reducing inference time to approximately one-sixth of the baseline. TOPS also yields consistent gains when transferred to the large-scale flow-based $π_{0.5}$ policy across all four evaluated settings. Real-world experiments further demonstrate improved task success, including an increase from $20\%$ with standard FM to $80\%$ with TOPS on the upper-drawer opening task.
Lean is increasingly used to verify mathematical proofs and software produced by LLM agents, but writing formal proofs remains hard. Existing prover systems take a divide-and-conquer approach that recursively decomposes hard proof goals into easier lemmas and splits again any lemma that resists proof. However, a lemma can resist proof for different reasons, including insufficient guidance, a false statement, or a flaw in the decomposition that produced it. Existing systems cannot tell these apart as a failed attempt reports little beyond its failure, and responses are confined to the failing node without considering the rest of the proof or revising the plan above it. The system may thus keep decomposing obligations along a flawed proof direction until the budget runs out, causing the whole proof attempt to fail. We introduce Maestro, a Lean prover harness that handles failures through global coordination. Provers report why they failed, not just that they failed, and a coordinator that sees the whole proof chooses an appropriate response, such as guiding a stuck prover, correcting a false lemma, or retracting a flawed subtree and replanning it. The same harness extends to verified code synthesis, where a checked counterexample sends the code back for repair. We evaluate Maestro with four frontier LLMs on seven mathematics and three software verification benchmarks. It proves more targets than each model's native coding agent (Claude Code or Codex) on every benchmark the agent has not already saturated, with the largest gains on research-level mathematics. With Claude Opus 5, it discovers its own proofs of 34 of the 48 theorems in the June 2026 release of arXivLean, against 19 for the native agent. It also outperforms all published prover systems on every benchmark they share, solving 669/672 PutnamBench, 59/60 Lean-IMO-Bench, and 6/6 IMO 2026.
Edge-cloud synergies provide a promising paradigm for privacy-preserving deployment of foundation models in heterogeneous web environments, enabling lightweight edge models to support domain-specific multimodal we b services while cloud-hosted models facilitate collaborative knowledge sharing. However, in practical edge-cloud web environments, collaborative multimodal learning is challenged by modality heterogeneity (different modality combinations across domains) and model-structure heterogeneity (different modality-specific encoders/fusion modules). To address these issues, we propose ML-ECS, a collaborative multimodal learning framework that enables joint training between a server-based model and heterogeneous edge models. This framework consists of four components: (1) cross-modal contrastive learning (CCL) to align modality representations in a shared latent space, (2) adaptive multimodal tuning (AMT) to preserve domain-specific knowledge from local datasets, (3) modality-aware aggregation (MAA) to robustly aggregate while mitigating noise caused by missing modalities, and (4) SLM-enhanced CCL (SE-CCL) to facilitate bidirectional knowledge transfer between cloud and edge. Experimental results on various multimodal tasks show that ML-ECS consistently outperform state-of-the-art baselines under varying modality availability, achieving improvements of 1.55% to 16.75% in Rouge-LSum and improving both client- and server-side performance. In addition, by communicating only low-rank LoRA parameters and fused representations, ML-ECS achieves high communication efficiency, requiring only 0.71% of the total parameter volume.
Monocular outdoor navigation requires robust camera motion estimation and a dense, navigable map, where 3D Gaussian Splatting SLAM serves this task through an explicit differentiable parameterization and closed-form projection. However, typically, the geometry is recovered by associating 2D pixels with 3D points, so a modular SLAM system absorbs tracking error into the map, leading to cascading degradation. To address these gaps, we present SG^2SLAM, which couples streaming spatial geometry with a ray-conditioned 3DGS kernel. Streaming spatial geometry replaces the per- frame cross-dimensional correspondence with a chunk-level metric prior that supervises tracking and mapping from one source. A ray-conditioned kernel extends each primitive with a viewing-direction subspace that senses the monocular camera viewpoints. The joint optimization over spatial geometry and poses ensures a globally consistent map and trajectory outcome. Experiments on the KITTI and Waymo Open datasets show that our SG^2SLAM surpasses the per-frame association baselines in the monocular outdoor navigation task. Code will be made publicly available upon acceptance.
