Paper archive

Soft-masking accelerates the convergence of Masked Diffusion Language Models (MDLMs). Existing formulations build this blend with linear interpolation (LERP) in the raw embedding space, which implicitly treats that space as Euclidean. We analyze the embedding space of MDLMs and find that the mask and predicted-token embeddings maintain a near-constant angle of (\approx 73^\circ) throughout training, while embedding norms remain essentially flat across vocabulary-frequency rank. These indicate a hyperspherical geometry, for which LERP is the wrong interpolation primitive. We introduce Spherical Soft-Masking (S-SM), a drop-in replacement that aggregates the top-(k) predictions with a Fr'echet mean on the hypersphere and blends this mean with the mask direction using spherical linear interpolation (SLERP), then restores the native mask norm. We evaluate S-SM on continued pre-training of a released 169M-parameter MDLM checkpoint across a wide range of inference-time step budgets, SLERP feedback avoids the training degradation that LERP feedback induces and delivers MAUVE gains of up to 2x over the vanilla MDLM baseline and 27.5-56.1% over TopK/LERP at various sampling budgets, alongside consistently lower generative perplexity (16.9-19.6% over the baseline), while leaving output entropy and convergence essentially unchanged.

Scaling 3D Gaussian Splatting (3DGS) to large outdoor scenes is costly in both data acquisition and computation. Adopting panoramic images with equirectangular projection (ERP) can reduce capture effort via their full $360^{\circ}$ field of view, yet the resulting omnipresent visibility invalidates existing partitioning strategies that rely on local camera frustums, causing block-wise optimization to degenerate into global training. Thus, we propose PanoLOG, a two-stage coarse-to-fine framework equipped with a Geometry and Gradient-based Partitioning Strategy tailored for large-scale panoramic 3DGS reconstruction. In the global coarse stage, PanoLOG leverages sky-sphere modeling and panoramic monocular depth supervision for reliable geometry, while in the refinement stage, G$^2$PS builds adaptive bounding volumes via parallax-driven uncertainty and assigns cameras via gradient-based importance scoring. Furthermore, we construct Pano360, the first benchmark on large-scale panoramic dataset for outdoor scene reconstruction. Extensive experiments demonstrate that G$^2$PS achieves state-of-the-art rendering quality while maintaining scalable, block-parallel training. Our models, training code, and dataset are publicly available.

In this work, we present Canvas360, a two-stage framework for in-context panoramic generation that combines geometry-aware pretraining with downstream task-specific fine-tuning. To address the lack of large-scale, high-quality training data tailored to in-context panoramic tasks, we propose Canvas360Dataset, a collection of 1M high-quality paired panoramic samples for style transfer, inpainting, outpainting, and editing, enabling effective supervision across diverse in-context generation scenarios. On the modeling side, Canvas360 enhances text-to-panorama generation through parallel depth generation, velocity circular padding, and similarity loss regularization, enabling the model to learn geometry-aware representations, capture object distortion details, and improve geometric consistency and global coherence. Furthermore, empowered by strong panoramic priors, Canvas360 enables a unified in-context panoramic generation framework that supports diverse downstream tasks via token-level concatenation, surpassing prior methods in both task coverage and modeling flexibility. Extensive experiments show that Canvas360 improves panoramic image fidelity, achieving particularly strong performance on the panorama-specific FAED metric and competitive or leading results across the reported quantitative evaluations. More information can be found on our project page: https://zry000.github.io/Canvas360/

Reasoning has become a core capability for large models, especially when reliable decisions require understanding logical consequences. Recent video generation models offer a reasoning path distinct from previous Chain-of-Thought (CoT): reasoning can unfold through temporally connected frames, known as Chain-of-Frame (CoF) reasoning. However, existing video generators are primarily trained on general video corpora, still lacking diverse supervision and dedicated designs for CoF reasoning. To address this gap, we introduce OpenCoF, a framework comprising the OpenCoF-17K dataset, a reasoning video dataset spanning 11 task families, and Wan-CoF, a fine-tuned video model for studying whether diverse temporal supervision improves CoF behavior. Across four video reasoning benchmarks, Wan-CoF achieves considerable gains over the Wan2.2-I2V-A14B baseline. Building on this, we empirically explore more advanced designs for CoF capabilities, i.e., equipping the model with visual and textual reasoning tokens. This mechanism respectively captures low-level visual cues and high-level semantic priors for spatial and temporal reasoning. Through performance comparisons and attention analysis, we examine how these tokens contribute across model depth, denoising steps, space, and time. Our results suggest that stronger video reasoning requires both broad temporal supervision and explicit mechanisms for organizing intermediate reasoning state. We open-source the dataset, model, and code to facilitate future research on reasoning-oriented video generation.

Programmable quantum simulators are beginning to access correlators of increasing complexity, ranging from four-point out-of-time-ordered correlators to even higher-order many-body correlators. The theoretical framework for interpreting such data, however, remains comparatively underdeveloped. Although a variety of higher-order correlators can be constructed straightforwardly, their physical meaning is often difficult to infer. A further complication is that different correlators are generally not independent: some may be mutually redundant, while others may encode genuinely distinct information. These features make it necessary to analyze correlators not as isolated quantities, but as a structured family. In this work, we develop a geometric framework for the collective analysis of higher-order correlator families. By representing correlators as inner products between operator words, we recast each family as a geometry in operator space. The key idea is to introduce conditioning subspaces that separate this geometry into reducible information, already explained by a chosen resolved sector, and irreducible information, encoded in the residual correlator geometry. Focusing on the latter component, we define irreducible volume profiles that quantify how broadly the unexplained part of a correlator family spreads over independent geometric directions. This perspective leads to several complementary forms of conditioning. Canonical conditioning optimally explains a correlator family. Targeted conditioning fixes the resolved sector to isolate a chosen physical feature. Krylov and cross conditioning extend the framework from a single correlator family to comparisons among correlator geometries. Our framework reveals irreducible structures hidden at the level of individual correlator values and establishes correlator geometry as a higher-level description of quantum many-body dynamics.

We introduce a novel monthly county-level consumption dataset constructed from spending data on over 350 million credit cards in the Federal Reserve's Y-14M reports, covering over 3,000 U.S. counties since 2014. We first show that the data closely approximate traditional consumption measures, explaining 92 percent of the variation in monthly adjusted personal consumption expenditures (PCE) growth at the national level and capturing meaningful cross-sectional variation in annual adjusted PCE growth at the state level. As a proof of concept, we use the county-month panel to estimate heterogeneous consumption responses to monetary policy shocks across the county-level income distribution, an analysis infeasible with traditional consumption data. We find that low-income counties exhibit larger spending declines than high-income counties, consistent with heterogeneous agent New Keynesian models. Finally, we provide practical guidance for researchers working with similar data, discussing coverage, sample composition, and the approximation of credit card spending from credit bureau data.

Score matching controls average error under the forward marginals, but a discretized reverse-time sampler evaluates the learned score along its own trajectory. We show that small forward-marginal error does not guarantee numerical stability. We construct a single smooth score field with arbitrarily small forward-marginal $L^2$ error. The learned reverse-time process is nonexplosive, has moments of every order, and can be arbitrarily close to the exact reverse-time process in path-space total variation. Yet its Euler--Maruyama discretizations converge in probability while every positive moment diverges. Thus weak convergence can hold even though every Wasserstein distance $W_p$, $p\ge1$, diverges. The same failure can occur within one fixed finite neural architecture. We construct a family of bounded, globally Lipschitz denoisers for which both the forward-marginal error and the path-space total variation distance tend to zero, while their Euler--Maruyama endpoints diverge in every $W_p$. For compactly supported data, we also give a simple positive result. Projecting the learned denoiser onto a known bounded closed convex set containing the support preserves pointwise accuracy, gives grid-uniform moment bounds, and yields Wasserstein convergence under mild local regularity. Experiments with a small fixed DiT-style network show large growth along rare numerical trajectories and its suppression by denoiser projection, while overall trajectory errors remain small.

We present MulTTiPop, a dataset of pop music segments and their associated multitrack MIDI recordings for the evaluation of automatic music transcription models. MulTTiPop contains 572 segments of popular music totaling 3.5 hours of audio, and contains songs from diverse genres and decades from the 1930s to 2000s. To collect this dataset, we perform metadata-based matching on song segments from the Lakh MIDI and TheoryTab datasets, manually identify an anchor beat between the audio and MIDI, then use beat tracking on the audio and warp the MIDI to match its tempo and timing. We evaluate state-of-the-art automatic music transcription models on MulTTiPop and find substantial room for improvement, with the best model achieving 38% Onset F1. More details and sound examples of MulTTiPop are available at https://gclef-cmu.org/multtipop.

In this study, we present a large-scale descriptive analysis of the use of an AI-based learning assistant (Syntea) in higher education. Based on objective log data from 77,543 students enrolled in distance studies, we examine usage patterns across gender, age group, study cluster, degree, and study mode. To date, existing research on educational chatbots has largely relied on comparatively small samples and self-reported survey data, while large-scale evidence on actual usage behavior remains limited. Our findings show that Syntea is already embedded in the study routines of many learners, but that usage differs across demographic and structural contexts. By identifying these patterns, our study provides an empirical basis for the further development of AI-based learning support and contributes a large-scale analysis of educational chatbot usage in higher education.

