Papers

A spatial answer depends on the reference frame specified in language and its supporting visual evidence. Item accuracy alone obscures whether correctness persists when either changes. We introduce Frame Evidence, a benchmark of 320 matched triplets and 960 Chinese multiple-choice questions from captured and rendered scenes. One family compares three grounded reference frames over fixed images. The other pairs image deletion annotated as insufficient with answer-preserving addition. Joint metrics require correct answers across all three frames and paired evidence variants. Across 20 configurations on 288 held-out questions, GPT three-shot reaches 50.00% item accuracy, but the best three-frame group accuracy is 5/48 (10.42%), reached by GPT two-shot. Few-shot prompting and task-specific adaptation improve item accuracy without consistently improving joint correctness. Input controls show where images support matched success, while repeated executions reveal answer instability beneath similar scores.

Safety alignment of text-to-image diffusion models aims to suppress harmful content without compromising useful generation capabilities. However, preferences over completed images compress multidimensional generation factors into a coarse signal, potentially driving changes beyond those required for safety. To address this, we propose Safety Transition Preference Optimization (SaTPO), which enforces step-wise alignment during denoising. SaTPO leverages the insight that an unsafe prompt and its safe rewrite naturally form a tightly coupled preference pair when conditioned on a shared parent state and noise realization. By anchoring optimization on identical parent states, SaTPO eliminates historical trajectory divergence, confining safety adaptation to an incremental alignment while preserving task-relevant semantics. Multi-scale drift constraints and benign replay further limit behavioral deviation and preserve ordinary generation. We additionally construct CoProPV, a refined dataset ensuring visually verified contrasts. Experiments on Stable Diffusion v1.5 and SDXL demonstrate improved safety-quality trade-offs. Compared with AlignGuard, PerspectiveVision unsafe-detection rates on Inappropriate Image Prompts decrease from 12.70% to 6.00% and from 15.04% to 9.30%, respectively, while FID improves from 32.28 to 25.95 and from 36.50 to 27.51. Warning: this paper includes potentially offensive content.

Parikh's famous theorem relates to the semi-linearity of context-free languages. We seek for expanding the scope of the semi-linear property from context-free languages to languages beyond them. Of those languages, this work focuses on the languages that are particularly recognized by one-way nondeterministic depth-k storage automata under a natural restriction of frozen blank sensitivity. Those languages (which are called $k$-storable languages) have been already proven to satisfy a natural generalization of the well-known pumping lemmas for regular and context-free languages. The semi-linearity of k-storable languages generalizes Parikh's theorem on the semi-linearity of context-free languages. As for bounded languages, Ginsburg and Spanier's well-known theorem characterizes context-free languages in terms of stratified semi-linearity. This notion is further generalized and is used to characterize the k-storable languages as well. Such a generalization is interesting in its own right. This helps us take an abstract treatment of k-storable languages without relying on "machines". Our results strengthen the belief that those languages are "natural" generalizations of context-free languages.

An early mistake in the generated reasoning can constrain all subsequent reasoning. Repeated-generation methods, like MCMC-based approaches, attempt to improve output quality by sampling multiple generations, but their effectiveness can be limited by randomly selecting where to regenerate. On MATH500 with Qwen2.5-Math-7B, we find that incorrect responses have substantially higher mean token entropy than correct responses, with an average within-question relative increase of approximately 112%, suggesting that entropy can signal reasoning errors. Motivated by this finding, we introduce REWIND, a training-free controller that uses entropy dynamics to propose targeted suffix regenerations. Across three model families and five benchmarks, REWIND outperforms all baselines, with gains of 0.51-10.00 percentage points over the strongest baseline. These results demonstrate the effectiveness of targeted, uncertainty-guided trajectory repair across models and tasks. Our code is available at https://anonymous.4open.science/r/rewind-D310.

