Decompose, Estimate, Select: Coverage-Aware Context Construction for Long-Form RAG
He Tu, Aleena Bosco Kurumthottam, Torsten Suel
AI 速览：提出覆盖感知检索流程，通过分解查询、估计覆盖矩阵和贪心选择提升长文档RAG的证据覆盖，并引入自反馈循环改善生成器对证据的利用。
Retrieval-augmented generation (RAG) grounds a language model's output in retrieved evidence, but for complex questions spanning multiple aspects of an information need, relevance-based retrieval may return redundant evidence while leaving some aspects unsupported. We propose a coverage-aware retrieval pipeline that decomposes queries into subquestions, ranks candidates with a coverage-trained bi-encoder, predicts a document\,$\times$\,subquestion coverage matrix with a cross-encoder, and selects documents by greedy maximum-marginal coverage. On TREC RAG 2024, our approach improves $α$-nDCG@10, which rewards newly covered nuggets and discounts redundant evidence, from $0.625$ to $0.702$, approaching the $0.718$ gold-nugget upper bound without sacrificing relevance. We then find that ground-truth oracle retrieval, despite substantially higher retrieval coverage, delivers no more nugget coverage to the answer than our estimated selection. This suggests that answer completeness depends not only on what evidence is retrieved, but also on how effectively the generator uses it. To address this, we introduce a \emph{grounded self-feedback loop} in which the model identifies underused evidence from its first answer and generates again without gold supervision, raising $V_{\mathrm{strict}}$, the fraction of vital reference nuggets fully supported by the answer, by $0.086$ over two passes while largely preserving faithfulness.
Reinforcement Learning with Verifiable Rewards has shown great promise in enhancing mathematical reasoning in large language models. However, existing approaches provide limited intermediate supervision for complex, multi-step reasoning tasks. To bridge this gap, we propose \textbf{Hierarchical Hint-based Reward}, a framework that leverages the intrinsic sub-question structure of mathematical problems as structured intermediate conditioning signals during RLVR training. Unlike chain-of-thought prompting, these hierarchical hints serve as logically grounded intermediate milestones that effectively reduce the reasoning horizon toward the final answer. To validate this framework, we introduce \textbf{HiMath}, a benchmark of hierarchically structured mathematical problems. Across diverse LLMs, hierarchical hints consistently improve reasoning performance even without task-specific training. When integrated into RL training via H$^2$R, they boost Qwen2.5-Math-7B accuracy on HiMath from 30.4\% to 39.1\% using hierarchical hints, and up to 45.6\% with Oracle Hierarchical Hints. This pronounced improvement demonstrates that incorporating reliable intermediate checkpoints into reward design is essential for sustaining complex mathematical reasoning.
We study \textsc{Max $k$-Weight CSP}: maximize the satisfied weight of Boolean constraints with nonnegative rational weights by setting at most $k$ variables to $1$. Each constraint uses at most $d$ distinct variables and a relation from a finite Boolean language $Γ$. This includes \textsc{Partial Vertex Cover}, \textsc{Densest $k$-Subgraph}, bounded-frequency \textsc{Maximum Coverage}, and cardinality-constrained \textsc{Max-SAT}. We completely classify the fixed finite languages admitting an efficient parameterized approximation scheme (EPAS) with parameter $k+d$. Only nonempty relations excluding the all-zero tuple matter. For such a relation, a coordinate is a \emph{unit witness} if the tuple with a single $1$ in that coordinate is accepted, and a \emph{control} otherwise. A relation is \emph{Direct} if setting any selected control to $0$, with all other coordinates unchanged, preserves satisfaction. It is \emph{Rooted} if every satisfying tuple has a selected unit witness such that keeping it, setting all other unit witnesses to $0$, and setting any subset of selected controls to $0$ preserves satisfaction. If all such relations are Direct, or all are Rooted, we give a deterministic $(1-\varepsilon)$-approximation for every rational $\varepsilon\in(0,1)$ in time $2^{O(d+k\log(1+dk/\varepsilon))}\ell_{\mathrm{in}}^{O(1)}$, where $\ell_{\mathrm{in}}$ is the total input length. The algorithms are uniform in $d$ and give an EPAS parameterized by $k+d$ whenever relation membership is polynomial-time decidable. If neither condition holds, no EPAS exists unless $\mathsf{FPT}=\mathsf{W[1]}$, even for unweighted instances where no two constraints involve the same set of variables.