Large language model (LLM) applications increasingly use explicit workflows for tool use, retrieval, branching, checkpointing, and human approval. Existing workflow systems already address many execution concerns. This paper proposes a Lisp-inspired but language-independent conceptual model: symbolic forms, object identity, and live-image thinking are used as explanatory lenses, not implementation commitments. In this model, workflow definitions, workflow instances, inference records, context snapshots, and dependency relations are represented as persistent knowledge objects in a shared knowledge substrate. Its central semantic distinction is between derive and infer: derive is deterministic computation over available state; infer is mediated LLM judgment under declared context and executor-controlled capability policy. The result is a preliminary conceptual account of semantic persistence: workflows do not merely produce knowledge and leave traces, but can themselves be represented as inspectable, resumable, and reviewable knowledge objects, while formal transition semantics remain future work.

The black hole information paradox has motivated extensive study of how and when information escapes from evaporating black holes. Here we address the analogous question for cosmological horizons: when does an individual Hawking pair begin to carry information out of a de Sitter horizon? We study this problem in a class of two-dimensional flow geometries that interpolate smoothly between an asymptotic AdS$ _2 $ boundary and a dS$ _2 $ static patch. Modeling the emission of a Hawking pair via a probe state constructed from local operators and their modular conjugates, we promote the centaur algebra of observables to a Type II$ _\infty $ factor through the crossed-product construction. This allows us to compute the entropy difference between the thermofield-double reference state and the Hawking-pair state. We find that this difference traces a characteristic mini-Page curve for the cosmological horizon: it starts near zero, reaches a minimum near $ τ\approx β/8 $ before increasing again. We interpret the location of the minimum as the time at which quantum information begins to escape the cosmological horizon. Extending the analysis to the microcanonical ensemble, we show that the algebraic entropy coincides with the generalized entropy of an entanglement wedge cut that tracks the emitted particle along the horizon. Furthermore, the relative modular flow generated between the two states yields a Lyapunov exponent $λ=2π/β$, identifying the scrambling time as the scale at which the information carried by the pair becomes accessible to a static-patch observer.

A national language model offers a linguistic community its own instrument for measuring what its citizens say and value. Portugal's AMALIA, a publicly funded 9B-parameter model for European Portuguese, appears competitive on agreement alone: asked to code the moral foundation of authority, it agrees with trained human coders to within six F1 points of open models eight to thirteen times its size. Yet agreement is reliability, not validity. For theoretical constructs that must be inferred rather than read from surface features, the question is whether the model follows the construct's theory or reaches the right code by correlated shortcuts. We test this with the recovery gap: the loss in performance when a holistic prompt is decomposed into the codebook's atomic clauses and recombined by the theory's explicit rule. If calibration closes that gap, some portability should survive across models and languages; where it does not, the construct-model instrument is the likely locus of failure. We ask whether a calibrated English instrument transfers to AMALIA-9B and to European Portuguese. For one construct and one corpus, it does not. Decomposition recovers only about half of AMALIA's holistic performance, and error analysis suggests reliance on surface correlates, especially moral outrage near authority figures. An open multilingual LLM closes the gap on the same Portuguese corpus under the same instructions, pointing away from the corpus as the main explanation. AMALIA can still screen and pre-code at scale, but it cannot yet measure this construct well enough to stand alone. The study is a single counterexample, not a verdict on national models; it argues that sovereign-LLM benchmark batteries should test not only agreement with human coders, but the evidential route by which that agreement is warranted.

Recent progress in 3D human pose estimation has made markerless recovery of skeletal motion increasingly accurate and scalable. However, most pose estimators remain optimized for geometric keypoint accuracy, while many real-world applications in rehabilitation, sports science, ergonomics, and clinical movement analysis require biomechanical quantities that describe how the body moves, loads, and activates. In this work, we propose BioModule, a lightweight plug-in temporal transformer that attaches downstream of any 3D pose estimator and predicts biomechanical attributes from standard 17-joint 3D skeletons. BioModule is estimator-agnostic and requires no modification of the upstream pose model, enabling existing pose estimators to be extended toward physically interpretable motion analysis. To train and evaluate BioModule, we construct a large-scale aligned dataset pairing Human3.6M video and 3D keypoints with the biomechanical label space of Human3.6Mplus. We establish and verify anatomical correspondence between coordinate systems of the two datasets, enabling frame-accurate cross-modal supervision. Using this aligned supervision, BioModule predicts biomechanical quantities. We further benchmark BioModule across seven state-of-the-art 3D pose estimators, providing the first systematic analysis of how upstream pose estimation quality propagates to downstream biomechanical prediction fidelity. The results position BioModule as a compact, modular bridge between vision-based pose estimation and biomechanically meaningful human motion analysis.

Negative and pathologically large Monte Carlo event weights strain the computing budgets of experiments at the Large Hadron Collider. Cell resampling algorithms locally redistribute event weights among nearby events in a metric space. We study the performance of metrics defined in terms of Optimal Transport, namely the Energy Mover's Distance and a spectral variant, in the context of such algorithms. As these metrics are insensitive to the addition of soft and collinear radiation, they may be applied directly to particles at any stage of event generation. When applied to samples simulated at next-to-leading-order in quantum chromodynamics, this approach reduces the observed bias relative to other cell resampling techniques presented in the literature. We also study the Cross-Section Mover's Distance as an unbinned, broadly-applicable figure of merit for quantifying the bias introduced by any full-phase-space reweighting.

N-of-1 trials are patient centered randomized controlled trials. Although the primary goal of N-of-1 trials is to obtain the results for each patient separately, pooling the results across patients also has relevance. In this paper, we present two analytical strategies to pool the results across N-of-1 trials, when the main outcome for each patient is a binomial variable. Our first method takes into account the extra-binomial variation, while as the second approach takes into account hierarchical clustering in addition to overdispersion. We illustrate the methods using real data analysis and compare the methods using simulations.

We study the functional inequality \[ f(r+s)\le g(r)+αf(s) \quad(r,s>0). \] Here $g:(0,\infty)\to[0,\infty)$ is a given decreasing function, $α$ is a constant such that $0<α<1$, and the problem is to determine whether the family of decreasing functions $f:(0,\infty)\to[0,\infty)$ that satisfy this inequality is bounded above by some finite function on $(0,\infty)$ and, if so, to find bounds for this function. We present a solution to this problem, and use it to give a new proof of a theorem of Domar on the uniform boundedness of certain families of subharmonic functions, in addition obtaining explicit bounds.

Asthma affects over 260 million people worldwide, yet diagnosis remains dependent on spirometry and specialist assessment, limiting accessibility in primary care and low-resource settings. Vocal biomarkers offer a promising non-invasive alternative, but prior studies have largely focused on acoustic features without integrating clinical context. We present a multimodal Mixture-of-Experts framework for asthma detection that adaptively combines acoustic embeddings from sustained vowel phonation and reading passage tasks with structured clinical and demographic data. The model was evaluated on a matched cohort of 1,218 asthma cases and healthy controls from the Colive Voice study. The multimodal model achieved an AUROC of 0.85 and Brier score of 0.17, outperforming unimodal and bimodal approaches. Adaptive gating analysis revealed increased reliance on audio features in participants with greater respiratory symptom burden, whereas clinical features contributed more strongly in less symptomatic individuals. These findings support scalable and explainable asthma screening using smartphone-collected voice recordings.

Gilbreath's conjecture asserts that if one starts with the sequence of primes and takes successive absolute differences to create a triangular array, then the left diagonal of this array consists entirely of ones after the first row. In this paper, we show that the analogue of this conjecture for a Cramér random model holds, in which the (normalized) prime gaps are replaced by independent random variables with geometric distributions of logarithmic size. We also give some preliminary analysis of the associated continuous probabilistic model for this problem, as well as a deterministic "inverse theorem" that isolates the specific obstructions to Gilbreath's conjecture (assuming a Cramér type bound on prime gaps), namely long blocks of zeroes, or very long shallow $\{0,d\}$-valued blocks for some $d \geq 2$.

Accurate 3D terrain maps are essential for emergency response when assessing wildfire hazards. However, wildfire-prone regions often span vast areas where conventional reconstruction methods underperform. Airborne LiDAR systems provide high-resolution terrain data, but they are expensive and infrequently updated. Image-based methods offer a lower-cost alternative, but struggle due to sparse visual features and limited image overlap. We propose a multi-modal reconstruction framework leveraging outdated Digital Elevation Models (DEMs) as geometric priors for image-based 3D reconstruction. Our key innovation is physics-based pixel-pixel alignment between images and DEM data, dramatically reducing computational complexity by eliminating expensive feature matching procedures. To validate our approach, we developed a large-terrain simulator based on a real wildfire-prone area, generating realistic images enabling a comprehensive evaluation. Given posed images and legacy DEMs, our method produces high-fidelity depth maps while maintaining real-time performance. We find significant improvements in reconstruction accuracy and computational efficiency over existing techniques, offering a scalable solution for wildfire response.

Quantum key distribution (QKD) brings the promise of communication with information-theoretic security but is limited in practice due to its susceptibility to noise, losses, and device imperfections. To address these challenges, we propose a robust high-dimensional (HD) one-sided device-independent QKD (1sDI-QKD) protocol and present a proof-of-principle experimental implementation using photons entangled in the transverse-spatial degree-of-freedom. We develop a systematic security analysis of HD 1sDI-QKD protocols, leveraging quantum steering to certify security, and evaluate achievable secret key rates for different measurement configurations and system dimensions using reverse reconciliation. Our analysis shows that increasing the dimension enhances robustness against both noise and loss. We then demonstrate the key experimental building blocks required for implementing the protocol: (a) a high-quality source of high-dimensional photonic entanglement, and (b) a fully programmable, high-dimensional multi-outcome measurement device operating in up to dimension 11. Using these components, we obtain positive key rates for all investigated dimensions under the fair-sampling assumption, with the highest key rates achieved for dimension d=7. Finally, we discuss the steps required for a practical, loophole-free implementation of 1sDI-QKD in realistic regimes of loss and noise.

