Papers

Extending reliable nowcasting of extreme precipitation could provide critical additional time for warnings and emergency response during high-impact events such as flash floods. Radar-based generative machine-learning models have enabled skilful hyperlocal precipitation nowcasting, but accurate prediction of intense precipitation remains confined to the first few hours. Because storm-scale structure is predictable for longer than individual cells, a natural strategy is to predict that structure while generatively modelling only the uncertain local growth, decay, reorganisation and initiation of storms. Here we present Microsoft Weather Nowcast (MW-Nowcast), a six-hour ensemble radar nowcasting model that jointly learns a deterministic predictor to capture organised precipitation structure shared across ensemble members, and a generator to produce diverse local residuals around this shared prediction. Across independent test data from the United States, Europe and China, MW-Nowcast achieves higher detection skill than leading methods for heavy and extreme precipitation throughout the 6 h horizon. For the most intense rainfall, MW-Nowcast doubles the available warning time across all three regions, delivering 6 h forecasts with skill previously limited to 3 h for the leading generative baseline. A cost-loss decision analysis shows that MW-Nowcast retains substantial value for a broad range of applications even at 4-6 h, where alternative methods offer little benefit. These additional hours can give forecasters and emergency managers the time to warn and act before extreme rainfall strikes, helping to protect lives and property.

On the upper hemisphere, we use the Obata-Escobar argument to establish an Obata theorem for $\sigma_k$-flat and positive constant boundary $\mathcal B_k$-curvature equation, which was initiated by Case-Wang [Adv. Math. 337 (2018), 83-106].

Transform-based methods provide an effective framework for point cloud attribute compression by representing attributes as transform coefficients. Introducing learned spatial context into this framework requires mapping spatial representations to the transform domain, but this known basis change is often left for the network to learn implicitly. We propose Transform-Aligned Learned Features (TALF) by applying the attribute transform to learned spatial representations, explicitly aligning them with the coding targets. Our analysis shows that the resulting features exactly represent the first-order prediction term of a smooth nonlinear model, with a bounded Taylor remainder. We integrate TALF into a transform-based attribute codec with explicit coefficient prediction and conditional residual entropy modeling under a unified coefficient-domain rate–distortion objective, while retaining explicit quantization-step control. Extensive experiments across three benchmark datasets and multiple transform bases demonstrate that TALF improves rate–distortion performance over conventional and learned baselines.

Semantic mapping is fundamental to embodied navigation, yet existing methods are developed for indoor environments, where objects exhibit relatively limited scale variation and are observed from a restricted range of viewpoints. Urban environments pose substantially greater challenges: agents must map objects ranging from pedestrians to buildings while navigating large spaces with highly diverse viewing distances. These conditions introduce two key difficulties that existing datasets and methods fail to cover. First, object scale and observation distance can be severely mismatched. For example, small objects may be viewed from far away, whereas large objects may be observed at extremely close range, resulting in unreliable observation likelihoods. Second, objects with substantially different sizes and geometries require distinct mapping behaviors, which are difficult to capture with a single shared value estimator. To investigate these challenges, we introduce a large-scale urban semantic mapping dataset featuring realistic city layouts, high-fidelity rendering, and instance-level annotations spanning multiple object scales. We then propose a category-aware likelihood calibration policy that identifies and alleviates unreliable observations according to object category and viewing distance. Because the calibration and motion policies are optimized toward the same mapping objective, they may learn redundant shortcuts and become excessively coupled. We therefore introduce a mutual-information (MI) regularizer that penalizes their estimated representation dependence and encourages complementary behaviors. To better model heterogeneous mapping strategies across object scales, we further employ category-wise value estimators. We formulate their joint optimization as a Pareto optimization problem to mitigate conflicting gradients across categories.