We show that established performance metrics in binary classification, such as Matthews' correlation coefficient (MCC), Cohen's $κ$, the F-score or the Jaccard similarity coefficient are not robust to class imbalance in the sense that if the proportion of the minority class tends to $0$, the true positive rate (TPR) of the Bayes classifier under these metrics {becomes very small or even tends to $0$}. Thus, in imbalanced classification problems, these metrics favour classifiers which ignore the minority class. To alleviate this issue we introduce robustified modifications of the MCC, of Cohen's $κ$ and of the F-score with an additional tuning parameter which allows to adapt the amount of robustness against class imbalance. As theoretical guarantee we show that the Bayes-optimal classifier for these robustified performance metrics, when expressed in terms of the density ratio $f_1/f_0$ of the class-conditional densities $f_i$, has a threshold parameter which is upper-bounded in terms of the tuning parameters. Therefore, even in strongly imbalanced settings, the TPR associated to this classifier will be bounded away from $0${, provided that $f_1/f_0$ exceeds this bound with positive probability under the positive class}. We numerically illustrate the behaviour of the various performance metrics and the effect of the tuning parameters in simulations as well as on a credit default data set. We also discuss connections to the receiver operating characteristic and precision-recall curves, which provide an alternative perspective on the proposed notion of robustness, and give recommendations on how to combine their usage with performance metrics.
Nonlinear structural system identification remains challenging when governing physics is only partially known and measurements are sparse. Deep Bayesian filtering addresses these challenges by combining learnable dynamical representations with probabilistic state-space models, physical inductive biases, and uncertainty-aware inference. This paper presents, for the first time, a unified tutorial treatment of deep Bayesian filtering for nonlinear structural system identification, connecting model formulation, learning, posterior inference, and evaluation. Amortized variational inference and differentiable Kalman filtering are introduced as complementary strategies within this broader framework. The Deep Markov Model (DMM) provides a data-driven reference, while the Physics-guided Deep Markov Model (PgDMM) and Neural Extended Kalman Filter (Neural EKF) serve as illustrative instances through which methodological choices and their implications are examined. Dynamic state estimation and joint state and parameter estimation are demonstrated on a three-dimensional shear-frame benchmark with hysteretic Bouc-Wen connections and sparse acceleration measurements. The comparative analysis yields new observations concerning the relationships among reconstruction accuracy, predictive fidelity, physical interpretability, and estimation robustness. Physical structure improves latent-state interpretability without necessarily ensuring faithful predictive dynamics, while comparable typical reconstruction accuracy can conceal substantial differences in estimation robustness. These findings motivate practical guidelines for model construction, training, and evaluation that distinguish reconstruction, forecasting, state recovery, and parameter identification. The tutorial concludes by outlining challenges in uncertainty calibration, identifiability, and validation on real structures.
Nizar Khalfet, Francisco A. Monteiro, Ioannis Krikidis
This paper investigates a quantum relay-assisted free-space optical (FSO) communication system in which a decode-and-forward relay is powered by time-switching optical energy harvesting and forwards the decoded information using displaced squeezed signaling (DSS). To obtain a physically con sistent description of turbulence-induced loss for non-classical signaling, each hop is modeled as a pure-loss bosonic chan nel whose random transmissivity follows a truncated Gamma Gamma law, i.e., the Gamma-Gamma irradiance conditioned on the physically admissible range of a transmittance. Under this channel, coherent states remain pure, whereas squeezed states are mapped to mixed Gaussian states. Under Helstrom-optimal detection, the average error of the coherent hop is derived in exact Laplace-series form. For the DSS hop, we derive the output states, a closed-form Gaussian fidelity with rigorous two sided bounds on the Helstrom error, and prove the existence of a loss-aware optimal photon partition. Furthermore, we prove that optimized squeezing preserves both the turbulence-induced diversity order and the leading asymptotic constant of coherent relaying, so that its benefit is a finite-photon-budget effect, and that an interior throughput-optimal harvesting fraction exists. A covariance-preserving benchmark with an exact series representation yields an analytical upper bound on the achievable throughput whose absolute gap vanishes as the relay photon budget grows. Numerical results based on the exact mixed-state Helstrom error quantify the benefit of squeezing and the regime in which it is useful.
Abhishek Malvankar, Paul Castro, Nick Mitchell, Lionel Villard
Speculative decoding speeds up large language model (LLM) serving only while the GPU has spare compute, so the best draft length, the number of tokens a small drafter guesses per step, shrinks as the batch grows. vLLM lets operators set the draft length for each batch size with a hand-written table, but nothing measures or corrects that table. We present TableTune. Its tuner times the decoding step for every batch size and draft length on vLLM's own serving path and computes, in minutes, the table that maximizes throughput. Its controller, a vLLM plugin, tracks how many guesses are accepted in live traffic and rebuilds the table while serving. We measure TableTune only against other ways to speculate. On an H100, the tuned table delivers at least 94% of the throughput of the best fixed draft length at every load in four of five drafter settings (88% in the fifth) without the operator having to know which length that is. Against reimplementations of TurboSpec- and Nightjar-style control on the same engine, its worst-case loss is never higher than theirs, and with the EAGLE-3 drafter it is 3% against 9% and 12%. On a mixture-of-experts model, vLLM's documented table loses up to 18% and the tuned table at most 3%. Starting from mistuned tables, the controller raises throughput by up to 58% (up to 45% on vLLM's default runner), and after a workload switch with the DFlash drafter it recovers 96% of a stale table's loss. In llm-d's optimized-baseline deployment as shipped, two Qwen3-32B replicas on four GPUs, the tuned table is up to 32% faster than a fixed draft length of 3 and up to 25% faster than vLLM's documented table at high load, and the controller raises the documented table's throughput by up to 29%.