We perform a semiclassical, large-spin S, analysis of the quantum approximate optimization algorithm (QAOA) on the Sherrington-Kirkpatrick (SK) model, using the truncated Wigner approximation. Fixing the QAOA angles to their previously determined optimal S=1/2 values, we observe a non-monotonic dependence of the final energy on the spin. At small S the semiclassics is dominated by noise, while the large-S limit is constrained by the exponential growth of the initial fluctuations. For a depth-p QAOA one achieves the optimal balance at S of order p, resulting in a convergence of the final energy to the Parisi value like log(p)/p. We find that the semiclassics slightly outperforms the true spin-1/2 QAOA, and thus suggest they both converge to the Parisi value in the same way. Finally, removing all the initial noise, and re-optimizing the parameters to account for that change, results in superior performance with 1/p convergence.

To best reproduce observed multiply-lensed lensed images, lens models usually incorporate shear attributed to objects unrelated to the lensing galaxy (i.e., external shear): whether it be neighbouring galaxies not explicitly included in the lens model or other cosmic structures along the sightline. When constrained solely by the positions of image counterparts, such lens models, even those utilising simple ellipsoidal mass distributions, can satisfactorily -- if not near perfectly -- reproduce the observed image positions, but often leave significant differences in flux ratios between the predicted and observed images. For the narrow-line regions (NLRs) of quasars, which are too large to be affected by micro-lensing from stars in the lensing galaxy, the flux ratio anomalies thus left are commonly attributed to small-scale structures (sub-structures) in Dark Matter associated with the lensing galaxy. Here, we show that external shear can always resolve, among the quadruply-lensed quasar NLRs studied, position anomalies in lens models constrained solely by the observed image positions, and in addition reduce although not fully resolving flux ratio anomalies when constrained by both the observed image positions and flux ratios -- provided, usually, that the external shear incorporated have strengths that far exceed (as is the common practise) those typically inferred from weak lensing along general sightlines (i.e., cosmic shear). Our work highlights the subversive role of excessive external shear in concealing lensing anomalies, undermining inferences on the characteristics of Dark Matter sub-structures -- and, correspondingly, the nature (mass and temperature) of the Dark Matter particle -- when not sensibly incorporated into lens models.

The quadratic cost of causal self-attention severely bottlenecks long-context transformer inference. While numerous post hoc linearization pipelines exist, it is difficult to identify which components preserve model quality. This work isolates the effect of state update design in a strict frozen-backbone regime. We show that softmax relies on key-dependent, rank-1 orthogonal projections, elucidating why delta-style networks outperform purely gated accumulation. We identify a potential source of approximation errors and introduce structural interventions, specifically sink tokens, short convolutions, and fixed-budget cache routing, which reduces the remaining gap. We scale this linearization approach across LLaMA and Qwen models up to 32B parameters, outperforming prior post hoc baselines on MMLU and matching the long-context retrieval of complex adaptive-caching frameworks.

The optimization of long-horizon agents increasingly relies on reflection-based mechanisms, where a large language model (LLM) acts as an optimizer to diagnose agent failures and improve agent policies. However, real execution traces are difficult to use directly for optimization: large trace collections are often redundant and heterogeneous, making optimization inefficient and prone to overfitting to low-value failures; meanwhile, each individual trajectory also contains many irrelevant steps, while naive context reduction methods such as truncation or sliding windows can discard causally important evidence and produce misleading optimization signals. To resolve this dilemma, we introduce STRACE (Structural TRajectory Analysis and Causal Extraction), a framework that constructs high signal-noise optimization contexts for more precise and effective optimization. At the batch level, STRACE mines failure patterns to filter redundant traces and retain representative failures; within each selected trace, it performs causal localization over a textual dependency graph to remove non-causal steps and identify the true root-cause module for optimization. Empirical results demonstrate that STRACE significantly outperforms standard context-filtering baselines. Notably, on a challenging formal verification task (VeruSAGE-Bench), it successfully optimizes human-expert designed agents, delivering $1.4\times$ success-rate improvement (42.5% to 58.5%). The code is available at https://github.com/moomight/STRACE .

The spectral condition number is a widely adopted measure of worst-case cost for quantum linear system solvers. Yet it can significantly overestimate the actual runtime for a typical problem instance. We present two quantum algorithms that produce the normalized solution $|x\rangle$ of linear system $Ax=| b \rangle$ to accuracy $ε$ with complexity independent of the condition number $κ=\lVert A^{-1}\rVert$. We focus on the standard input model where $A$ is accessed through a block encoding and $| b \rangle$ is prepared by a unitary. But we also introduce an affine dilation model that encodes $A$ and $| b \rangle$ jointly, allowing further refinements of the query complexity. Our truncation-based solver makes an optimal number of queries to $| b \rangle$ and $\operatorname{\mathbf{O}}\left(κ_{\mathrm{eff}}\operatorname{polylog}\left(\frac{κ_{\mathrm{eff}}}ε\right)\right)$ queries to $A$. We prove a family of upper bounds on the effective condition number, including $κ_{\mathrm{eff}}\leq\frac{\lVert(A^\dagger A)^{-t/2}|x\rangle\rVert^{1/t}}{ε^{1/t}}$ for positive even integer $t$ and $κ_{\mathrm{eff}}\leq\frac{\lVert A^{-1\dagger}(A^\dagger A)^{-(t-1)/2}|x\rangle\rVert^{1/t}}{ε^{1/t}}$ for positive odd $t$, overcoming the $κ$-barrier. Our filtering-based solver is extremely simple with a favorable runtime prefactor. In particular, the solver has query complexity $6\frac{\lVert A^{-1\dagger}|x\rangle\rVert}ε\ln\left(\frac{1}ε\right)$ to leading order when the solution norm is known. We then present a similarly simple solution norm estimator with the same asymptotic cost up to logarithmic factors. Our quantum linear system solvers thus substantially improve a recent algorithm of Li, enabling faster quantum linear system solving beyond the condition number.

We address the circular motion Unruh effect in 2+1 spacetime dimensions, as probed by a pointlike Unruh-DeWitt detector coupled to a massless scalar field. The effective temperature due to circular acceleration, operationally defined in terms of the detector's excitation and de-excitation probabilities, is known to be much smaller than the linear acceleration Unruh temperature when the detector's energy gap is small and the interaction lasts for a long time. It was shown by Parry et al. [Class. Quant. Grav. 42, 245012 (2025), arXiv:2508.19987] that a temperature of the order of the linear acceleration Unruh temperature can nevertheless be recovered in a simultaneous long-time-small-gap double limit, using suitable classes of detector-field couplings described by asymptotically scaled switching families (ASSFs). The successful constructions presented there required the coupling to change sign. Here we prove, within the ASSF framework and under certain technical boundedness and localisation conditions, that sign changes in the detector-field coupling are in fact *necessary* for obtaining a nonvanishing limiting effective temperature. Our analysis is motivated by current work towards an experimental verification of the circular motion Unruh effect in analogue spacetime experiments.

In magnetically ordered materials, magnetic field and temperature variations modify the magnetic texture through their coupling to the local energy landscape, imprinting distinct fingerprints in the resulting magnetic domain patterns. Retrieving these conditions from the pattern remains challenging, as stochastic nucleation and hysteresis produce a nonlinear, multivariate, and ambiguous relationship between magnetic domain morphology and external stimuli. To decode these fingerprints, we designed a controlled magneto-optical inference experiment that reconstructs magnetic field, temperature, and magnetic history from a single fine-scale, high-contrast, pixel-resolved optical polarization map of feature-rich magnetic domain textures in a bismuth-substituted yttrium iron garnet film. Deep convolutional neural networks are complemented by feature-based neural-network inference using hand-crafted, physically interpretable descriptors of measured magneto-optical image data, linking the decoded information to material-dependent features and exploring their contributions. Together, these results establish magnetic texture as a high-fidelity record of external conditions enabling accurate single image multiparametric sensing and paving the way for data-driven explorations of complex magnetic states. Uncovering the physically interpretable features that encode this record sheds new light on the physics of magnetic domain formation.