Diffusion drafters accelerate speculative decoding by proposing multiple tokens in parallel. Despite recent advances in speculative decoding through sequence-level drafting and verification, existing training objectives remain largely designed around token-level verification. To address this mismatch, we introduce Block Verification-aware loss (BV loss), a training objective designed to maximize the expected acceptance length of a drafted sequence. BV loss is directly derived from the block verification acceptance rule, providing a principled connection between the drafter training objective and the inference-time verification mechanism at the sequence level. Across math, code, and chat benchmarks, BV loss increases the mean number of tokens accepted per verification call under block verification by 13.0–21.0% over cross-entropy loss training for DFlash and DSpark with Qwen3-4B and Qwen3-8B without changing the inference procedure. BV loss also outperforms tokenwise acceptance objectives such as TV loss and LK loss, and its gains extend to token verification and greedy decoding. These results demonstrate the benefit of training block diffusion drafters with an objective aligned with sequence-level verification, rather than optimizing each token independently.

Neural Stochastic Differential Equations (Neural SDEs) provide flexible continuous-time generative models, but generic neural drift and diffusion networks are costly to simulate on long horizons and can give unstable gradients when the training signal is a path functional rather than a pointwise observation. We introduce SLiSDE, a family of Neural SDE models built from structured linear stochastic layers. Parallel-in-time simulation is obtained at the layer level, while expressivity is recovered by gated in-flow stacking: previous-layer paths modulate the next layer's latent flow through learned gates. For functional calibration tasks in which rare paths dominate the loss, we add an optional Girsanov tilt that acts as a learned importance sampler with an exact likelihood-ratio correction. We prove well-posedness, a discretisation error bound, validity of the change of measure, and a universality result: the terminal laws of the gated stack are dense in the space of square-integrable laws. Experiments on functional calibration benchmarks show that the structured model outperforms fully neural SDE baselines while retaining parallel-time simulation and stable importance weights.

We study deformations of foliated complex analytic structures defined by locally free subsheaves of tangent sheaves and locally free subsheaves of cotangent sheaves on the basis of Kodaira-Spencer's deformation theory. We prove theorems of existence and completeness for deformations of foliated complex analytic structures defined by locally free subsheaves of tangent sheaves as analogues of theorems of existence and completeness for deformations of complex analytic structures by Kodaira-Spencer. We prove theorems of existence and completeness for deformations of foliated complex analytic structures defined by locally free subsheaves of cotangent sheaves. We also prove theorems of existence and completeness for simultaneous deformations when singular holomorphic foliations on compact complex manifolds are defined by locally free subsheaves of tangent and cotangent sheaves at the same time.

Most methods that optimize LLM prompts and agent workflows assume that task-specific output schemas, extraction instructions, and evaluation criteria are predefined. For scientific extraction agents, however, a short task goal may not fully determine these components, while specifying them manually is costly. We study the upstream problem of constructing the task-specific configuration from a weak specification containing only a short goal and unannotated reference documents. Rather than treating automatic construction as a fixed preprocessing step, our framework constructs a task-specific schema, extraction instructions, and base training rubrics, then keeps schema construction and extraction instructions editable during optimization. Failure-focused updates concentrate textual-gradient feedback on lower-scoring documents, while training-time evaluation criteria adapt to recurring failures. On a heterogeneous-catalysis literature corpus, automatic construction remains improvable, and optimizing both schema construction and extraction instructions performs best across all four judge-rubric settings, with ablations and blinded human evaluation supporting the proposed formulation.

Large language models are increasingly used to simulate response distributions in social surveys. Prior work has achieved accurate population-level simulation for individual questions. Real questionnaires, however, ask each respondent a sequence of related questions. A simulated respondent should show coherent preferences across the whole questionnaire, not merely accurate distributions for isolated items. Existing single-item methods cannot accurately reproduce how the same person answers a complete survey. We propose FullRespondent-LLM (FR-LLM), which fine-tunes two specialized LLMs: a marginal model for each item's response distribution and a respondent-level autoregressive model for dependencies across answers. Marginal-Constrained Joint Projection (MCJP) then projects the autoregressive joint distribution onto the set satisfying the item-level marginals learned by the first model. This yields complete questionnaires with realistic cross-item relationships while retaining strong item-level accuracy. On two real-world social survey datasets, FR-LLM more accurately reproduces multi-question response patterns, maintains competitive single-item accuracy, and generalizes better to unseen populations and questions. In a small commercial-survey dataset, we use simulated responses to make pricing and stocking decisions; FR-LLM achieves the highest realized profit.