In long-horizon tool-calling tasks, LLM agents can fail not only because of tool-use errors, but also because of ineffective interaction with users. Yet existing benchmarks largely assume cooperative and behaviorally stable users, leaving the effect of emotional variation on agent reliability underexplored. We introduce $\tau$-Emotion-Bench, a ten-domain benchmark for evaluating tool-calling agents under controlled emotion-conditioned user interactions while preserving the underlying task objective. To scale both benchmark construction and agent training, we introduce Tracer, a scalable agentic data-generation pipeline that constructs grounded, policy-valid tasks and collects trajectories through an interactive agent–user–environment harness. By allowing dissatisfied users to terminate inefficient interactions, the harness provides an interaction-aware signal that filters trajectories with excessive user dependence and favors effective tool use. Experiments with two user simulators show that frontier agents remain unreliable, with substantial gaps between average task success and consistent task completion. Tracer-generated supervision consistently improves Qwen3 models across scales, generalizes to unseen domains, and transfers to $\tau$-Bench and $\tau$-Trait. Behavioral analyses further show that higher task success is associated with a better balance between tool execution and necessary user follow-up. Overall, $\tau$-Emotion-Bench and Tracer provide a scalable framework for evaluating agent reliability under diverse user behaviors and improving tool-calling agents through interaction-aware supervision.

Lipreading remains challenging due to the inherent ambiguity of visual speech cues and their sensitivity to head pose and video resolution. In this work, we propose FLARE, a framework for robust visual speech representation learning via cross-modal distillation and joint face-lip modeling. FLARE leverages semantically rich Whisper representations as the acoustic teacher signal, converting them into discrete distillation targets through a random-projection quantizer and pretraining the model with a masked prediction objective. To capture complementary visual information, FLARE employs a dual-stream architecture that jointly processes lip region-of-interest (ROI) and full-face inputs, where the lip stream captures finegrained articulatory motion and the face stream encodes broader facial dynamics. During pretraining, auxiliary audio input is incorporated to facilitate cross-modal alignment, and modality dropout is applied to encourage robust visual-only representations. For downstream lipreading, FLARE serves as a visual encoder coupled with either a lightweight Transformer decoder or an LLM-based decoder. Evaluated on LRS2, LRS3, and the out-of-domain WildVSR benchmark, FLARE achieves strong lipreading performance, with word error rates of 11.7%, 13.2%, and 34.3%, respectively. A progressive design study confirms the contribution of each major component. Further analyses demonstrate that joint face-lip modeling improves robustness under degraded video resolution and large head pose variation, while yielding more phoneme-discriminative visual representations. Code and pretrained models are available at https://github.com/jxzhanggg/FLARE.

Poset games form a fundamental class of combinatorial games that includes classical games such as Nim and Chomp. Determining the outcome of a poset game is known to be PSPACE-complete even for posets of height at most three. However, the exact computational complexity of Two-Level Poset Game, played on posets of height at most two, has remained an open problem for over a decade. We resolve this problem by proving that Two-Level Poset Game is PSPACE-complete via a polynomial-time reduction from Avoid True on positive CNF formulas.

Contrastive learning for time series depends on deciding which pairs of inputs should be treated as similar and dissimilar. Because few augmentations reliably preserve meaning in time series, prior methods have relied on two disjoint notions of similarity between windows: structural similarity, which relates windows with similar morphology, and temporal similarity, which relates windows that occur close together in time. We introduce TEX-MEX, to our knowledge the first approach to unify the two into a single relationship. Given a random batch of windows drawn across the dataset, TEX-MEX uses the encoder to identify which windows are structurally similar or dissimilar to each other. It then pulls each window toward the temporally shifted versions of similar windows and pushes it away from those of dissimilar ones, learning what typically accompanies similar windows in time. Across PPG, EEG, ECG, accelerometry, and continuous glucose monitoring, TEX-MEX broadly achieves state-of-the-art performance against contrastive, reconstructive, JEPA-based, and modality-specific approaches, establishing it as a strong general SSL objective. We further train a 560K-parameter PPG foundation model that achieves state-of-the-art performance across a broad range of downstream benchmarks while being roughly 50x smaller than the next-best unimodally pretrained PPG foundation model in our evaluation. We will release this foundation model open-source and open-weight to the community to support future work and practical applications.