First-derivative edge detectors such as Sobel, Prewitt, Roberts and Scharr measure absolute grey-level differences. We revisit the log-gradient magnitude, a relative-gradient edge operator, and motivate it by analogy with the superderivative, a multiplicative derivative in which the independent variable is perturbed multiplicatively rather than additively. In one and two dimensions the logarithm of the superderivative is a radial logarithmic derivative; its local, origin-free, isotropic analogue is the magnitude of the logarithmic gradient. The resulting operator measures relative rather than absolute change, giving an edge-strength map that is invariant under global intensity rescaling. We present the theory, show how the operator can be implemented in two steps---a classical gradient kernel plus a pointwise division---and compare it with standard first-derivative filters on synthetic images with ground-truth edges and on the official BSDS500 benchmark. We also study a parametric family that interpolates between the absolute gradient and the log-gradient; empirically, the best members are small intermediate exponents, which edge the classical kernels by a small but statistically detectable margin in the per-image metric and, with the same Gaussian smoothing as the single-$σ$ Canny, match it in ODS and lead it in OIS and AP, while the tuned Canny remains ahead in ODS. The single-threshold margin over the kernels does not replicate on the held-out test split, and the fully relative endpoint is never the best member, its ranking depending on the edge-map normalisation. The contribution is the construction and the invariance it buys, not a new state of the art.
Zongyuan Sun, Shuai Zhao, Yingjun Du, Ke Shu, Ziyue Zhu, Haoran Luo
Large language models (LLMs) can answer many factual questions correctly, but a correct answer is not necessarily a supported one. For an LLM agent, a knowledge graph (KG) plays two roles: as a source of answers, it supplies facts that the model lacks; as a source of evidence, it supplies the supporting triples that make an answer traceable and auditable. However, existing KG agents treat the KG only as a source of answers: their training data carries answers but not evidence, and their training rewards check the answer but not the retrieved evidence, so agents often answer correctly without retrieving the supporting triples. We propose FoG, a framework that trains KG agents to retrieve the evidence behind their answers. FoG consists of evidence-grounded synthesis, a KG interaction agent, and an evidence-coverage process reward (ECPR): the synthesis recursively hides explicit entities behind verifiable sub-questions, so that every question carries its supporting triples by construction; the agent records the triples returned by the KG; and ECPR credits each supporting triple to the turn that first retrieves it. Experiments on three KGQA benchmarks demonstrate that FoG enables 3B--4B open-source agents to outperform substantially larger baselines in answer accuracy and evidence coverage, with zero-shot cross-KG transfer.
Jiawei Zhou, Bo Peng, Yuxuan Huang, Ming Zhang, Youbang Sun, Xiang Wang, Chaochao Lu
Skill optimizers adapt an agent to long-horizon tasks, such as repairing a terminal environment, by rewriting its skill, a natural-language procedure in its context, and keeping the edits that raise success on a small dev set. On such tasks success is noisy, so a gain on a small dev set does not show that an edit helps, and a success score does not show what the edit costs. We introduce BiPA, which replays an optimizer's acceptance rule on candidates identical to its seed to measure what selection alone reports, and records each episode as a bipartite graph of turns and the text they read to trace extra input tokens to the behavior behind them. Across three optimizers, three agentic benchmarks and two models, the learned skills succeed about as often as their hand-written seed, 0.1 points higher on average. Their dev gains vanish on test: keeping the best of several candidates on at most 30 tasks reports gains this large even when none changed. They read about 16% more input tokens: the optimizers almost always add rules, half to re-check work and some to act one call at a time, and the agent follows them as extra turns, each re-sending the whole conversation. Four principles follow: accept an edit only if its gain exceeds what selection alone reports; charge it for the turns it adds, not for its length; let the editor delete and merge rules; show it what earlier rules cost. Code is available at https://github.com/valleysprings/BiPA.git.