We present a detailed photometric and spectroscopic analysis of Janus-11, a $z=10.7933 \pm 0.0006 $ galaxy magnified by a factor 1.6 by the galaxy cluster MACS J0416. It was observed with JWST/NIRCam as part of the PEARLS program and with JWST/NIRSpec MSA as part of the DREAMS program. It has M$_{\rm UV}=-19.78^{+0.12}_{-0.11}$, and its stellar mass of log(M*/M_sun) = 8.75 (+0.08/-0.10) places it among the most massive known $z>10$ galaxies, while its $A_V=0.54\pm0.09$ mag and UV slope of $\beta=-1.71\pm0.26$ are compatible with a red monster classification, albeit among the least obscured members of this population: this could hint at a transition from the red monster to the blue monster phase. Moreover, the [OIII]4363 detection allows for the highest-redshift metallicity measurement based on the [OIII]4363 direct method to date of $12 + \log(O/H) = 7.74^{+0.11}_{-0.10}$, together with an electron temperature of $T_e$(O$^{++}$) = $(2.1 \pm 0.2)\times10^4$ K. Its spatially resolved nature, with an effective radius of $136 \pm 9$ pc, together with the narrow FWHM of the emission lines ($\lesssim 230$ km s$^{-1}$, or $\sigma_v \lesssim 100$ km s$^{-1}$), and the lack of significant detections of high-ionization emission lines and broad components, favor an SFG classification, although an AGN contribution cannot be excluded. Janus-11 joins the small sample of spectroscopically confirmed and characterized sources at $z>10$, providing new information on massive systems with non-negligible dust attenuation and well-measured metallicity in this epoch.

Humans often solve spatial problems by mentally simulating visual transformations. In contrast, conventional vision-language models (VLMs) reason primarily through language. We investigate whether VLMs can solve spatial problems by reasoning with both text and generated visual states. To this end, we introduce WM-VLM, which equips a pretrained VLM with a lightweight world model branch for generating intermediate visual states. Our two-stage training first teaches the model to generate the next visual state and then to use that state for reasoning. We programmatically construct spatial reasoning tasks with verifiable intermediate visual states. These tasks allow us to evaluate how well the model generates visual states and how much it relies on them to answer the question. On 2D and 3D mental rotation tasks, WM-VLM consistently outperforms the supervised fine-tuned backbone, with gains of up to 39.25 percentage points. Ablations suggest that these gains depend on the generated visual states, as removing or corrupting them sharply reduces performance. Together, these results suggest that internal world models offer a promising path toward VLMs that reason in both language and visual space. Code is publicly available at: https://github.com/yuh-zha/WM-VLM

Gaussian neural networks (GaNNs) are proposed as a novel regularization mechanism for neural networks. From a Bayesian perspective standard regularization techniques can be viewed as imposing priors over weight-space. Assuming priors over activation-space remains a largely unexplored possibility. GaNNs assume such priors. They do this by treating activities from earlier layers like signals with Gaussian noise and predicting the properties of the noise distribution using an additional unsupervised loss. While training, the unsupervised loss acts as a penalty on unexpected activities, allowing greater weight updates in less surprising directions. The paper demonstrates the superiority of Gaussian neural networks over standard neural networks on a variety of classification and regression tasks. We also investigate the ability of GaNNs to quantify uncertainty.

Under a shared-factor model, the conditional law of the next image patch factors into a posterior over the shared scene factor and a residual kernel given that factor. A sufficient statistic of the past replaces the raw past in the posterior and does not replace the kernel. The conditional entropy splits into residual entropy, which no observation of the factor can remove, and a posterior term, which a better representation of the past can remove. Next-embedding prediction is a directional likelihood on a shallow map, so the fiber of that map is unidentified and a constant embedding remains a minimizer. The same split is an equality in a scalar Gaussian model, evaluated in closed form.