Confirmation of overdose deaths in the United States is often delayed due to the complexities of certifying drug-involved deaths. Death investigation reports represent a rich but underutilized data source that could be used to predict whether a death is a suspected overdose (i.e., a death preliminarily identified as an overdose prior to the final cause and manner of death determination). We present an automated framework that classifies suspected overdoses from death investigation reports, while producing supporting explanations guided by expert knowledge. Through an expert-guided design process, we investigate 1) what type of explanations are best for model prediction performance, and 2) which architecture is better suited for this task: pipeline models that rationalize first and then predict, or self-rationalizing models that jointly predict and rationalize? Our experiments show that a pipeline model with explanations grounded in the input death investigation reports outperforms the other settings. Our best model also outperforms a rule-based system currently in deployment by 15%, while allowing expert interventions for final decisions.

Recent full-stream text-to-speech (TTS) models enable low-latency personalized synthesis as input text arrives. However, sustaining generation beyond the lengths seen during training without additional training remains largely unexplored. We systematically analyze VoXtream2, which extends generation by periodically discarding past context while retaining the acoustic prompt. We find that this strategy does not prevent degradation: pronunciation accuracy deteriorates near the end of the training range due to the range of positions in which the model operates rather than the loss of past context, while speaker similarity gradually declines throughout generation. Based on these findings, we propose two training-free modifications that reduce the attention window and strengthen attention to the acoustic prompt. Experiments show that these modifications preserve pronunciation accuracy and speaker similarity throughout long-form generation while retaining the naturalness of the original model. Ablations further show complementary effects of the two modifications.

Foundation models for dynamical systems can forecast unseen partial differential equations (PDEs), ordinary differential equations (ODEs), and time series during inference. However, it remains an open question whether these foundation models truly learn generalizable in-context forecasting strategies. We show that looped transformers are capable of emulating propagators for dynamical systems in-context by first establishing a link between linear attention and (extended) dynamic mode decomposition. We then generalize this insight into the $\textit{Test-Time Forecasting}$ (TTF) framework in which a one-step forecaster leverages iterations of gradient descent as recurrent depth. We demonstrate that lightweight TTF models trained on ODEs and PDEs generalize to out-of-distribution system dynamics during autoregressive rollout after being trained for one-step prediction. Our results reinforce TTF as a theoretically motivated framework for designing recurrent depth models capable of forecasting unseen dynamical systems at test-time.

Text-to-image models increasingly generate readable Chinese text that still contains local rendering defects, such as a character with an extra or missing stroke or a malformed component. Such errors are invisible to OCR and vision-language models, which are trained for recognition and are therefore invariant to the very structural deviations that define a rendering error. We formalize text rendering verification as a target-conditioned problem: given a target character and an image crop, decide whether the crop is a correct rendering of that character. Our verifier TEXTVERIFIER feeds the text tower a character's Ideographic Description Sequence (IDS) rather than a single-character symbol, and is trained on 12.7M crop-IDS pairs plus label-free "same-character, different-quality" hard pairs sampled from the training trajectory of a text-region inpainting model. To evaluate this ability we build TEXTRENDERBENCH-ZH, the first per-character benchmark for generated Chinese text, annotating position, identity, and correctness for 7,358 characters. TEXTVERIFIER reaches an F1 of 0.844, far above OCR confidence (0.526), a general VLM (0.329), and TextPecker (0.590). Used as a per-character reward for Flow-GRPO post-training, it further reduces rendering errors, showing that its score can both detect and correct errors.

Closing the loop of AI-driven science requires reliably translating natural-language protocols into executable laboratory workflows despite protocol defects and heterogeneous hardware. Existing approaches often couple scientific interpretation with device-specific code generation, making errors difficult to localize and limiting reuse across laboratory platforms. We introduce LabCompiler, a robust protocol-to-execution framework that transforms imperfect natural-language protocols into verifiable human-machine workflows. Its protocol firewall diagnoses and repairs defects before compilation while preserving unaffected content, guided by troubleshooting knowledge accumulated through perturbation-guided self-evolution. A two-level, device-agnostic graph representation separates experimental semantics from hardware-specific implementation and supports capability-based device assignment with coordinated human execution. We further introduce a protocol recovery benchmark covering four controlled perturbation types and naturally occurring errors collected from protocols written by iGEM participants. Experiments show that LabCompiler outperforms frontier LLMs in protocol error correction while enabling coordinated human-machine execution across heterogeneous, partially automated laboratory settings.