Electrocardiography (ECG) is one of the most widely used tests for diagnosing cardiovascular disease. Yet several remote clinics still utilize paper ECG printouts for their analysis due to limited connectivity and computational capacity. As a result, vast numbers of physical ECGs obtained in remote areas still remain incapable of being accessed by contemporary artificial-intelligence (AI)-based decision support as they require high computational resources or strong high-speed internet connectivity. This causes several cases where conditions like acute coronary occlusion (ACS) is overlooked and reperfusion therapy delayed. Although prior work has tackled digitization and diagnosis separately, and utilized advanced AI models for them, there still remains a lack of a compute-light, on-device framework that reconstructs paper ECGs at high fidelity, while accurately supporting multiple clinically relevant endpoints. We address this need with an end-to-end lightweight on-device digitization-to-diagnosis pipeline that converts a smartphone photo or scan of a paper ECG into a calibrated 12-lead signal and screens for Myocardial Infarction (MI) pathologies, with SHapley Additive exPlanations (SHAP) to support interpretability. Trained and evaluated on 21,799 ECGs from the PTB-XL dataset and further validated on hospital-acquired ECG-Matrix dataset, the complete system runs in <30 s per ECG on CPU-only resources, achieving 95.51% accuracy (F1 = 0.9519) for MI detection on PTB-XL and 88.89% accuracy (F1 = 0.8862) for OMI detection on ECG-Matrix. This work showcases that legacy paper records can be reliably democratized in any part of the world, providing a scalable decision support when digital ECG export, connectivity, or high-end compute are unavailable

Medicine is inherently multimodal, requiring clinicians to synthesize information across diverse data streams. Yet the development of multimodal foundation models is constrained by limited access to large-scale, high-quality clinical data. Although PubMed Central (PMC) offers a complementary source of expert-authored image-text data, existing PMC-derived resources remain limited in fidelity, reproducibility, and clinical validation. We introduce MedPMC, an automated, continuously updatable framework that transforms permissively licensed literature into high-fidelity infrastructure for medical multimodal models. Applied to 6.1 million PMC articles, MedPMC curated 11 million medical image-text pairs. Component evaluations showed strong performance for initial screening (F1 = 93.2), multi-panel figure detection (F1 = 96.5), figure separation (mAP = 89.8), caption separation and alignment (F1 = 81.4; ROUGE-L = 85.3), and medical figure classification (F1 = 96.5). Manual review by five annotators, three with medical training, found 95.3% of MedPMC images medically relevant, versus 19.7% in a prior PMC-derived dataset. Across 26 benchmarks spanning 11 specialties, a MedPMC-trained CLIP-style model improved average zero-shot AUC by 7.1 percentage points over the strongest architecture-matched biomedical CLIP baseline despite using fewer than half as many image-text pairs. As the vision encoder in a multimodal large language model, it improved medical visual question-answering by 1.9 and 16.9 percentage points across two benchmarks. In 10,524 Yale New Haven Health System dermatology photographs, it improved morphology-to-image retrieval Recall@5 by 11.7 percentage points. These findings show that high-fidelity literature curation strengthens medical multimodal foundation models across benchmark and clinical settings. We publicly release the framework, corpus, benchmarks, and pretrained models.

We study first-order locally convex Lie algebroids in the setting of Bastiani calculus. The first-order condition is automatic in finite dimensions, but is an additional regularity hypothesis for general locally convex vector bundles. Under this condition, we define sheaves of scalar-valued and vector-valued Lie algebroid forms as fiberwise continuous alternating maps with smooth local representatives. We define morphisms by requiring the induced pullback on inverse-image sheaves of scalar-valued forms to commute with the Lie algebroid differentials, and prove that first-order locally convex Lie algebroids form a category. We also study representations and the induced cohomology sheaves. We show that locally convex Lie groupoids have first-order Lie algebroids and that Lie groupoid morphisms induce morphisms in this category. As applications, we prove that the current algebroid associated with a first-order Banach Lie algebroid is a first-order Fréchet Lie algebroid, and we prove a Lie II theorem in the Banach setting: first-order morphisms between the Lie algebroids of Banach-Lie groupoids under source-connected and source-simply connected hypotheses integrate to unique Lie groupoid morphisms.

Large language models increasingly \emph{understand} dialectal English, yet still \emph{produce} only standard, US-leaning English, leaving dialectal generation, the harder half of the problem, largely unaddressed. We introduce \textbf{DiaLLM}, which continually pretrains three open-weight language model families on the International Corpus of English and applies implicit and explicit post-training paradigms, each combined with three model alignment strategies, giving the first controlled comparison of these components across Australian, Indian, and Northern British English. Our results reveal that dialectal robustness and generation are \emph{dissociated}: benchmarks are shaped by continual pretraining and SFT, while alignment visibly reshapes generation in ways benchmarks do not capture. Explicit variety-targeted adaptation produces output reliably recognised as dialectal and preferred over broad alignment, yet the method that most aggressively optimises the dialectal reward is not preferred by human evaluators. Independent linguistic analysis corroborates this reward-quality gap, most clearly on two of the three families. No single alignment method dominates, and closing the gap will require richer reward designs and continued investment in dialectal resources. We release all code, checkpoints, and preference datasets.

Classifier-free guidance (CFG) is the standard way to strengthen class-conditioning in diffusion and flow-matching samplers, yet at large guidance it oversaturates and destabilizes, symptoms practitioners suppress with more steps or limited-interval schedules. We analyze CFG through an asymptotic-preserving, numerical-analysis lens. Building on a recent result that the deterministic DDIM step is the unique fitted operator for the unguided terminal layer, exact on the final small-sigma stretch of sampling, we show that guidance re-stiffens exactly the discriminative subspace to an anomalous exponent 1+w. DDIM is therefore no longer fitted there, and on coarse meshes its guided residual diverges as sigma_min goes to zero. We prove a guided clock barrier with three ordered step-size thresholds, and read one-step oversaturation as its endpoint: a solver artifact on the calibration model rather than the continuous guided law. The same analysis yields a one-coefficient, zero-extra-NFE repair: replace CFG's w(r-1) by r^(1+w)-r on the guidance direction. On the calibration model's discriminative crossover, this removes CFG's sigma_min-divergent blow-up and is first-order accurate against the exact guided flow as sigma_min goes to zero. On learned CIFAR-10 checkpoints, and as a cross-domain smoke test on Stable Diffusion 1.5 DDIM, it acts as a high-guidance stabilizer at no extra cost rather than a universal quality knob: it cuts residual amplification and saturation, gives 9/9 point-FID wins over CFG on the tested grid, and preserves classifier-proxy target accuracy in the hard-cell blocks. We report the limits alongside: it is not a universal image-quality win, and against a dense vanilla-CFG reference it is not a uniformly better integrator of that field.

Hadronic vacuum polarization (hVP) enters simple atomic systems at a level that is small yet decisive for the precision spectroscopy now underway. We evaluate the hVP contributions to the Lamb shift and the hyperfine splitting (HFS) in ordinary and muonic hydrogen (H and $μ$H) and hydrogen-like helium-3 ions ($^3$He$^+$ and $μ^3$He$^+$), using the dispersive data-driven approach and state-of-the-art empirical parametrizations of the $R$ ratio. At the centre of the analysis is the interplay of recoil and finite-size effects: the recoil corrections that dominate the HFS in muonium (Mu), where both constituents are pointlike, are shown to be suppressed by the nuclear elastic form factors (FFs). Our results for the leading hVP contribution to the Lamb shift agree with the literature within uncertainties. Furthermore, we present a first evaluation of the subleading $O(Z^5α^6)$ hVP-finite-size correction, which is by no means negligible in $μ^3$He$^+$. Our results for the hVP contribution to the HFS deviate significantly from all previous evaluations. For the ground-state HFS, we obtain $2.153(11)~μ$eV in $μ$H and $-15.19(57)~μ$eV in $μ^3$He$^+$, as well as $0.0860(4)~$kHz and $-0.476(17)~$kHz in ordinary H and $^3$He$^+$, respectively. Notably, our result for $μ$H differs from previous evaluations by roughly ten times the experimental precision anticipated by the upcoming CREMA and FAMU measurements.

Surface plasmon polaritons (SPPs) and magnetic polaritons (MPs) are fundamental resonance modes that are widely used to tailor the thermal radiation properties of micro/nanostructured metamaterials. Lumped circuit models (LCMs) are usually constructed empirically to describe the MP resonance conditions, and different LCMs have to be constructed for different orders of MPs, but these are difficult to be built for high-order MP modes due to the complex electromagnetic field distribution. This work proposes a new type of circuit model, distributed circuit model (DCM), to describe and predict multi-order MP resonances inside the structure based on the minimum total impedance condition. This allows both fundamental and high-order MP resonances to be predicted with a unified circuit, significantly simplifying the analysis of high-order MPs. More importantly, the DCM shares a similar and clear physical picture as the LCM for describing MPs. The MP resonance conditions for four typical structures are derived. Theoretical predictions based on DCMs are compared with and validated by rigorous numerical simulations. This study deepens the understanding and facilitates the design of MP-based thermal radiation metamaterials.

Single-cell molecular characterization remains a bottleneck in scalable biological analysis because of labeling requirements, limited multiplexing, and reagents that perturb physiology. Raman spectroscopy addresses these limits by providing chemically specific, label-free vibrational fingerprints, but long acquisition times and specialized instruments restrict high-throughput use. Here, we overcome this barrier by showing that spectral fingerprints can be reconstructed from brightfield microscopy using generative modeling. We introduce Pic2Spec, a framework that learns a shared latent biochemical representation linking image morphology to vibrational spectral structure, enabling virtual Raman spectroscopy without hardware. We validate Pic2Spec across mammalian and bacterial cells, generating high-fidelity spectra that reproduce measured Raman fingerprints with 98% cosine similarity and Pearson correlations of ~95%, while preserving biochemical peaks and population distributions. Beyond spectral similarity, Pic2Spec provides molecular-level resolution in bacterial systems: generated spectra discriminate mutation-driven transgenic states and predict GFP expression with accuracy approaching true Raman measurements, outperforming conventional image analysis by 20%. These findings establish Pic2Spec as a first demonstration of chemically informative virtual molecular fingerprinting from brightfield images, complementing slow, hardware-intensive spectroscopy with computational inference. By redefining microscopy as an inference-enabled molecular profiling platform, Pic2Spec democratizes label-free biochemical phenotyping and overcomes the hardware and time constraints that have confined spectroscopy to specialized laboratories. This enables high-throughput molecular analysis for clinical diagnostics, screening, and monitoring at the scale and accessibility of standard microscopy.