XinYu Piao, Myunghoon Ryu, Joongheon Kim, Jong-Kook Kim
A large language model maintains conversational content in an external storage that is outside the model and grows as the dialogue continues. This work assumes an environment where a language model has no external memory and no prior knowledge of the future question or its answer. The motivation is to research whether a dialogue can be stored in the model parameters and later answer a question about it. This work proposes self-keyed in-parameter memorization (SKIM), a closed-form method in which the model generates its own questions about each dialogue turn and writes those questions into its parameters as the keys that later retrieve the turn's content. Generating many varied questions for each turn allows a later user question to match one even when phrased differently, and its answer is recalled from the parameters alone. Evaluated on five language models, SKIM recalls most stored dialogue turns from later user questions, and it answers from the model parameters without any external storage.
On the theoretic analyses on transforms, they mainly focus on the computability analysis, such the Turing-completeness of unbounded precision transformers, the advantage of parallelizability of unbounded precision transformers and the limitations of bounded precision transformers. This paper give transformers an algebra for reasoning, which relies heavily on the functional decomposition results of transformers. This algebra uses functional decompositions as atomic actions, includes sequential composition and parallel composition, recursion for feedback and abstraction for hidden layers. As we know, it is the first attempt to model transformers logically and can be used to verify the correctness of transformer based applications.
Yesung Cho, Ji Hwan Park, Chan Il Kim, Hyewon Kim, Honglan Li, Yumin Lee, Geongyu Lee, Sujeong Hong 等 31 位作者
Outcome informed AI can localize prognostically relevant regions in digital pathology, but directly characterizing the molecular basis of these spatial predictions remains challenging. Here, we developed an outcome informed, AI-guided deep visual proteomics framework that uses recurrence-risk heatmaps to guide physical tissue sampling and mass spectrometry-based spatial proteomics in triple-negative breast cancer (TNBC). In 156 patients, high- and low-risk regions coexisted within the tumor compartment and showed distinct nuclear and architectural features, indicating intratumoral variation beyond broad tissue identity. Guided by these maps, we physically isolated and profiled 46 AI-defined high- and low-risk tumor regions from two recurrent cases. High-risk regions were enriched for mitotic programs, whereas low-risk regions showed immune- and antigen-presentation-associated programs. We used proteins that consistently differed between high- and low-risk regions to construct a 13-protein composite. The composite showed a significant effect to discriminate highly recurrent patients from less recurrent patients in an expanded TNBC cohort, while the corresponding transcript-based score showed a similar trend in METABRIC. Integrating the protein composite with the H&E-derived risk score improved out-of-bag discrimination, increasing the C-index from 0.679 to 0.739 and the 5-year time-dependent AUC from 0.675 to 0.770. These findings establish outcome-informed H&E risk maps as spatial guides for molecular discovery, extending prognostic AI from prediction toward direct biological interrogation of recurrence-associated tissue states.
The fibrational framework of Licata, Shulman, and Riley gives a uniform account of substructural and modal logics for single-conclusion judgments. This thesis extends it to a multiple-output setting in which cartesian resources may be used freely, while linear resources occur exactly once on either side of the turnstile. Judgments are indexed by descriptors: directed acyclic graphs generalizing the tree-shaped expressions of the original framework. Descriptor $2$-cells encode structural transformations, and explicit gates provide weakening, contraction, coweakening, and cocontraction at linear modes. We define a sequent calculus generated by two connectives, $F$ and $U$. We prove identity, invertibility, and cut admissibility. The mode and type theories present linear-non-linear $2$-polycategories, and the projection is a local discrete fibration and a fibration, with opcartesian lifts for one-output base morphisms. Two adequacy results validate the framework: its cartesian fragment recovers the Licata--Shulman--Riley calculus, and its purely linear fragment presents classical multiplicative linear logic. Finally, we relate indexed and fibrational semantics. After developing representability correspondences for multicategories and polycategories, we establish an equivalence between multi-indexed categories over a fixed representable base and multicategorical opfibrations over it, together with its biclosed bifibrational version. We also give a one-way construction from planar poly-indexed data to a planar polycategory.
Yao Huang, Zhiyuan Zeng, Yu Wang, Zhitong Zhu, Dingcheng Duan, Hao Li, Shangzhi Xu, Yifan Wang 等 17 位作者
As frontier language models grow increasingly capable and autonomous, independent third-party evaluations are critical for characterizing their systemic safety risks. While closed-source releases are routinely accompanied by comprehensive system cards, open-weight models receive less scrutiny regarding frontier risks. To address this gap, we present an independent safety evaluation of DeepSeek V4.1 Flash-1017, the latest version of V4.1 Flash that departs from V4 with a new architecture, scaled-up training, and strong agentic capabilities. Drawing on the EU General-Purpose AI Code of Practice, we evaluate the model across 34 benchmarks, including 5 in-house suites, spanning evaluation integrity, loss of control, and harmful manipulation. Conforming to the AEF-1 standard, we compare V4.1 Flash-1017 against V4.1 Flash, earlier DeepSeek V4 models, and leading closed-source systems (GPT-6 Astra, GPT-5.6 Sol, and Gemini 3.8 Flash). Overall, V4.1 Flash-1017 shows substantial safety improvements over the V4 generation and approaches or matches leading closed-source baselines on several dimensions, while still exhibiting notable behaviors that warrant continued attention. Building on our findings, we highlight three key observations and recommendations. First, because safety can shift and even regress across minor updates, developers should integrate safety evaluation into incremental releases. Second, as final-turn outputs and end-state checks alone cannot reliably detect loss-of-control behaviors, deployments require process-level observability across both reasoning traces and action logs. Third, because public benchmarks elicit strong evaluation awareness and face data contamination risks, developing private benchmarks is essential for trustworthy frontier evaluation.