Graph retrieval-augmented generation (GraphRAG) uses document-derived knowledge graphs to assemble evidence for large language model (LLM) answers. Its security depends on both the seed chunks retrieved for a query and the graph evidence assembled from them, yet attacks on this evidence path remain underexplored. We introduce GraphRAG-LTM, a Layered Threat Modeling framework that tests three nested levels of attacker access. BB Query uses ranking feedback to make query edits under a semantic-similarity constraint and demote relevant seed chunks. WB Query uses retriever gradients for the same objective. Graph-aware WB keeps the white-box adversarial query and deletes bridge triplets from the query-specific evidence graph, disrupting multi-hop evidence propagation through localized graph edits. Across datasets, retrievers, and LLM backbones, rank feedback alone reduces answer quality, gradient access strengthens query attacks, and bridge-triplet deletion further impairs supporting-fact retrieval. Further analyses show how seed retrieval and graph evidence jointly contribute to these failures.

Digital libraries and knowledge graphs need a way to organize very large collections so that related works are close and any work is quick to reach. The classifications libraries inherited for this, such as LCC and DDC, are maintained by editorial staff, are partly closed, and were not built against a stated, measurable objective. We propose the opposite: a general classification that is fitted, under the constraints that make a classification usable on a shelf or in a browse list, to the same quantity that a companion survey and benchmark (Awad, 2026) ranks existing schemes by. The objective is J = D + 0.005 N, where D is the normalized Dasgupta cost of the class tree on a fixed similarity graph (Dasgupta, 2016) and N is an expected navigation cost in bits with a per-level charge. Two constraints are hard: every class is a contiguous run of one fixed shelf order (the optimal-alphabetic-tree constraint of Hu and Tucker, 1971), and new works are filed by splitting leaf classes only, so that every existing class keeps its code, label and parent. On a corpus of 10,000 arXiv papers the fitted prototype has 110 hand-labelled classes and reaches cost 0.1722 in sample (average linkage 0.1518; arXiv's own categories 0.3333; the best existing scheme reached through the same lexical classifier 0.6414, from the survey). Filing 500 library books as new works created 16 classes and changed no existing identifier. Fit, filing and scoring share one embedding: the held-out skill of 0.922 falls to 0.896 when the same fit is scored on the graphs of the two other models, and on books the average falls from 0.806 to 0.731. The scheme has also seen only arXiv, and labels for inserted classes are weak. We state those limits plainly, release the scheme as CC0 open data, and ask whether a learned, open, order-preserving classification can become a general one.

In medical annotation and crowdsourcing, multiple annotators assign labels to each example, but these labels may be incorrect. Treating such noisy labels as ground truth can bias calibration evaluation. To address this problem, we propose methods for evaluating and recalibrating a fixed binary predictor under the underlying clean-label distribution, using only noisy annotations. First, given annotator-specific noise probabilities, we propose Noise Corrected ECE (NC-ECE), a binned estimator of expected calibration error (ECE) under class-conditional noise. We derive non-asymptotic bounds on the estimator's bias. Then, we propose practical algorithms for estimating unknown noise probabilities under conditional independence and label-orientation assumptions. We also develop loss-based and histogram post-hoc recalibration with clean-risk and calibration guarantees, respectively. Experiments on semi-synthetic benchmarks show that the proposed ECE estimator closely tracks its clean-label oracle counterpart and that the recalibration methods substantially improve latent-label calibration, whereas uncorrected recalibration against noisy annotations can degrade it.

Image-goal navigation requires a global planner that connects visually specified endpoints while providing observations that a local policy can follow. Three-dimensional Gaussian Splatting (3DGS) provides map-registered RGB, opacity, and depth renderings, but does not explicitly encode navigable topology. We introduce GS-PRM, a Gaussian visual roadmap whose milestones, local connections, and visual references are constructed directly through 3DGS rendering without deriving an occupancy grid or signed distance field. We define render-based collision-free (RC-free) configurations and paths. Panoramic depth supplies conservative clearance for milestone validation and adaptive coverage, while edge-aligned orthographic rendering tests the robot's swept volume. We establish conditional soundness and sufficient conditions for deterministic query completeness and probabilistic construction completeness of GS-PRM. To validate the performance of GS-PRM, we conduct experiments in both simulated and real-world environments, demonstrating long-horizon navigation using goal-conditioned policies guided by GS-PRM. Compared with a state-of-the-art 3DGS-based navigation system, our GS-PRM-based navigation system achieves up to a twice higher average success rate and reduces travel time by up to 57.2% when both systems complete the navigation task.