Does RL post-training merely amplify primitive skills already latent in a base model, or can it compose primitive skills into new higher-level strategies? We study this question in a fully observable rewrite-grammar environment where the pretraining distribution is known and every generated rewrite can be audited. A Transformer is pretrained on primitive symbol-rewrite chains and post-trained on a Trace-based reasoning task with only a binary final-answer reward. RL solves held-out problems that remain rarely solved by the pretrained model even under much larger sampling budgets, while rejection fine-tuning improves early but plateaus. Trace analysis shows that RL reorganizes primitive competence through a phased compositional mechanism: it first strengthens primitive reductions, then discovers valid composed procedures. These include sequential compositions, which collapse ordered chains of primitive contractions, and parallel compositions, which combine independent primitive contractions in a single step. The composed procedures are not isolated samples; they are reused and consolidated into a stable repertoire. Comparing RL with rejection fine-tuning shows that the key difference is not exploration volume but selectivity: RFT produces many shortcut-like rewrites, much of them invalid, whereas RL concentrates exploration into valid reusable structure. Pretraining ablations show that the emergence of compositional strategies is gated not by primitive exposure alone, but by whether pretraining organizes primitive competence into reduction procedures that RL can later compress. The base model provides weak procedural ingredients; RL builds them into reliable higher-level strategies.

We prove rates of decay at infinity for solutions to variable-coefficient Schrödinger equations of the form $-\text{div}(A \nabla u) + W \cdot \nabla u + V u = λu$ in cylinders, $\mathbb{T}^d \times \mathbb{R}^m$. We assume that $W$ and $V$ are bounded and that $λ\in \mathbb{C}$. Our rates depend on the decay of $|\nabla A|$ at infinity. In particular, we prove a range of quantitative unique continuation-type results at infinity when $|\nabla A(θ, x)| \le C (1 + |x|)^{-τ}$ for $τ\in [0,1]$. By adapting the methods in [KLP25], we construct explicit solutions to demonstrate the sharpness of our estimates for each such $τ$.

This work presents a novel approach for adapting neural network architecture along the depth based on a posteriori error estimation. By formulating neural network training as a continuous-time optimal control problem, we derive rigorous error estimates that quantify how approximation error distributes across network layers. This error decomposition enables a principled depth adaptation strategy: new layers are inserted at locations of maximum estimated error, allowing the network to efficiently capture complex, nonlinear variations in the underlying problem. Our framework introduces a novel network architecture that treats weights and biases as piecewise linear functions varying across layers, with the error estimator bounding the discrepancy between this discrete representation and the true continuous optimal control solution. The approach leverages dual weighted residual methodology from finite element analysis to derive computable upper bounds on the functional error. A key theoretical contribution is the derivation of explicit error bounds that decompose the total approximation error into interval-wise contributions, providing a rigorous basis for targeted architecture refinement. We demonstrate the effectiveness of our method on scientific datasets, including learning the observable-to-parameter map for the Navier-Stokes equation. Numerical results reveal that our approach consistently outperforms existing architecture adaptation methods in terms of generalization performance.

Time series analysis plays a vital role across a wide range of scientific and engineering domains but poses substantial computational challenges. A major difficulty arises from the time reparameterization invariance of time series data, which complicates the extraction of meaningful temporal features. In this work, we address the problem of time series classification by exploring the application of quantum computation techniques. We propose a hybrid quantum-classical architecture that integrates recent advances in quantum neural networks with the mathematical framework of path signatures, mitigating the impact of time reparametrization invariance. The architecture employs feature layers that compute a signature kernel between pairs of input paths, consisting of a reference path and a target path for classification, using either classical or quantum variational linear solvers (VQLS). These feature layers are followed by a Quantum Convolutional Neural Network (QCNN) to perform downstream learning tasks. We evaluate several realizations of the proposed architecture, differing in QCNN configurations, on a binary classification task involving time series representations of handwritten digits. Our experiments demonstrate the potential advantages of implementing path signature kernel layers within quantum circuits and provide an analysis of the computational limitations associated with the VQLS component.

Existing RLC circuit model leads to inaccurate predictions of the quality factor (Q-factor) of magnetic polariton (MP) resonances under imperfect absorption conditions due to the omission of radiation loss. Moreover, the lumped-parameter nature of RLC models also limits their applicability for predicting higher-order MP modes. In this letter, we propose a distributed circuit model (DCM) for predicting the Q-factor of MP resonances, which overcomes these limitations. By introducing a radiation resistance to characterize radiation loss and establishing a mapping between distributed and lumped parameters, we derive a unified analytical expression for the Q-factor of arbitrary-order MPs. Validation via rigorous coupled-wave analysis (RCWA) demonstrates that this model can accurately predict the Q-factors of MPs of all orders in various metal-insulator-metal (MIM) structures. This work provides a simple yet effective tool for designing metamaterial emitters/absorbers and advances the understanding of MP loss mechanisms.

This work investigates a Hénon-type equation with the cap-shaped weight $V_α(|x|)=(4|x|(1-|x|))^α$, which concentrates on the sphere of radius $1/2$ as $α\to\infty$. Particular attention is devoted to ground state solutions and the phenomenon of symmetry breaking in the large-$α$ regime. A complete description of the asymptotic behavior of ground state radial solutions is obtained as $α\to\infty$. The analytical results are further supported by numerical simulations of the radial ground states.

Masked diffusion language models (MDLMs) generate text by iteratively unmasking tokens, but their standard decoder reduces each step to a binary action: a position is either committed to a single token or left fully masked, discarding rich predictive information rather than carrying it forward, and forcing premature, irrevocable commitments that lead to poor performance under a limited decoding budget. In this paper, we reinterpret mask prediction as a clean-state prediction ($x$-prediction) and show that it can be used to induce a continuous flow in the input embedding space. Building on this view, we propose a continuous decoding framework for MDLMs where tokens can accumulate partial progress at each diffusion step and remain revisable. To match the uneven contextual constraints across positions in language, we replace the globally synchronous schedule in image diffusion with a confidence-based asynchronous update in which the diffusion progress is token-wise accumulated. Additionally, we introduce a lightweight policy network and formulate its training as a reinforcement learning problem. Applied to pretrained LLaDA, our decoder retains 83--97% of full-budget accuracy using under 15% of the diffusion steps, largely outperforming discrete mask-prediction decoding at matched budgets.

In this work, we focus on extending SHARP, the popular photorealistic view synthesis method, for universal monocular rendering across a continuum of camera systems, from conventional perspective cameras to wide-field-of-view, fisheye and omnidirectional panoramic settings. To overcome the pinhole-specific assumptions of SHARP, our key idea is to align various images in a unified omnidirectional latent space. Thus, we propose UniSHARP, which performs implicit alignment in both feature and Gaussian spaces. Specifically, Gaussian primitives are arranged along rays and radial distances in a ray-based universal representation, while 2D semantic and 3D spatial features extracted from UniK3D-inspired encoders are jointly decoded to generate the complete Gaussian cloud. To comprehensively evaluate our method, we construct a benchmark covering diverse imaging systems across various scenes. The benchmark is further stratified by field of view (FoV) to enable fine-grained assessment of the universal monocular rendering task. Extensive experiments on the proposed benchmark demonstrate the effectiveness of UniSHARP, outperforming alternative methods by a large margin. The project page can be found at: https://insta360-research-team.github.io/Unisharp-website/

Current Vision-Language Models struggle with hours-long videos because processing full-length visual sequences induces prohibitive token explosion and attention dilution. To overcome this, we introduce MemDreamer to decouple perception and reasoning, shifting long-video understanding into an agentic exploration process. As a plug-and-play framework, it incrementally streams videos to construct a Hierarchical Graph Memory, a top-down three-tier architecture for semantic abstraction, anchored by a foundational graph capturing spatiotemporal and causal relations. During inference, the reasoning model employs agentic tool-augmented retrieval, navigating hierarchies, searching nodes, and traversing logical edges via an Observation-Reason-Action loop. Experiments show MemDreamer achieves SOTA results across four mainstream benchmarks, narrowing the gap with human experts to only 3.7 points. It constrains the reasoning context window to merely 2% of full-context ingestion while delivering a 12.5 point absolute accuracy gain. Furthermore, statistical analysis uncovers a strong positive linear correlation between an VLM's performance on logic reasoning and long-video understanding benchmarks, establishing agentic capability scaling as a new paradigm for multimodal comprehension.

Despite their large abundance, it is still unknown whether and under what conditions rocky planets around M dwarf stars can host atmospheres. This open question motivated the on-going Rocky Worlds DDT survey focused on searching for atmospheres on relatively low-temperature rocky exoplanets by systematically probing for the presence of day-night heat redistribution and CO2 absorption through JWST/MIRI 15 $μ$m eclipse observations. Here we present the analysis of the first full data set from this survey, consisting of four observations of the warm Earth-size exoplanet GJ 3929 b, with a planetary mass of 1.75+0.44-0.45 M$_\oplus$ and instellation flux of 17.3+/-0.7 S$_\oplus$. In our analysis, we include two previously unpublished eclipse observations and find an overall eclipse depth of 118+/-22 ppm and a dayside surface brightness temperature of 641+59-64 K. This is marginally lower than the eclipse depth of 160+26-27 ppm previously reported based on only the first two observations. While the full data set remains consistent with bare rock scenarios, it also leaves more room for thin atmosphere scenarios. Only thick CO2 atmospheres without thermal inversion remain ruled out at greater than 3$σ$. We also continue with lessons-learned in robustly analyzing these kind of high-precision JWST/MIRI 15 $μ$m eclipse observations. Notably, we find that the Frame Normalized Principal Component Analysis (FN-PCA) method appears more robust against the choice of extraction aperture size, which otherwise can have a significant impact on the inferred eclipse depth and scientific conclusions when using a standard polynomial baseline detrending method.