Streaming 3D perception is well suited to robotics and augmented reality, where long visual streams must be processed efficiently and consistently. Recent recurrent models offer a promising solution by maintaining fixed-size states and enabling linear-time inference, but they often suffer from temporal forgetting over long sequences due to the limited capacity of compressed memories. We propose Mem3R, a streaming 3D reconstruction model with a hybrid memory design that decouples camera tracking from geometric mapping to improve temporal consistency over long sequences. For camera tracking, Mem3R employs an implicit fast-weight memory implemented as a lightweight Multi-Layer Perceptron updated via Test-Time Training. For geometric mapping, Mem3R maintains an explicit token-based fixed-size state. Compared with CUT3R, this design not only significantly improves long-sequence performance but also reduces the model size from 793M to 644M parameters. Mem3R supports existing improved plug-and-play state update strategies developed for CUT3R. Specifically, Mem3R combined with TTT3R reduces absolute trajectory error by up to 39% relative to CUT3R combined with TTT3R on 500- to 1000-frame sequences from the TUM Dynamics dataset. The resulting improvements also extend to other downstream tasks, including video depth estimation and 3D reconstruction. Project page: https://lck666666.github.io/Mem3R/.
We measure the emotion geometry of six LLMs spanning Google's Gemma 2 family. Using an extraction pipeline established by Sofroniew et al. (2026) on Claude Sonnet 4.5 and replicated by Hollowell (2026) on Gemma 2 27B base, we compare emotion geometry across base and instruction-tuned variants of Gemma 2 2B, 9B, and 27B. First, all six model cells exhibit the core emotion geometry. A linear map fit between two cells places held-out emotions nearest their own vectors in the target model. Across all fifteen model pairs, fit in one direction each, this failed exactly once, and the median rank was 1 in every pair. Linear maps refit on cleaned embedding-table vectors recover every held-out emotion at rank 1 in all thirty ordered pairs. Second, scale spreads the geometry's variance across more directions without weakening its tie to valence. The leading component's share of variance falls from 31.6 percent (2B) to 29.8 (9B) to 26.7 (27B) in the base models and from 31.9 (2B) to 29.2 (9B) to 27.3 (27B) in the tuned ones, while its correlation with human valence ratings stays between .71 and .78 in all six models. Third, we project each emotion vector onto the output vocabulary and ask at what depth the vector's own emotion word enters the top 10,000 of the 256,000 tokens. In the 2B and 9B models the median emotion word is there from layer 0, while in the 27B models it begins near chance, stays outside the top 10,000 for half the network, and enters at layer 23 in one step. Fourth, tuning leaves every registered analysis-layer measure within its floor. At every size, valence correlation rises and arousal correlation falls by amounts within their floors, the latter against prediction. PC2's variance share also falls, with a pooled decrease significant after correction. Mid-late layer stability exceeds its floor and traces to base 27B's final layer. Code, data, and audit reports are released.
Quadrotor controller comparisons commonly report a single tracking error, although the resulting ranking also depends on tuning effort, reference information, actuator prediction, optimizer fallback, and command timing. We quantify these effects for cascade PID, hover LQR, boundary-layer sliding-mode control (BL-SMC), linear MPC with LQR fallback (LMPC-F), a preview-enabled LMPC-F, and a geometric $SO(3)$ controller on a common nonlinear plant with rotor gyroscopic coupling, motor lag, and saturation. A primary study (556 episodes at 50~Hz) combines matched twelve-candidate tuning, 400 paired evaluation episodes, and a 48-cell MPC factorial; a confirmation study (1462 episodes) repeats the evaluation on 50 fresh realizations with tuning budgets of 12, 24, and 48 candidates. A matched twelve-candidate tuning protocol changed the mean-RMSE ordering obtained with the fixed baseline settings, moving BL-SMC from last to first; the new ordering recurred on the fresh realizations, and a 48-candidate budget left it unchanged. Reference preview substantially reduced MPC tracking error and, on the figure-eight task, removed fallback activity. On an abrupt step, the outcome depended on the interplay between fallback policy, predictor, and actuator authority. Command timing also changed the outcome: replaying measured command-call delays terminated three of four nominal MPC configurations on a desktop computer and all four on a Raspberry Pi~5, whereas capping the solver at 100 iterations removed every deadline miss in the tested nominal timing profiles. Controller rankings therefore depend not only on tracking accuracy but also on the tuning protocol, reference information, fallback behavior, and implementation timing. We propose a reporting checklist covering tracking accuracy, tuning effort, completion and failure rates, fallback activity, applied-input variation, and command timing.
Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale Policy Optimization (MPO), combining Multiscale Advantage Estimation (MAE) and turn-level importance sampling to align credit assignment and policy updates at the turn scale. Extensive experiments across search, tool-use, and automation tasks with 4B and 30B-A3B models demonstrate consistent gains over strong RL baselines.
Emanuele Rimoldi, Riccardo Neumarker, Tomaso Poggio, Daniel Mitropolsky
Can a model learn by trying to fool another model? We show that it can. Inspired by the Turing Test and its recent generalization to arbitrary distinguishers, we introduce a domain-based teacher-Turing-Test, in which a stronger teacher model serves as both the target of imitation and the distinguisher, and a weaker student model is trained to make the teacher attribute responses to itself. We use this self-attribution (or fooling) probability as a reinforcement learning reward, without ever providing the student ground-truth answers, nor requiring the teacher to generate responses. We test our approach independently in four domains with verifiable answers: CommonsenseQA (reasoning), GSM8K (mathematics), KodCode (programming), and Spider (SQL), training a Qwen2.5-1.5B student/imitator model using DeepSeek V4 Flash as teacher/distinguisher. Beyond improving at passing the Turing Test, the student improves on independent, held-out evaluations in all domains and runs. To probe why being mistaken for the teacher can improve capability, our analyses reveal that self-attribution is associated with correctness -- substantially more than formatting changes -- and that it rewards partial correctness. Our results demonstrate a new, simple teacher-student training paradigm with a private and/or high-cost teacher. They also suggest that indistinguishability need not merely measure behavior -- it can also be a mechanism for learning.
Latency-sensitive spoken agents must know not only which tool to call, but also when they have heard enough to act. Waiting for the turn to complete delays execution even when the request is already actionable. However, neither existing spoken agents nor spoken dialogue corpora define the earliest point at which a request becomes actionable. We introduce a synthetic spoken-dialogue pipeline that provides earliest-actionable-point supervision, marking when the observed speech grounds a tool and its required arguments. Through controlled task generation, speech synthesis and alignment, and gated annotation, we construct 26,694 dialogues totaling 786 hours across 48 domains, with disjoint training and test argument values. We additionally create an audio rendering of $τ^2$-bench for external evaluation. We train an omni-modal language model on this corpus, enabling causal, chunk-level decisions and partial-transcript generation for downstream call completion. The model achieves tool-state macro-F1 scores of 0.48 and 0.47 on held-out dialogues and the external benchmark, respectively. For correct calls, the estimated mean lead over a turn-end cascade, using the same formatter, is 1.8-2.0 seconds in-domain and 1.1 seconds on $τ^2$-bench. These results demonstrate a data-driven path toward spoken agents that act while listening. We will open-source the pipeline and the training scripts.
Many-query applications such as design optimization, uncertainty quantification and digital twins call for repeated full-field predictions from expensive partial differential equation (PDE) simulations. Grassmannian reduced-order models interpolate parameter-dependent subspaces but commonly vectorize the spatial field, and the ambient dimension then grows with the product of the grid sizes. Tensor-based models preserve the multiway structure but often use shared mode-wise bases or a fixed core across parameters. We propose a Tucker-based surrogate that keeps the spatial directions and time as separate modes. The factor subspaces are interpolated on a product Grassmann manifold, Procrustes alignment places the Tucker cores in one coordinate system, and a truncated singular value decomposition models their variation. Across five PDE benchmarks in two and three spatial dimensions, including public computational fluid dynamics data, the interpolation size is reduced by factors of 13 to 482 and the stored model by factors of 13 to 464 relative to a single-subspace Grassmannian surrogate with the same local interpolation. Cached queries run 3.3 to 8.5 times faster, the mean errors over ten resampled splits differ by at most 0.22 percentage points (0.36 when every model choice is re-selected inside each split), and the prediction-error terms of the two methods differ by less than 0.1 percentage points. The remaining accuracy differences are consistent with the larger representation error of the Tucker format. In a three-dimensional grid study the interpolation-size ratio rises to 2,260 and the query speedup to 11.9 at 64^3.