Automated defect detection of industrial boards must remain highly reliable under cross-scenario variations in lighting, surface contaminants, and production line conditions. This makes cross-scenario defect detection a domain generalization problem, wherein models are required to detect, localize, and classify defects across unseen scenarios while minimizing false alarms on clean boards. For the IDA 2026 Track 1 challenge, we propose a cross-scenario defect detection method that includes four complementary modules: cross-scenario robust training, small-defect-aware learning, score calibration and inference threshold selection, and metric-aligned checkpoint selection. These four modules respectively introduce cross-scenario robust normalization and DSU-style statistic perturbation with progressive random convolution, NWD and parameter-free spatial attention, the VFL learning objective, and a custom SWAD variant driven by the Track 1-oriented proxy. Experimental results demonstrate that our model achieves a Track 1 score of 0.564 with a screening score of 0.696 on the test dataset on a single RTX 4090 GPU. The code is available at https://github.com/GZHU-DVL/Dikai.

Aggregate performance metrics do not identify local modelling inaccuracies in machine learning (ML) surrogates used for thermal-system optimisation. This study assesses local modelling strength through validation-based reliability screening and post-optimisation residual assessment. A two-stage residual learning (TSRL) framework, built on artificial neural networks (ANNs), is developed for modelling-strength-informed and domain-constrained optimisation of thermal systems. Within the TSRL framework, a base ANN predicts the process output, and a residual ANN learns the signed residuals left by the base ANN. The TSRL framework is implemented on two case studies of thermal systems, including 817 simulation observations of a crude distillation unit and operating records from a 395 MW industrial gas turbine. The TSRL framework reduces the test root mean square error (RMSE) from 10.3 g/s to 4.7 g/s for Fuel Oil in the refinery dataset and from 151 kJ/kWh to 105 kJ/kWh for turbine heat rate (THR). The base ANNs trained within TSRL are embedded in domain-constrained optimisation problems, and the modelling strength at each computed solution is assessed using the locked residual-score threshold. Each computed solution receives a good modelled region (GMR) or weak modelled region (WMR) label to inform ML model-based decision-making for safety-critical operation of thermal systems.

3D Gaussian Splatting (3DGS) enables high-quality, real-time novel view synthesis but requires significant storage and transmission resources. Mapping Gaussian attributes to 2D representations enables efficient compression using mature image codecs. However, Gaussian representations that achieve comparable rendering quality are generally treated similarly during optimization, despite substantial differences in coding rate, leading to limited rate-distortion (R-D) performance after compression. To address this limitation, we propose RDCompGS, a 3DGS compression method that enables overall R-D optimization of Gaussian representations with respect to subsequent coding. A differentiable estimate of the coding rate is incorporated into the R-D optimization with rendering distortion. The rate constraint adjustment model further regulates the R-D trade-off across different compression conditions. Extensive experiments on three benchmark datasets demonstrate that RDCompGS achieves average Bjøntegaard Delta Rate reductions of 37.84% and 49.31% with H.265/HEVC and JPEG XL, respectively. The estimated rate also shows a strong correlation with the actual bit-rate under different compression conditions.

The Emergency Vertex Cover problem (Em-VC) assigns budgets to stations at graph vertices to cover every edge while minimizing the total budget. The cost of covering an edge from a station is the shortest-path distance from the station to the nearer endpoint plus the edge length. We study how restricting the number of stations affects the cost of ordinary and fault-tolerant coverage. In ordinary Em-VC, we prove that the optimal solution using at most two stations costs at most $5/4$ times the unrestricted optimum, improving the known $4/3$ guarantee for a single station. A recurrence extends this bound to larger station limits, giving approximation factors that converge to $\sqrt{5}-1$. Budget-threshold enumeration computes the restricted optimum in polynomial time for every fixed station limit. Under $k$-fold coverage, every edge must be covered by at least $k$ distinct stations, ensuring coverage after any $k-1$ station failures. We prove that selecting the $k$ vertices with the smallest full-coverage costs yields a $2$-approximation relative to the unrestricted optimum for $k$-fold coverage. When $k=2$, we combine upper bounds derived from the structure of the covering balls to improve the approximation guarantee to $5/3$.