Mingkang Liu, Zhanghao Zhouyin, Guanghui Wang, Daye Zheng, Lei Shen, Qiangqiang Gu
Density functional theory (DFT) gives the electronic structure of materials, but its cost limits large-supercell calculations and materials screening. Machine-learned Hamiltonians bypass the self-consistent cycle but are usually system-specific; universal ones could cover new compositions and disordered solids. Such models must generalize across chemistry and keep downstream properties physical, while staying inexpensive for large cells. Here we introduce UniTB, a universal E(3) equivariant Hamiltonian representation framework that integrates physical priors rooted in density-functional tight-binding theory with deep learning. UniTB's network, which corrects a prior Hamiltonian, generalizes across chemistry through a mixture of experts routed by that prior and a charge-equilibration head as in self-consistent-charge tight binding. A projection bounds band shifts from the prior, suppressing spurious states and keeping downstream properties physical. We provide the fastest SO(2) tensor-product GPU kernels for the network. The variant UniTB-Flash assembles its prior from two-center tables at $O(N)$ cost for $N$ atoms, against $O(N\log N)$ on a grid. Trained on 29,242 crystals, UniTB reaches a 0.48 meV matrix-element error on nonmagnetic test crystals spanning 68 elements. Without amorphous training data, both models agree with DFT optics and conductivity of amorphous Ge-Sb-Te. UniTB-Flash ranks candidates with no significant difference from UniTB, whose property values are more accurate; together they screen 16,195 crystals for five application domains of industrial relevance.
Xiucheng Wang, Nan Cheng, Zhisheng Yin, Conghao Zhou, Ruijin Sun
Dense low-Earth-orbit (LEO) networks increasingly reuse spectrum across moving satellites and overlapping beams, making a spatial radio map essential for interference diagnosis and coordination. Constructing such a map from sparse ground power measurements is difficult because the active satellite subset and its cardinality are unknown, candidate footprints can be highly coherent, and beam pointing, width, effective power, Doppler, and background must be inferred jointly. Known ephemerides do more than enumerate visible candidates. Propagated positions determine candidate-specific ranges, viewing geometry, and spatial footprints, while radial velocities induce distinct time-frequency signatures. Their joint spatial--Doppler structure rejects candidates whose geometry or time--frequency signature conflicts with the observations before continuous beam inference. SateBeam therefore screens all candidates through Doppler-resolved nonnegative fitting. GPU-accelerated grid correlations initialize continuous off-grid refinement; a Gamma-likelihood Bayesian information criterion (BIC) selects model cardinality, and backward deletion and exchange repair greedy support errors. Across paired real-ephemeris scenarios, SateBeam recovers the exact support in every case and reduces independent-query RMSE from $0.655$ to $0.050$ dB relative to orthogonal matching pursuit, with consistent advantages under model mismatch and larger candidate sets. The code is available at https://github.com/UNIC-Lab/SateBeam.
Zhichao Shi UCAS, Cehao Yang The Hong Kong University of Science and Technology, Hao Zhou, Xiaojun Wu The Hong Kong University of Science and Technology and IDEA Institution, Huajie Li International Digital Economy Academy, International Digital Economy Academy, Xuhui Jiang International Digital Economy Academy, International Digital Economy Academy, Chengjin Xu International Digital Economy Academy, Yuanzhuo Wang Chinese Academy of Sciences 等 9 位作者
Synthetic data has emerged as a crucial solution to the data scarcity bottleneck in large language models (LLMs), particularly for specialized domains and low-resource languages. However, the broader adoption of existing synthetic data tools is severely hindered by convoluted workflows, fragmented data standards, and limited scalability across modalities.To address these limitations, we develop DataArc-SynData-Toolkit, an open-source framework featuring: (1) a configuration-driven, end-to-end pipeline equipped with an intuitive visual interface and simplified CLI for exceptional usability; (2) a unified, quality-controllable synthesis paradigm that standardizes multi-source data generation to ensure high reusability; and (3) a highly modular architecture designed for seamless multimodal, multilingual, and multi-task adaptation.We apply the toolkit in multiple application scenarios. Experimental results demonstrate that our toolkit achieves an optimal balance between generation efficiency and data quality. By offering an end-to-end and visually interactive pipeline, DataArc-SynData-Toolkit significantly lowers the technical barrier to synthetic data generation and subsequent model training, accelerating its practical deployment in real-world applications.