We extend the study of boundary criticality in the Gross-Neveu-Yukawa universality class beyond leading order. Using the hyperbolic space formulation of boundary conformal field theories, we compute the first subleading corrections at large $N$ to the free energies of the ``normal", ``ordinary" and ``special" boundary universality classes. We also determine the order $1/N$ correction to the dimension of the boundary fermion at the normal fixed point. In the Gross-Neveu-Yukawa theory in $d=4-ε$, we perform a higher-order analysis of the boundary free energy, and use it to extract estimates for the boundary central charge in $d=3$. The large $N$ and $ε$-expansion results are shown to be precisely consistent in overlapping regimes, providing nontrivial consistency checks for the identification of the boundary universality classes. Our calculations rely on a combination of AdS harmonic analysis and boundary conformal field theory techniques.

In this paper, the large N behavior of a supersymmetric matrix model is compared with its exact continuum description. We concentrate on the large N limit of a supersymmetric matrix model describing a supermembrane with central charge on a toroidally compactified target space. We analyze, on the one hand, the supermembrane model formulated on a differentiable compact torus without boundary, with structure group given by the area-preserving diffeomorphisms, and, on the other hand, the associated regularized $SU(N)$ model. We emphasize in our analysis the structure of the constraints of the regularized model, which generate the $SU(N)$ algebra and reproduce the area-preserving diffeomorphism algebra in the large N limit, together with the topological information associated with the central charge of the model. We explain the role of the central charge in the compactified supermembrane and how it allows a top-down $SU(N)$ regularization. It is known that the regularized model has discrete spectrum. We prove, in the semiclassical approximation of the models, that in the large N limit the eigenvalues of the Hamiltonian of the supersymmetric matrix model are in one to one correspondence with, and converge exactly to, the eigenvalues of the Hamiltonian of the supermembrane (M2-brane) with central charge. Finally, we discuss some physical consequences of this result.

We propose Differences in Detection (DnD), an intuitive method to compare two object detection models. Based on the same matching algorithm, it complements the standard metrics of mean Average Precision ($mAP$) and TIDE error analysis with the ability to compare two models directly. More specifically, we calculate the intersection of ground truth labels that are recognized by both models, followed by the corresponding difference sets and the complement set of ground truth labels that are missed by both models. The resulting comparison is more direct and intuitive than a comparison of independent summary statistics. It reveals individual and shared mistakes and becomes particularly interesting when combined with error types. In this case, the differences in detection errors can be analyzed naturally in a standard confusion matrix. While valuable in itself, we believe that one of the best applications of DnD is to guide explainability methods such as ODAM towards metric-relevant examples, grounded in structured subsets. The code for our method is available here: https://github.com/JohannesTheo/differences-in-detection

In this paper we construct numerically a new class of unstable self-similar solutions of the incompressible Navier-Stokes equations in $\mathbb{R}^3$. Our solutions are axially symmetric and homogeneous of degree $-1$ at $\infty$, and are unstable in the sense that the linearization around these solutions contains unstable modes. Solutions of this type have been discovered numerically by Guillod and Šverák and Hou, Wang, and Yang, and have applications to proving non-uniqueness results. The main novelty in this paper is that we discover the existence of such solutions in the space of axially symmetric swirl-free (ASSF) vector fields. These approximate solutions are defined on all of $\mathbb R^3$ and achieve global pointwise residuals of order $10^{-10}$. We discuss the numerical construction of these solutions in detail, as well as their relevance to the problem of non-uniqueness of solutions of the incompressible Navier-Stokes equations in 3D, in the space of ASSF solutions.

Ultra-hot Jupiters (UHJs; $T_{\rm eq} \gtrsim 2000$ K) enable simultaneous detection of volatile (ice-forming) and refractory (rock-forming) species in planetary atmospheres, providing a powerful diagnostic of planet formation and atmospheric processing. We present a comprehensive high-resolution cross-correlation spectroscopy (HRCCS) analysis of the UHJ MASCARA-1b ($T_{\rm eq} \approx 2600$ K) using the IGRINS and IGRINS-2 spectrographs. We detect robust (SNR$>$4) signals from H$_2$O, CO, OH, Fe I, Mg I, Ca I, and Ti I, marking the most complete atmospheric inventory of MASCARA-1b to date. Using a chemically consistent atmospheric inference framework, we constrain elemental abundances to a typical precision of $\approx$0.2 dex, retrieving a solar atmospheric metallicity ([M/H]$_\odot$ $= 0.07^{+0.17}_{-0.13}$ $\approx 1.2\times$ solar), a C/O ratio (C/O $= 0.65^{+0.08}_{-0.08}$) consistent with solar value (C/O $=$ 0.59), an enhanced refractory abundance ([R/H]$_\odot$ $= 0.40^{+0.23}_{-0.17} \approx 2.5\times$ solar; $\approx 3.8\times$ stellar), and a moderately super-solar refractory-to-volatile ratio ([R/V]$_\odot$ $= 0.36^{+0.11}_{-0.09}$ $\approx 2.3\times$ solar). Comparison with formation models suggests that MASCARA-1b most likely accreted material between the soot-H$_2$O or H$_2$O-CO snowlines (at 68$\%$ confidence). We additionally find stellar values for atmospheric Ti/Mg and Ca/Mg ratios (at 68$\%$ confidence). The Mg/Fe is also found to be consistent with stellar value at 95$\%$ confidence. Therefore, we do not find strong indication of nightside cold trapping in MASCARA-1b. As homogeneous refractory-to-volatile measurements expand across the UHJ population, particularly with upcoming Extremely Large Telescopes, these diagnostics will enable statistically robust tests of emerging trends in giant planet formation and atmospheric evolution.

Urban trip-planning systems are commonly optimized for travel time and cost, but they offer limited support for the heterogeneous needs that real travelers bring, such as personalized preferences, multi-stop itinerary construction, and end-to-end wheelchair accessibility. We present openpaths, a supervisor-specialist multi-agent system that handles all of these tasks within a single architecture. openpaths adopts a deliberate division of labor: LLM agents parse natural-language input, classify request intent, and orchestrate execution, while classical algorithms perform route optimization over curated mobility and accessibility data. This design ensures that the resulting trip honors heterogeneous user preferences and enforces strict accessibility requirements when requested. Beyond per-user planning, openpaths doubles as a measurement instrument for city-scale accessibility analysis: applied to NYC, the system reveals substantial ADA infrastructure gaps and quantifies their effect on job accessibility for wheelchair users. Overall, this study shows how a supervisor-specialist LLM agentic framework can support heterogeneous trip planning and transparent, equitable transportation analysis in real urban environments.

QBism recasts quantum theory as a normative framework for an agent's probability assignments, with the Born rule taking the form of a consistency condition known as the Urgleichung. Motivated by this perspective, qplex theories provide a broader class of probabilistic models in which the sets of valid states and measurements are constrained by QBist-inspired geometric conditions. While qplexes have been extensively studied for single systems, their implications for bipartite correlations remain largely unexplored. In this work, we investigate bipartite correlations in qplex theories by expressing joint expectation values as inner products between suitably defined $C$-vectors. This geometric formulation allows Bell-type inequalities to be studied as optimization problems over qplex-compatible probability assignments. We first analyze the CHSH scenario and show that the shared inner-product structure of the $C$-vectors restricts the maximal value to the Tsirelson bound $2\sqrt{2}$. We then turn to the three-outcome CGLMP inequality $I_{2233}$ and find that the same qplex-derived norm and inner-product constraints allow the algebraic maximum of 4, thereby exhibiting superquantum correlations. These results show that qplex geometry captures enough structure to reproduce an important quantum bound in the two-outcome case, but not enough to recover the full set of quantum correlation constraints. The analysis therefore suggests that additional principles are needed to complete the QBist reconstruction of quantum theory.

We prove $β(\mathbb{Z}_2 \times \mathbb{Z}_4) = 14$, where $β(G)$ denotes the minimum cardinality $|P|$ among finite posets $P$ with $\Aut(P) \cong G$. The lower bound is established by a complete case analysis of orbit decompositions of $P$ under faithful $G$-actions, organized by the largest orbit size. The upper bound is realized by an explicit $14$-element poset whose automorphism group is computed by a height-function argument together with a rigidity analysis of its covering relations.

We introduce and analyze a mixed Virtual Element Method for the $p$-Laplace equation in a non-Hilbertian setting, covering the full range $p \in (1, \infty)$. The discrete framework combines standard mixed Virtual Element spaces with a novel non-linear stabilization term designed to mimic the power-law structure of the continuous operator. We establish discrete inf-sup stability under non-Hilbertian norms and rigorously prove the continuity and coercivity of the discrete form. This guarantees the well-posedness of the problem and allows us to derive a priori error estimates for the primal variable and the flux. A set of numerical tests supports the theoretical derivations.