Typed decision models are language models that answer a question with a fixed set of possible answers by returning a probability for each answer instead of generating text. We study them on tabular data and compare four decision models with three large language models (LLMs) on row classification, fact verification and anomaly detection. Every request is sent in its original form and under 21 types of rewrite that keep its content, such as renaming the options or writing the table in another format. The decision models perform as well as the LLMs at a lower cost. Three of them are also far more robust to how the question is formulated. No model we test is robust to how the table is organized. For every model, at least a third of the row decisions change under some rewrite of the table or the question. The decisions to which a decision model gives low probability are the ones most likely to change. Decision models are a practical choice for tabular decisions, but their robustness remains an open problem.

Existing student models in educational applications largely focus on student responses to questions. However, as tutoring becomes more accessible at large scale, it is important to analyze dialogues between tutors and students, which may reveal fine-grained evidence on student knowledge. In this work, we propose a unified framework that models both student responses to questions and tutor-posed subtasks during dialogues on a shared item response theory (IRT) scale. We first identify mathematical subtasks in tutoring dialogues, rewrite them into self-contained questions, and identify the correctness of student responses to these subtasks. We then develop an IRT framework that estimates question/subtask difficulty from text, tracks students' evolving abilities across question-answering and dialogue interactions, and a novel method to leverages tutor moves between consecutive subtasks to regularize their difficulty values. We evaluate our framework on two real-world math education datasets containing both human- and AI-tutored dialogues. Results show that our framework outperforms strong baselines on both question difficulty prediction and response correctness prediction.

We establish nearly sharp $W^{1,\infty}$ stability estimates for the Legendre spectral Ritz projection associated with Laplacian on a square domain. Its norm grows proportionally to the square root of the polynomial degree, up to a cubic logarithmic factor. The upper bound uses an integral representation of the discrete gradient projection in terms of one-dimensional semigroup kernels and a rank-two Duhamel comparison between the mass-matrix semigroup and its infinite-dimensional counterpart. The lower bound follows from the endpoint concentration of the Legendre kernel. As an application, we derive gradient maximum-norm error estimates for the Poisson and heat equations, with spatial factors optimal up to logarithmic terms.

As a lightweight and highly accurate method for nearest-neighbor search in 3D point clouds, approaches based on space-filling curves (SFC) have been proposed. However, due to the inherent properties of SFC, a gap arises between proximity relationships along the SFC and those in the original Euclidean space. In this study, we propose a loss function that reduces the influence of this gap. Specifically, we compute the distances between adjacent points along the SFC and train the model to optimize feature similarity between pairs of points that are distant from each other in the original 3D space. We incorporate the proposed method into a high-performing 3D point cloud segmentation model and evaluate its effectiveness on three datasets. The results demonstrate that the proposed method outperforms conventional methods on both indoor and outdoor point cloud datasets.

A large number of theories in international relations attribute importance to the rhetoric of leading foreign policy actors. Moreover, in recent years, there has been a significant growth in interest for producing machine-readable text of remarks from key foreign policy actors. Yet, researchers have, to date, focused primarily on the rhetoric of executive leaders and members of legislatures. To understand the construction of foreign policy in the public arena and dynamics in it, especially in the United States, we need a collection of fine-grained textual representation of the remarks of US Secretaries of State. An especially promising use for such a corpus of text is to measure attention areas in US foreign policy and its implication for the US's policies, including in foreign aid. In this note, We introduce the United States Secretary of State Dataset, a corpus of 17,700 remarks delivered by US Secretaries of State from 1997 to 2026. Using this dataset, we demonstrate that Secretaries of State frequently make public remarks, the likes of which can have substantively meaningful implications.