Augmented dexterity has the potential to reduce the fatigue experienced by surgeons during repetitive surgical tasks. In this paper, we propose the first AGentic RObotics framework for SUrgical VIscoelastic DEbridement (AGRO-SUVIDE), the repeated removal of small fragments attached to a viscoelastic substrate. Leveraging the self-improving and coding capability of agents, AGRO-SUVIDE adopts a modular framework. Specifically, the demonstration analysis module automatically identifies recurring skills from a single expert demonstration, using both visual and kinematic information. The construction module then builds each skill, either as a procedural model-based skill the agent codes against a scaffolded library or as a model-free policy-based skill. At runtime, the monitoring module composes the skills into a loop-style graph sized to the number of fragments it observes, then verifies pre- and post-conditions of each skill to decide whether to advance or retry. We evaluate AGRO-SUVIDE through 340 physical trials on the da Vinci Research Kit (dVRK). AGRO-SUVIDE achieves an average single-fragment removal success rate of 85%, completing consecutive three-fragment removal at 60% and at 95% with one human intervention. It further generalizes to unseen five-fragment scenarios with an average success rate of 80% for single-fragment removal. Project page: https://surgical-robotics.github.io/AGRO-SUVIDE/

We extend Sullivan's rational homotopy theory by constructing geometrically and algebraically minimal models for diagrams of simply connected spaces over a class of finite indexing categories. Both classes classify rational diagram homotopy types, but their minimality conditions can differ. We also show that, in contrast to classical rational homotopy theory, H-diagrams need not split into products of Eilenberg–MacLane diagrams, and characterize the indexing categories for which every H-diagram admits such a splitting. Also we apply this to computation of automorphism groups of rational homotopy types of certain diagrams of spaces.

Long-horizon competition tests agents' ability to coordinate business decisions under uncertainty and adapt to changing rival strategies. We introduce CEO Arena, a benchmark that uses matched replacement evaluation to assess operating returns alongside an agent's effects on rivals and the market. Each CEO agent is compared with a reference policy in the same company under the same economic seed, holding other agents' identities and assignments fixed while all agents adapt. In a shared eight-company market spanning 500 simulated days, CEOs make sequential decisions on pricing, procurement, marketing, research and development, and service using private company information and noisy market signals, under resource constraints and delayed feedback. We evaluate eight LLM-based CEO agents in 27 main runs and 26 robustness runs. In the main evaluation, most agents have negative mean returns, and private gains can accompany market losses. Robustness analyses suggest that aggregate patterns extend beyond the original rule-based baseline; four of the 56 directed pairs show relatively stable effects. Memory, action, and accounting traces suggest demand capture and rivals' pricing and spending responses as possible explanations. CEO Arena provides a controlled testbed for studying long-horizon agent competition, strategic interaction, and market externalities.

It has long been believed that newborn quark stars are inefficient emitters because the quark plasma frequency is so high that it suppresses electromagnetic radiation across the entire spectrum up to energies of several tens of MeV. Here, we show that significant electromagnetic emission can instead originate from the electrosphere of a quark star, where the plasma frequency is much lower, approximately 40 keV. As a result, the cooling of a newborn quark star is expected to proceed through two distinct observational epochs. During the first epoch, blackbody radiation is emitted from the photosphere of an optically thick relativistic wind generated near the electrosphere. Over the course of a few days following the star's formation, the photospheric temperature decreases from several MeV to about 22 keV. During the second epoch, the blackbody spectrum evolves into a narrow feature centered around 40 keV, and the emission gradually fades. This spectral evolution provides a characteristic signature of the electrosphere surrounding compact astrophysical objects and may be detectable with current or future X-ray observatories.

The (naive) de Rham cohomology of a zero-dimensional scheme is the homology of the Kähler differential algebra of its coordinate ring, viewed as a complex. Since it is well-known to vanish in higher degrees if the ring is quasi-homogeneous, we concentrate on the affine case. For a general affine $K$-algebra $R=P/I$, where ${\rm char}(K)=0$ and $P=K[x_1,\dots,x_n]$, we prove a vanishing theorem for $H_{\rm dR}^m(R)$ based on the shape of a Macaulay basis of $dI\wedge \Omega^m_{P/K}$. For an Artinian local algebra $A=P/\langle f_1,\dots,f_n\rangle$, where $\{f_1,\dots,f_n\}$ is a super regular sequence, we show that $H_{\rm dR}^{\bullet}(A)$ is non-trivial in general, but trivial when the natural system of generators of the relation module of $\Omega^m_{A/K}$ is a standard basis. Moreover, we provide a detailed study of the dimension of $H_{\rm dR}^0(A)=\ker(d_A)$ for Artinian local rings $A=P/I$. The case of arbitrary affine zero-dimensional schemes is reduced to this case using a Galois splitting, and as a result we obtain the de Rham cohomology of a fat point scheme. Many explicitly computed examples and counterexamples support the results and indicate how subtle the de Rham cohomology of a zero-dimensional scheme is in general.