Node classification in graph neural networks (GNNs) has been widely applied in various fields of graph analysis. GNNs achieve high-accuracy node classification in homophilous graphs, where nodes with the same class label tend to be connected. However, their performance remains limited in heterophilous graphs, where nodes with different class labels are more likely to be connected. In particular, current GNNs derived from graph convolutional networks cannot capture higher-order class label connectivity, which is frequently observed in real-world heterophilous graphs. To address this issue, we propose a novel classifier, Label Context Classifier (LCC), designed to capture higher-order class label connectivity in directed graphs. LCC estimates the class label of a target node by leveraging label context embeddings that are generated through four distinct types of walks. In addition, our approach allows the integration of LCC and any GNN by adaptively learning their importance. Experimental results demonstrate that GNNs integrated with LCC outperform SOTA methods and the label context embeddings improve the node classification performance in heterophilous directed graphs.

This paper sets out the most general computational and econometric implications of the rigorous stochastic equilibrium theory from SELCKE (Staines (2024a)) arXiv:2312.16214. The analytical backbone is the discovery that the system converges geometrically to long-run equilibrium, at a rate given by the greater of the eigenvalue or inverse eigenvalue (from outside) closest to the unit circle and the maximum shock persistence. High-order shocks converge faster. I develop a simulation procedure to test, with asymptotic power, whether stochastic equilibrium exists for a particular model. The fundamental approximation result asserts that, whatever the order of expansion or loss function, the stochastic steady state delivers the most accurate perturbation solution. I also show that super-consistent parameter estimators $O(1/T)$ arise whenever second-order terms vanish. Besides Calvo, I study stochastic equilibrium in two alternative pricing models. Dynamics simplify considerably. I bound the time the impulse response peaks, by the maximum lag in the errors. This lends empirical support to Taylor contracts, although there are issues surrounding unit roots and the strong cost-channel. For menu costs, I demonstrate that the initial price distribution decays away super-exponentially, producing a system equivalent to Calvo with an endogenous reset probability. The impact of idiosyncratic disturbances appears as an additional wedge between actual and efficient output. Blow-up of the objective function at the boundary is proven, with the help of new distributional arguments, so the model meets existing eigenvalue existence conditions for the recursive equilibrium. Along the way, new light is shone on existing theoretical models and statistical procedures.

We construct locally minimizing $(1,2)$-clusters whose exterior interfaces are asymptotic to various prescribed singular area-minimizing cones. For $n+1 \leq 7$, Bronsard & Novack characterized all minimizing $(1,2)$-clusters as standard lenses, whose exterior interface is planar. For $n+1 \in [8,2700]$, the authors together with Bronsard showed the existence of a locally minimizing $(1,2)$-cluster whose exterior interface blows down to some (unknown, possibly non-unique) singular area-minimizing hypercone. For $n+1=8$, this was shown independently by Novaga, Paolini & Tortorelli. Here we develop a refined construction using the Hardt-Simon foliation that realizes prescribed cones. For a singular area-minimizing hypercone $C$ that has an isolated singularity or is cylindrical, we show that if $C$ satisfies an explicit energy bound, then there is a locally minimizing $(1,2)$-cluster whose exterior interface is asymptotic to $C$ with quantitative rates. In fact, if $C$ is an area minimizing Lawson cone satisfying this energy bound, we produce a countably infinite family of distinct locally minimizing clusters asymptotic to $C$, distinguished by their prescribed asymptotic decay to leading order. We verify this energy bound for the generalized Simons cones $C_{k,k}$ in every even ambient dimension $n+1 = 2k+2\geq 8$, and for the cylindrical cone $C_{3,3}\times\mathbb{R}$ in $\mathbb{R}^9$, where $C_{3,3}$ is the Simons cone, therefore answering the cone realization problem in these cases. This in particular removes the upper bound of 2700 on the ambient dimension when $n+1$ is even in our preceding work.

Pre-layout design space exploration (DSE) for high-speed signal integrity (SI) analysis is often limited by the computational cost of simulations and iterative optimization algorithms within modern electronic design automation (EDA) workflows. While machine learning surrogate models accelerate the simulation step, optimizing designs still requires utilizing iterative black-box search methods. This iterative nature scales poorly, making multi-corner sweeps computationally expensive. As a solution, this paper proposes amortized neural optimization (ANO) for pre-layout SI design. ANO entirely eliminates iterative black-box inference by utilizing fully differentiable neural network surrogate models. ANO extracts analytical gradients from the surrogate to train a global optimization policy. Instead of solving the optimization problem repeatedly at inference, the optimization process is learned offline and therefore amortized. Once the ANO policy is trained, it maps different channel contexts directly to near-optimal design parameters in a single deterministic forward pass. The efficiency and accuracy of the ANO framework are demonstrated based on three complex SI design scenarios, including DDR5 decision feedback equalization (DFE), 9-dimensional SerDes Tx/Rx co-equalization, and DDR3 DQS differential pair routing to optimize eye diagram metrics under intra-pair skew constraints. By trading roughly 10% in optimality compared to instance-specific black-box algorithms, it realizes speedups of three to four orders of magnitude. For a large-scale 320,000-instance multi-corner SerDes sweep optimization, ANO collapses what would have taken days of computation using iterative search algorithms into a single batched forward pass that completes in milliseconds. This transforms computationally expensive SI optimization into real-time and interactive pre-layout DSE.

Quantum metasurfaces comprised of subwavelength atomic arrays emerged as a promising platform for enhanced atom-photon interaction. However, realizing such a system with solid-state emitters has been considered impractical due to strong inhomogeneous broadening, which was expected to suppress the photon-mediated interactions that underpin collective emission. Here we report the observation of collective emission from subwavelength arrays of silicon-vacancy centres in diamond -- solid-state emitters whose inhomogeneous broadening exceeds the natural linewidth by two orders of magnitude -- demonstrating that collective effects such as resonance shifts, modified decay rates and directional coherent emission survive this disorder. A crucial enabling element is the implantation of a high density of silicon ions at each array site. This creates so-called superatoms, local ensembles that probabilistically achieve frequency matching across the array and enhance the collective response. We support our observations with a theoretical analysis explaining the mechanisms that preserve the collective effects even in the presence of inhomogeneity. These observations have direct implications for the realization of subwavelength arrays in any solid-state system, paving the way for quantum-emitter metasurfaces that are naturally integrated into nanophotonic environments.

Diabetes affects millions of people and requires reliable continuous glucose monitoring for early hypoglycemia warning. However, medical AI systems must be not only accurate and energy efficient, but also explainable, noise robust, and uncertainty aware. This work presents a 65 nm hypoglycemia forecasting engine based on probabilistic decision trees for trustworthy medical inference. The proposed hybrid architecture combines exact arithmetic evaluation for shallow tree layers with sampling based inference for deeper layers, reducing soft decision tree complexity from exponential to sample efficient traversal. A reconfigurable 4 by 24 by 24 probabilistic node array supports arbitrary tree structures with a maximum depth of 12, coordinated by an on chip low power RISC V core. Fabricated in 65 nm CMOS, the chip achieves 11.3 nJ per inference and a state of the art 30 min forecasting F1 score of 0.825 on continuous glucose monitoring data. Compared with conventional decision tree and random forest models, the proposed engine improves robustness to sensor noise and data point drop off by 4.1x to 16.1x. These results demonstrate an energy efficient, explainable, and uncertainty aware edge AI engine for trustworthy hypoglycemia forecasting.

Vision-language models such as CLIP are highly useful for diverse tasks due to their shared image-text embedding space. Despite this, the image and text embeddings are often poorly aligned, affecting downstream performance. Recent work has shown that this can be attributed to an information imbalance: images contain more information than their captions describe. In this work, we propose TEVI, a framework that uses captions as a signal for what to retain from image embeddings. Specifically, we use sparse autoencoders to disentangle image embeddings and train a masking module to selectively reconstruct the embedding based on a given caption. In a controlled setup with synthetic captions, we show that TEVI is effective at preserving caption-described attributes while discarding others. By applying TEVI to CLIP models trained on natural images, we further achieve improved retrieval performance across coarse-grained short-caption (MS COCO, Flickr) and fine-grained long-caption (IIW, DOCCI) benchmarks, with stronger gains on richer captions, and improved robustness on the RoCOCO benchmark.

This paper addresses orbital stabilization of a circular motion primitive for a dynamic extension of the Dubins car model within a transverse-linearization framework. We show that the corresponding transverse linearization is unstable and not stabilizable by linear state feedback. Therefore, the standard linearization-based approach to orbital stabilization cannot be applied directly. The main contribution is a set of explicit and verifiable conditions that characterize when a controller design based on transverse linearization remains applicable. These conditions rely on the specific structure of the dynamics in a neighborhood of the motion and on the use of non-standard transverse coordinates for controller design and analysis. Numerical simulations illustrate the proposed design procedure.

Few-Shot Medical Image Segmentation (FSMIS) aims to segment novel object classes in medical images using only minimal annotated examples, addressing the critical challenges of data scarcity and domain shifts prevalent in medical imaging. While Diffusion Models (DM) excel in visual tasks, their potential for FSMIS remains largely unexplored. We propose that the rich visual priors learned by large-scale DMs offer a powerful foundation for a more robust and data-efficient segmentation approach. In this paper, we introduce SD-FSMIS, a novel framework designed to effectively adapt the powerful pre-trained Stable Diffusion (SD) model for the FSMIS task. Our approach repurposes its conditional generative architecture by introducing two key components: a Support-Query Interaction (SQI) and a Visual-to-Textual Condition Translator (VTCT). Specifically, SQI provides a straightforward yet powerful means of adapting SD to the FSMIS paradigm. The VTCT module translates visual cues from the support set into an implicit textual embedding that guides the diffusion model, enabling precise conditioning of the generation process. Extensive experiments demonstrate that SD-FSMIS achieves competitive results compared to state-of-the-art methods in standard settings. Surprisingly, it also demonstrated excellent generalization ability in more challenging cross-domain scenarios. These findings highlight the immense potential of adapting large-scale generative models to advance data-efficient and robust medical image segmentation.