A spatially aware dictionary-free eigenfunction discovery (SADFED) framework is proposed for identification of low-rank Koopman models from data without prescribing a lifting dictionary, kernel, or neural-network eigenfunction architecture. A reference trajectory is selected and used to determine the Koopman modes by regularized least squares (LS). Then, a transformed temporal basis allows the eigenfunction values at all sampled initial conditions to be obtained by a second regularized LS projection. Consequently, only the real and imaginary parts of the eigenvalues remain as the optimization variables. Interpolation of the identified eigenfunction samples reveals their spatial structure, enabling numerical estimation of their gradients. A joint objective combines trajectory reconstruction error with a normalized Koopman partial differential equation (KPDE) residual, promoting spatial consistency with the KPDE over the sampled region and serving as a physics-informed regularizer. The method is evaluated on a system with analytical Koopman eigenfunctions, the FitzHugh-Nagumo system, the van der Pol oscillator, the Duffing system, and a two-spool turbojet engine. The examples demonstrate recovery of known eigenfunctions, sensitivity to reference trajectory and hyperparameters, limit-cycle harmonics and isochrons, discontinuous indicator eigenfunctions and isostables, symmetry exploitation, and construction of state-dependent lifted input dynamics. For the turbojet example, the identified model is further used for state estimation and design of a gain-scheduled tracking linear quadratic Gaussian controller. The results indicate the applicability of SADFED to Koopman spectral identification and control-oriented modeling of nonlinear dynamical systems.
Jin Zhao, Marta Knežević School of Computer and Communication Sciences, EPFL - EPF Lausanne, Tanja Käser EPFL
Large Language Models (LLMs) are increasingly used in education, yet their default helpfulness often conflicts with pedagogical principles. Prior work evaluates pedagogical quality via answer leakage–the disclosure of complete solutions instead of scaffolding–but typically assumes well-intentioned learners, leaving tutor robustness under student misuse largely unexplored. In this paper, we study scenarios where students behave adversarially and aim to obtain the correct answer from the tutor. We evaluate a broad set of LLM-based tutor models, including different model families, pedagogically aligned models, and a multi-agent design, under a range of adversarial student attacks. We adapt six groups of adversarial and persuasive techniques to the educational setting and use them to probe how likely a tutor is to reveal the final answer. We evaluate answer leakage robustness using different types of in-context adversarial student agents, finding that they often fail to carry out effective attacks. We therefore introduce an adversarial student agent that we fine-tune to jailbreak LLM-based tutors, which we propose as the core of a standardized benchmark for evaluating tutor robustness. Finally, we present simple but effective defense strategies that reduce answer leakage and strengthen the robustness of LLM-based tutors in adversarial scenarios.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Mengyu Xiang, Tinghao Chen, Boxu Han, Qiudan Li, Shu Wu Institute of automation, Chinese academy of sciences, Daniel Dajun Zeng Institute of Automation, Chinese Academy of Sciences
As social media grows, harmful information spreads rapidly across platforms and evolves over time, showing cross-platform and cross-temporal variations. Existing methods rely on fixed model parameters during training, which fail to handle substantial semantic discrepancies, leading to Out-Of-Distribution (OOD) problems. While test-time tuning enables dynamic parameter adjustment, it may lead to excessive adaptation to individual samples. The key challenge is how to adapt to semantic variations during testing while preventing overfitting from continuous tuning. To tackle this issue, this paper proposes RLAT, a reinforcement learning (RL)–guided adaptive tuning method for harmful text detection. First, a tuning joint optimization module is designed to update parameters and adapt to semantic variations during testing. It tunes the model by optimizing consistency loss and applying word-level attention constraints to reduce over-reliance on local words and learn a more robust global representation. Then, to mitigate overfitting caused by continuous tuning, a RL–guided adaptive decision model is introduced to direct the tuning process. It reduces the influence of local samples by selecting data and controlling parameter updates, thereby improving overall test performance. Experimental results show that the RLAT outperforms state-of-the-art baselines in cross-platform and cross-temporal scenarios across multiple public datasets.
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 models (LLMs) have demonstrated better safety performance in high-resource languages than in low-resource languages. We attribute this issue as a mismatch gap between language-agnostic semantic understanding ability and language dominant safety alignment biased toward high-resource languages. Based on above insights, we empirically identify the semantic bottleneck in LLMs: intermediate layers in which the geometry of model representations is governed primarily by shared semantic content rather than language identity. Then, we propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks. Experiments show that LASA substantially improves safety across all languages: average attack success rate (ASR) drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains within 3–4% across Qwen2.5 and Qwen3 Instruct models (7B–32B). Besides, our analysis and method offer a representation-level perspective on LLM safety, suggesting that safety alignment requires anchoring safety understanding not in surface text, but in the model’s language-agnostic semantic space.