AlphaFold2 is written in JAX, so the same inference code compiles and runs unchanged on CPUs, GPUs and Google Cloud TPUs. That portability makes the accelerator look like the main decision a user has to make. We show that it is not. Running one AlphaFold2 inference workload across a Colab CPU runtime, an NVIDIA T4 GPU and a dedicated eight-chip Cloud TPU v5e slice, we find a large hardware advantage for the TPU, 0.47 s per call in steady state on a single chip against 13.1 s on the T4 in the same measurement campaign, and three ways in which the software layer decides how much of it a user actually gets. The default execution path uses one chip of the eight, and at list prices the idle capacity makes the slice cost about as much per prediction as the GPU. Batching with JAX's vmap never exceeds single-query throughput, while mapping queries across chips with JAX's pmap gives eight chips 6.5-7.9x the throughput of one on a matched grid; automatic sharding leaves the per-chip footprint unchanged, consistent with replication, most plausibly because AlphaFold2 carries no sharding annotations. Our retained trace analysis of a first call at a new input shape reports about three quarters of the traced span in JAX tracing and compilation rather than execution. Reruns five weeks later reproduced neither cloud baseline, the GPU one off by roughly a factor of two, so the hardware ratio above is specific to one campaign.

Human dexterity is guided by two eyes watching two hands: binocular vision supplies the metric 3D structure that fine-grained manipulation consumes. Egocentric stereo is therefore the natural perceptual interface for robots, AR, and VR-yet metric 3D hand reconstruction from this very signal still has neither an end-to-end model nor an in-the-wild benchmark. We propose ESTHER, a model whose stereo geometry, temporal reasoning, and output representation are designed for wearable egocentric stereo. It is trained on pseudo-labels from a calibrated labeling pipeline and in turn assembles our benchmark ESTHER3D, an egocentric stereo hand dataset pairing a large in-the-wild training set of model-generated labels with a motion capture test set of true metric ground truth. Experiments show state-of-the-art accu?racy, superior external generalization, and robustness to the missing views, dropped frames, and lighting and motion blur extremes of real egocentric capture that break existing meth?ods. This robustness runs deeper than graceful degradation: stereo guidance teaches the model to bind apparent hand scale to metric depth, so it not only adapts to different stereo rigs and modalities with minimal fine-tuning, but more strikingly preserves true metric scale even after collapsing to a single monocular view.

GRB 240825A shows unprecedented temporal and spectral features that constrain the properties of its inner engine: (1) a prompt emission three-component spectrum consisting of a Band function continuum, a quasi-thermal bump, and a hard MeV tail; and (2) an X-ray afterglow fitted by a power-law decay with index $1.29 \pm 0.02$; and (3) a $6.37 \pm 0.05$ Hz quasi-periodic oscillation (QPO) identified in the 100-300 keV band during the 2.07-3.25 s time interval, coinciding with the photospheric radiation phase. We compare two central-engine candidates: (a) a newborn millisecond magnetar undergoing multipolar spin-down, free precession, and global magnetoelastic oscillations of its interior; and (b) a Kerr black hole powering a Blandford-Znajek jet subject to Lense-Thirring disk precession. In principle, both channels can reproduce the observed QPO frequency but impose different constraints on the energetics, temporal evolution, and the origin of the oscillation. We first demonstrate that the X-ray afterglow is quantitatively reproduced by a hexapolar magnetar spin-down model with an initial spin period of $P_0 \simeq 1.38$ ms and a surface field of $B_{\rm hexa} \simeq 2.04 \times 10^{16}$ G. The decay index measures a braking index rather than the field multipolarity, does not exclude a fallback-regulated flow, and leaves $P_0$ and $B_{\rm hexa}$ uncertain by a factor of a few. In this scenario, magnetic or magnetoinertial dynamics of the star interior provide a plausible explanation of the QPO without invoking extreme stellar deformations. Although a black hole engine cannot be firmly excluded, the combined prompt and afterglow observational properties of GRB 240825A favor a long-lived magnetar central engine.