Retrieving external knowledge can improve large language model (LLM) translation by providing supplementary information such as idiom interpretations and terminology. However, existing retrieval-augmented machine translation systems typically apply predefined retrieval strategies uniformly across all inputs. We argue that translation operates in a partial-evidence regime: LLM's parametric knowledge suffices for many inputs, while external evidence benefits a subset and usually supports rather than directly provides the desired translation. This creates two challenges: deciding when to retrieve and what evidence to acquire. We introduce Selective Retrieval-Augmented Translation (SRAT), an agentic reinforcement learning framework that learns when to translate directly or retrieve, what to retrieve, and how to integrate these decisions by optimizing final translation quality. To align evidence retrieval with downstream translation quality rather than simple semantic similarity, we further develop joint Plackett–Luce optimization and translation-quality contrastive training for retriever-training. Experiments on idiom, knowledge-intensive, and document-level translation with offline or online retrieval show that SRAT consistently outperforms never- and always-retrieve baselines, and quality-aligned retriever training provides further gains when similarity-based retrieval is insufficient. The results demonstrate the benefit of treating retrieval as a learnable translation action rather than fixed preprocessing.

We devise a dual-arm cooperative planning framework for wiping table surface in the presence of scattered objects. Unlike conventional surface-cleaning strategies that only cover free areas and detour around obstacles, our framework explicitly assigns one arm to rearrange objects while the other performs coverage wiping, thereby enabling access to previously obstructed regions. Specifically, we first construct a set of wiping and object-rearrangement action primitives. Based on these primitives, we introduce Geo-MILP, a geometry-aware task planner formulated as Mixed-Integer Linear Programming (MILP). By integrating Logic-Based Benders Decomposition (LBBD), Geo-MILP iteratively optimizes action sequencingand timing to resolve geometric conflicts, promoting efficient cooperation between wiping and object rearrangement. We then employ K-Order Markov Optimization (KOMO) to generate smooth, collision-free motions to ensure feasible execution. Simulation and physical experiments demonstrate that the proposed framework not only increases reachable coverage by enabling access to obstructed areas, but also significantly improves time efficiency compared to Task and Motion Planning (TAMP) baselines, reducing planning time by up to 9 times and execution makespan by approximately 40%.

Temporal action detection in long untrimmed videos suffers from inaccurate localization and low-quality proposals, especially with sparse annotations and ambiguous boundaries. For open-ended video understanding and LLM-driven event analysis, reliable proposals connect temporal localization, event annotation, and semantic understanding. However, conventional temporal action detection (TAD) pipelines lack explicit modeling of proposal localization quality, cross-representation support, and inter-proposal relations. We propose Hybrid Rank-Calibration of Action Proposals (HyrCap), a proposal-centric query-based TAD framework for temporal event understanding. Hybrid combines complementary video features with different localization cues, while Rank-Calibration adjusts proposal ranking confidence to reflect localization reliability and candidate competition. HyrCap treats dense query-based predictions as an over-complete temporal event hypothesis space. Cross-Representation Consensus Calibration, Localization Quality Estimation, and Proposal Redundancy Suppression improve proposal ranking reliability, localization quality, and candidate discrimination. Inspired by LLMs' capabilities in fine-grained semantic recognition, event description, and reasoning, we introduce HyrCap for Semantic Action Understanding, using calibrated proposals for standard TAD and as inputs to LLM-based semantic understanding. We define a staged evaluation protocol to separately measure event localization, segment-level semantic understanding, and combined event understanding performance. Experiments on public benchmarks show that HyrCap achieves state-of-the-art performance across multiple metrics and provides a finer-grained paradigm and evaluation protocol for query-based event segmentation and annotation.

Verifiable instruction-following benchmarks emphasize output-constraint satisfaction but can overlook whether responses meaningfully answer the underlying request, creating an evaluation blind spot. We show that frontier models can receive full credit for responses with little substantive answer content. Using the same verifier-based criterion as a training reward can induce reward hacking, as publicly released checkpoints exhibit higher constraint satisfaction but lower response quality after reinforcement learning with verifiable rewards. To address this reward hacking, we introduce CALIF, a lightweight method that augments verifier rewards with a calibrated likelihood-based bonus to preserve response quality. CALIF evaluates response likelihood under the request alone using the existing frozen reference policy, requiring no additional judge or learned reward model. Across models ranging from 1.7B to 8B in reasoning and non-reasoning modes, CALIF improves constraint satisfaction on IFBench while maintaining response quality near or above pre-RL levels. It also outperforms corresponding base policies in 40 of 45 model-metric comparisons across three rubric-based instruction-following benchmarks, demonstrating gains beyond training verifiers.