尚无 AI 解读，展示原文摘要
cs.CL · cs.CV
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) · 2026 · 正式发表
Xiangxu Zhang Renmin University of China, Lei Li, Yanyun Zhou Renmin University of China, Xiao Zhou Renmin University of China, Yingying Zhang Tencent, Xian Wu Tencent
Medical diagnostics is a high-stakes and complex domain that is critical to patient care. However, current evaluations of large language models (LLMs) remain limited in capturing key challenges of clinical diagnostic scenarios. Most rely on benchmarks derived from public exams, raising contamination bias that can inflate performance, and they overlook the confounded nature of real consultations beyond textbook cases. Recent dynamic evaluations offer a promising alternative, but often remain insufficient for diagnosis-oriented benchmarking, with limited coverage of clinically grounded confounders and trustworthiness beyond accuracy. To address these gaps, we propose DyReMe, a dynamic benchmark for medical diagnostics that provides a controlled and scalable stress test of diagnostic robustness. Unlike static exam-style questions, DyReMe generates fresh, consultation-style cases that incorporate clinically grounded confounders, such as differential diagnoses and common misdiagnosis factors. It also varies expression styles to capture heterogeneous patient-style descriptions. Beyond accuracy, DyReMe evaluates LLMs on three additional clinically relevant dimensions: veracity, helpfulness, and consistency. Our experiments show that this dynamic approach yields more challenging assessments and exposes substantial weaknesses of stateof-the-art LLMs under clinically confounded diagnostic settings. These findings highlight the urgent need for evaluation frameworks that better assess trustworthy medical diagnostics under clinically grounded confounders.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Expanding Large Language Models(LLMs) to new languages is a costly endeavor, demanding extensive Continued Pre-Training(CPT) and data-intensive alignment. While recent data-free merging techniques attempt to bypass alignment by fusing a multilingual CPT-enhanced model with its instruct counterpart, they are plagued by a critical trade-off: mitigating parameter conflicts to preserve original abilities inevitably dilutes new language acquisition, and vice-versa. To resolve this conflict, we introduce , which upcycles a dense model into a Mixture-of-Experts(MoE) architecture, allocating different experts to different languages. Alignment ability is then transferred by grafting a MoE-expanded parameter delta(\Delta_{\text{instruct}}) to the CPT-enhanced base model, bypassing the complex alignment phase. Experiments demonstrate ’s superiority even against baselines with similar FLOPs or number of parameters; it improves performance on expanded languages while effectively preserving original capabilities. We further show our approach is highly applicable across different models and Post-training deltas.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Shufan Yang, Zifeng Cheng Nanjing University, Zhiwei Jiang Nanjing University, Qingfeng Qi, Yafeng Yin Nanjing University, Cong Wang Nanjing University, Ao Zhou, Qing Gu Nanjing University
Extracting embeddings directly from Mixture-of-Experts (MoE) models is a promising yet underexplored direction that requires no additional data or fine-tuning. While previous studies have utilized semantic compression prompts or expert routing information to improve sentence embeddings, they typically allocate a fixed number of experts uniformly across all layers and tokens, ignoring inter-layer and inter-token heterogeneity. In this work, we identify two key observations in MoE models: (1) layer-wise variations in expert homogeneity, suggesting that different layers require different expert budgets, and (2) token-wise contribution imbalance, indicating that different tokens should also be allocated different numbers of experts. To address these issues, we propose an Adaptive Expert Allocation (AEA) framework that dynamically performs both layer-wise and token-wise expert allocation to enhance embedding quality. Specifically, AEA allocates fewer experts to layers with higher homogeneity and to tokens with lower attention importance, where layer-wise homogeneity is determined by the similarity among embeddings produced by the experts in each layer. Notably, our method is plug-and-play, seamlessly integrates with existing prompt engineering methods, and introduces no additional time overhead. Experiments on the STS tasks demonstrate that AEA consistently improves embedding quality across multiple MoE models.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Large Language Model (LLM) based Chinese Grammatical Error Correction (CGEC) systems face two critical challenges: general-purpose models lack specialized linguistic priors for subtle grammatical distinctions, and Supervised Fine-Tuning (SFT) with Maximum Likelihood Estimation fails to optimize for precision-focused metrics, leading to systematic over-correction. We propose CSRP, a three-stage framework that progressively builds correction capability through Continual Pre-training (CPT) on 5.9M balanced samples to internalize domain knowledge, Chain-of-Thought SFT with explicit error reasoning for diagnostic transparency, and Group Relative Policy Optimization with a novel Efficiency-Aware Reward that explicitly penalizes unnecessary edits. On the NACGEC benchmark, CSRP achieves state-of-the-art performance with 50.99 F_{0.5} and 57.17 precision, substantially outperforming previous best results while effectively mitigating the over-correction bias inherent in MLE-trained models. Our method also advances CSCD spelling correction to 59.61 F1, surpassing GPT-4 by 5.20 points. Comprehensive ablation studies demonstrate that the RL alignment stage contributes a 8% relative gain over the SFT baseline, and that this gain is orthogonal to the contribution of large-scale CPT, validating that explicit optimization for edit efficiency is essential for high-quality grammatical error correction. Our code is available at https://github.com/TW-NLP/ChineseErrorCorrector.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表