We consider the following singular Liouville equation $-\Delta v=\lambda V(x)|x|^2e^v \quad \text{in } B_1, \quad v=0 \quad \text{on } \partial B_1, $ where $B_1\subset\mathbb R^2$ is the unit disk, $\lambda>0$ is a small parameter, and $V$ is a positive smooth function. We first prove a non-degeneracy result for the bubbling solutions constructed in \cite{D-W-Z2026} by the local Pohozaev identities. Then we use this non-degeneracy result to establish the local uniqueness of bubbling solutions whose concentration parameters are sufficiently close.

Long-video generation with diffusion transformers (DiTs) produces extremely long token sequences, making attention a dominant inference bottleneck. Dynamic sparse attention reduces computation, but its realized speedup remains limited because irregular query-row execution degrades L2 cache locality and increases HBM traffic. We present WaveAlign, a lightweight, cache-aware query-row reordering framework for dynamic sparse attention. WaveAlign formulates row ordering as an optimization problem and approximates it with two stages. The first stage derives a low-rank SVD representation of sparse-mask rows and groups query rows with similar K/V access patterns, increasing K/V overlap among concurrently scheduled rows. The second stage exploits streaming GPU scheduling by sorting rows within each wave in descending order of their K/V-block counts, so that short rows from the current wave are followed by long rows from the next. This aligns K/V accesses across wave boundaries and enables shared blocks to be reused before eviction. An adaptive skip module avoids unprofitable reordering. By only permuting query and mask rows, WaveAlign preserves sparse-attention semantics and requires no changes to existing methods or backend kernels. Across two GPU architectures, two video DiTs, and four sparse-attention methods, WaveAlign raises the L2 cache hit ratio from 28.48%–36.35% to 79.38%–89.06%, reduces HBM read traffic by up to 92.11%, and achieves up to 1.25x kernel and 1.17x end-to-end generation speedup without quality loss.

Given a monoid that can be embedded in a group and that acts partially on a set, we show that under certain conditions the category associated with the partial action has a groupoid of fractions. By imposing some topological conditions on the action, we show that the groupoid is actually étale. This étale groupoid arises naturally in $C^{\ast}$-algebra theory.

We study Nash regret minimization in unknown finite matrix games with bandit payoff feedback and observed opponent actions. We develop Optimistic Payoff Balancing (OPB), which achieves instance-dependent $\mathcal{O}(\log^2 T)$ Nash regret against arbitrary adaptive opponents, including games with nonunique equilibria. This resolves the open problem posed by Maiti et al. (2025), extending their polylogarithmic guarantee under bandit feedback from $2\times2$ games to arbitrary finite dimensions. To handle nonunique equilibria, we construct a reference strategy that leaves room for local adjustments. We order independent payoff differences by estimation accuracy and scale these adjustments by uncertainty, allowing the learner to exploit the opponent's imbalance to offset estimation costs. Our result thus shows that observing opponent actions suffices for polylogarithmic Nash regret in general finite matrix games.

The exact one-point function of Liouville theory on the real projective plane was derived long ago from the bootstrap, yet it has never been verified against a semiclassical path-integral computation. The check is less straightforward than one might expect. On the real projective plane every constant-curvature metric is positively curved, whereas classical Liouville theory produces metrics of constant negative curvature, so no real classical solution exists. Instead, as we show, the semiclassical path integral receives contributions from infinitely many complex saddles with a negative-definite metric. Once the integration contour is chosen so that the path integral converges, these saddles reproduce the exact one-point function. Complex saddles arise in the same way in timelike Liouville theory and in two-dimensional de~Sitter gravity, and the example treated here offers a rare opportunity to test their use against a known exact answer.