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
Zechen Li, Qiannan Zhu Beijing Normal University, Mei Wang Beijing Normal University, Jia Li Beijing Normal University, Hua Huang Beijing Normal University
Equipping Large Language Models (LLMs) with pedagogical tutoring capabilities holds significant promise for education. Existing approaches simulate tutor behaviors or preferences and use them to prompt or fine-tune LLMs for dialogue tutoring. However, such methods often fail to sustain high-quality pedagogical conversations that provide explicit stepwise scaffolding and adapt to learners’ evolving cognitive states. To address this, we propose ScaffoldLM, a planning-guided tutoring framework with an assessment-driven memory for multi-turn math dialogue tutoring. ScaffoldLM first generates a stepwise pedagogical plan from solution steps, which serves as a stable backbone for explicit scaffolding. During tutoring, the tutoring memory is updated by an assessment-driven control loop that infers the learner’s cognitive state, evaluates whether the current step target is met, and adaptively selects tutoring actions. The plan, step-level progress, inferred learner states, and dialogue history are maintained in memory to support coherent multi-turn guidance. Experiments on multi-turn math tutoring benchmarks demonstrate that ScaffoldLM substantially improves pedagogical tutoring quality over strong baselines. Code is publicly available at https://github.com/BNU-ERC-ITEA/ScaffoldLM.
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
Lijun Wu Shanghai AI Laboratory, Wentao Zhang Peking University, Conghui He Shanghai AI Lab
As the development of Large Language Models (LLMs) matures, the focus of the research community is undergoing a critical shift from a purely model-centric to a data-centric paradigm. It is now evident that the quality, diversity, and composition of training data—not merely its scale—are the primary drivers of a model’s advanced capabilities, from complex reasoning to reliable instruction following. However, acquiring and curating such high-quality data remains a significant bottleneck. This tutorial provides a comprehensive and practical guide to the state-of-the-art in data research directions for LLMs. We structure the tutorial around the two core pillars of modern data strategy: intelligent data selection and advanced data synthesis. In the first part, we delve into methods for curating the most valuable information from vast, noisy datasets, covering techniques like LLM-as-a-judge for automated quality filtering and active learning for maximizing annotation efficiency. The second part explores the synthetic data revolution, detailing paradigms that range from generating complex reasoning traces (e.g., Chain-of-Thought) to deploying sophisticated multi-agent workflows that can autonomously create high-quality, diverse instruction data from raw seeds. Finally, we will conclude with a practical overview of open-source tools and platforms that facilitate these data-centric workflows, empowering researchers and practitioners to build better models through better data. Attendees will leave with a principled framework and actionable insights for designing and implementing the advanced data strategies required to build the next generation of powerful, specialized, and aligned LLMs.
尚无 AI 解读，展示原文摘要
cs.CL · cs.MA
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 5: Tutorial Abstracts) · 2026 · 正式发表
Akhil Arora Aarhus University, Vishrav Chaudhary Meta, Julia Kreutzer Cohere Labs, Nearchos Potamitis Aarhus University, Nouha Dziri Allen Institute for AI, Niket Tandon Microsoft Research
As large language models (LLMs) increasingly tackle reasoning-heavy tasks, from mathematics to commonsense to multilingual understanding, researchers face three pressing questions: How well do models reason? How can we make them reason better? What are the next frontiers in LLM reasoning? This tutorial answers these questions through a unified view of LLM reasoning. This tutorial explores comprehensive evaluation strategies to assess the reasoning abilities of models and discusses two types of methods to improve models’ reasoning: advanced inference time methods, such as structured and self-improvement inference methods, and (ii) post-training methods, such as RLHF, DPO, and GRPO that aim to make LLMs think more like humans. The tutorial explores these technical discussions while maintaining a practical outlook through illustrative demos and short guided hands-on exercises. The tutorial is designed for both researchers and practitioners seeking practical insights into LLM reasoning.
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 5: Tutorial Abstracts) · 2026 · 正式发表
Joanne Boisson Cardiff University, Jose Camacho-Collados Cardiff University, Elisa Sanchez-Bayona University of the Basque Country, Rodrigo Agerri University of the Basque Country
This tutorial will be of the type Introductory to CL/NLP topics, providing an overview of the metaphor processing field. The tutorial will discuss the influence of various metaphor theories on the creation of annotated resources and models. We will particularly focus on recent directions opened by LLMs for metaphor interpretation in multilingual and multimodal settings. Two types of audience may benefit from this tutorial: researchers in the humanities and computational social sciences interested in automatic or semi automatic metaphor analysis, and NLP researchers interested in understanding metaphor processing and improving metaphor modeling.
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 5: Tutorial Abstracts) · 2026 · 正式发表