Reinforcement learning has become an effective approach to training language model agents, but sparse and delayed outcome rewards provide limited guidance for credit assignment across long interaction sequences. Recent work on on-policy self-distillation (OPSD) offers complementary supervision by evaluating a policy's sampled responses under privileged training-time context. However, our diagnostics show that positive average agreement between outcome and hindsight feedback coexists with substantial local disagreement, raising the question of how to allocate influence between them at each decision. We introduce UniOPSD (Unified On-Policy Self-Distillation), which unifies these feedback sources through adaptive local credit arbitration. UniOPSD constructs comparable credit estimates from environmental returns and successful-peer hindsight at shared interaction anchors. Historical agreement determines the global mixing level, while current signal availability and relative precision adjust each source's influence at individual decisions. The episode-level outcome contribution is retained, and bounded token modulation refines the fused step credit for policy optimization. With Qwen2.5-3B-Instruct and Qwen2.5-7B-Instruct, UniOPSD achieves ALFWorld success rates of $82.8\%$ and $83.6\%$, WebShop success rates of $75.0\%$ and $82.0\%$, and Search-QA aggregate accuracies of $45.3\%$ and $49.8\%$, respectively. On 3B WebShop, UniOPSD improves over SDAR by $7.0$ percentage points. Our code is available at https://github.com/Zenghuang-Fu/Uniopsd

Vision-language models (VLMs) can answer simple visual questions, but often struggle when one question requires several visual judgments. We study this gap with controlled tasks for feature binding, numerosity, spatial relations, and amodal completion, together with a Composite task that combines them. Matched counterfactual image pairs isolate changes in the visual evidence needed to answer. Across four models, direct answers, hidden-state readouts, and state interventions show that the individual judgments can be made without explicit reasoning and that intervening on the corresponding states can affect the answer. During reasoning, the Composite answer becomes decodable from hidden states and usable from shortened traces, often before the model stops on its own. We train a small detector to predict this readiness and stop reasoning at that point. On MMStar and RealWorldQA, this reduces mean reasoning tokens by 79.1% and 74.5%, while average accuracy rises by 3.13 and 3.30 percentage points, respectively. These findings connect the internal development of answer readiness to a practical rule for allocating reasoning computation.

Nowadays, over-the-air computation (AirComp) exploits the waveform superposition property of the wireless multiple access channel to directly compute a function of distributed data from simultaneously transmitted signals. In this paper, we propose Histogram-State Coded Multiple Access Computing (HiCoMAC), a fundamentally new convolutional-coded method for digital AirComp that directly recovers the arithmetic-sum sequence from the superposition of coded and modulated user signals. HiCoMAC exploits the permutation invariance of the sum function to represent the multi-user convolutional encoder state by the numbers of users occupying the individual encoder states. This central idea enables log-max maximum a posterior (MAP) histogram-state Viterbi decoding with substantially reduced state complexity compared to joint full-state decoding. We further introduce the computational free distance to distinguish histogram paths that produce different arithmetic-sum sequences and develop a weighted product graph for its shortest path evaluation. This distance is then used to optimize the rectangular ratio of a fixed labeled quadrature amplitude modulation (QAM) constellation to improve the computational accuracy. Simulation results show that the computation-oriented modulation design increases the computational free distance and reduces the normalized mean square error (NMSE), while HiCoMAC outperforms the considered multi-symbol digital AirComp baselines under equal transmission and energy budgets.

Spatially conditioned diffusion models can embed words and contours in natural-looking images, but vision-language models (VLMs) may fail to recognize the hidden content. Transformation-based recovery depends on parameter and view selection. To evaluate hidden-content recovery and recognition, we construct FreqBlind, a 6,000-image benchmark spanning contours, real words and non-words across three conditioning strengths. The evaluated transformation-based methods show limited recognition of contour patterns and weakly conditioned hidden content. To address this limitation, we propose ControlTrace to recover the grayscale control field used during generation. An 8.4M-parameter U-Net predicts this field from the carrier image, and a VLM then identifies its content. With Qwen2.5-VL-7B-Instruct, ControlTrace achieves 60.2% open-ended contour recognition accuracy across the three conditioning strengths, exceeding the best of the three evaluated prior methods by 26.9 percentage points. On an A100 GPU, the complete pipeline adds only 7.4 ms (5.3%) to direct VLM inference. Recovered fields have lower pixel errors and higher structural similarity than the evaluated transformation views. Across four evaluated VLMs, ControlTrace retains its overall contour recognition advantage. Recognition remains stable under the tested JPEG compression, Gaussian noise and downsampling. These results support control-field recovery for hidden-content recognition in the evaluated setting.