Conventional few-shot medical image segmentation (FSMIS) approaches face performance bottlenecks that hinder broader clinical applicability. Although the Segment Anything Model (SAM) exhibits strong category-agnostic segmentation capabilities, its direct application to medical images often leads to over-segmentation due to ambiguous anatomical boundaries. In this paper, we reformulate SAM-based FSMIS as a prompt localization task and propose FoB (Focus on Background), a background-centric prompt generator that provides accurate background prompts to constrain SAM's over-segmentation. Specifically, FoB bridges the gap between segmentation and prompt localization by category-agnostic generation of support background prompts and localizing them directly in the query image. To address the challenge of prompt localization for novel categories, FoB models rich contextual information to capture foreground-background spatial dependencies. Moreover, inspired by the inherent structural patterns of background prompts in medical images, FoB models this structure as a constraint to progressively refine background prompt predictions. Experiments on three diverse medical image datasets demonstrate that FoB outperforms other baselines by large margins, achieving state-of-the-art performance on FSMIS, and exhibiting strong cross-domain generalization.

Semi-supervised medical image segmentation (SSMIS) aims to alleviate annotation scarcity, but general methods, often developed on few-class datasets, suffer performance degradation in class-imbalanced multi-organ scenarios. Existing class-imbalanced SSMIS methods also struggle, as their single-decoder architecture is forced to handle vastly different scales with shared parameters. This process is easily dominated by majority classes, fundamentally limiting tail-class segmentation capability. To address this, we propose a "Divide, Conquer, and Aggregate" (DCA) framework, featuring a unified encoder, three expert decoders, and an aggregation decoder. First, we Divide by applying a Logarithmic Gap Analysis to statically partition foreground classes into stable Head, Medium, and Tail sets, which aligns with anatomical priors. Then, we Conquer by training the three architecturally asymmetric experts independently using a label-split strategy. This fundamentally alleviates the burden on a single decoder. The experts' predictions on unlabeled data are fused via logit stitching to generate high-quality pseudo-labels. Finally, we Aggregate using an aggregation decoder with a Dynamic Feature Aggregation Module (DFAM), which dynamically fuses priors from all three experts to achieve unbiased predictions and fully leverage unlabeled data. Experiments demonstrate that our DCA framework significantly outperforms state-of-the-art general and class-imbalanced SSMIS methods.

Although foundation models (e.g. SAM) perform remarkably in medical image segmentation, its high computational complexity limits deployment. Knowledge distillation (KD) allows lightweight models to inherit the representational capabilities of large models, thereby mitigating this issue. Existing KD methods enhance student performance, but due to teacher-student different feature advantages, they neglect to internalize and integrate student's semantic information adaptively after knowledge transfer, causing poor knowledge assimilation and limiting gains and generalization. To address this limitation, we propose a novel medical image segmentation framework, which is infusion to assimilation distillation (IAD). In Knowledge Infusion Stage (KIS), to semantically align teacher-student prediction distributions, soft-label distillation is combined with class-weighted prototype alignment strategy. In Knowledge Assimilation Stage (KAS), to promote adaptively semantic assimilation, a contrastive semantic self-optimization strategy refines student predictions through positive and negative sample pairs and imposes reverse constraints on encoder features to enhance semantic consistency. IAD achieves DICE gains of 4.32% on Synapse, 1.85% on ACDC, and 2.42% on Polyp datasets, and delivers an average 4.16% generalization gain on ISIC2018, PH2, BUSI, and STU datasets, outperforming mainstream KD methods. Code is available at https://github.com/hjklearn/IAD.

Medical image segmentation demands both high accuracy and computational efficiency, yet existing methods face a critical trade-off: CNNs lack global context while transformers incur prohibitive costs for deployment on resource-constrained devices. To address this challenge, we propose a Physics-informed Multi-scale Refinement Network (PMRNet), integrating symplectic geometry, renormalization group theory, and entropy diffusion to guide feature learning. PMRNet features three innovations: (1) a physics-informed encoder with Enhanced Symplectic Convolution for boundary detection and Renormalization Group-inspired Downsampling for information preservation; (2) a Pseudo-Global Receptive Field module achieving near-global context with linear complexity through entropy-driven diffusion; and (3) a boundary-aware decoder for precise delineation. With only 0.87M parameters and 3.43 GFLOPs, PMRNet achieves 87.25% IoU and 92.56% Dice on the challenging Clinic dataset, outperforming state-of-the-art (SOTA) models with even 100x more parameters across 12 medical imaging datasets while maintaining computational efficiency. Code is available at https://github.com/KangBoce/PMRNet.

Unsupervised Domain Adaptation (UDA) is essential for deploying medical segmentation models across diverse clinical environments. Existing methods are fundamentally limited, suffering from semantically unaware feature alignment that results in poor distributional fidelity and from pseudo-label validation that disregards global anatomical constraints, thus failing to prevent the formation of globally implausible structures. To address these issues, we propose SHAPE (Structure-aware Hierarchical Unsupervised Domain Adaptation with Plausibility Evaluation), a framework that reframes adaptation towards global anatomical plausibility. Built on a DINOv3 foundation, its Hierarchical Feature Modulation (HFM) module first generates features with both high fidelity and class-awareness. This shifts the core challenge to robustly validating pseudo-labels. To augment conventional pixel-level validation, we introduce Hypergraph Plausibility Estimation (HPE), which leverages hypergraphs to assess the global anatomical plausibility that standard graphs cannot capture.This is complemented by Structural Anomaly Pruning (SAP) to purge remaining artifacts via cross-view stability.SHAPE significantly outperforms prior methods on cardiac and abdominal cross-modality benchmarks, achieving state-of-the-art average Dice scores of 90.08% (MRI - CT) and 78.51% (CT - MRI) on cardiac data, and 87.48% (MRI - CT) and 86.89% (CT - MRI) on abdominal data.

Vision-Language Models (VLMs), such as CLIP, have demonstrated remarkable zero-shot capabilities in various computer vision tasks. However, their application to medical imaging remains challenging due to the high variability and complexity of medical data. Specifically, medical images often exhibit significant domain shifts caused by various confounders, including equipment differences, procedure artifacts, and imaging modes, which can lead to poor generalization when models are applied to unseen domains. To address this limitation, we propose Multimodal Causal-Driven Representation Learning (MCDRL), a novel framework that integrates causal inference with the VLM to tackle domain generalization in medical image segmentation. MCDRL is implemented in two steps: first, it leverages CLIP's cross-modal capabilities to identify candidate lesion regions and construct a confounder dictionary through text prompts, specifically designed to represent domain-specific variations; second, it trains a causal intervention network that utilizes this dictionary to identify and eliminate the influence of these domain-specific variations while preserving the anatomical structural information critical for segmentation tasks. Extensive experiments demonstrate that MCDRL consistently outperforms competing methods, yielding superior segmentation accuracy and exhibiting robust generalizability.

Medical image segmentation remains challenging due to limited annotations for training, ambiguous anatomical features, and domain shifts. While vision-language models such as CLIP offer strong cross-modal representations, their potential for dense, text-guided medical image segmentation remains underexplored. We present MedCLIPSeg, a novel framework that adapts CLIP for robust, data-efficient, and uncertainty-aware medical image segmentation. Our approach leverages patch-level CLIP embeddings through probabilistic cross-modal attention, enabling bidirectional interaction between image and text tokens and explicit modeling of predictive uncertainty. Together with a soft patch-level contrastive loss that encourages more nuanced semantic learning across diverse textual prompts, MedCLIPSeg effectively improves data efficiency and domain generalizability. Extensive experiments across 16 datasets spanning five imaging modalities and six organs demonstrate that MedCLIPSeg outperforms prior methods in accuracy, efficiency, and robustness, while providing interpretable uncertainty maps that highlight local reliability of segmentation results. This work demonstrates the potential of probabilistic vision-language modeling for text-driven medical image segmentation.

Medical image segmentation is challenging due to limited annotated data, high labeling costs, and substantial image heterogeneity. Although large-scale vision foundation models (e.g., SAM) have shown great potential in this field, existing SAM-based methods typically rely on expert-defined geometric prompts or complex clinical text prompts, which limits their generalizability across diverse medical image segmentation tasks. To overcome these challenges, we propose Simple-ViLMedSAM, a CLIP-SAM integration framework that enables high-accuracy segmentation in zero-shot and few-shot settings using only simple text queries, that is, using only basic anatomical or disease-related text labels. At its core is an Implicit Pos-Prompter (IPP), which generates attribution maps containing implicit positional cues to replace traditional geometric prompts. IPP incorporates a multi-modal information bottleneck and an affinity-based refinement strategy to ensure high-quality guidance from CLIP-SAM interactions. To further enhance segmentation, we introduce a Bidirectional Interaction Decoder (BID) that employs bidirectional cross-attention to align IPP's positional maps with SAM's pixel-level features. By jointly modeling global semantics and local details, BID significantly improves segmentation accuracy. Extensive experiments on four public datasets demonstrate that Simple-ViLMedSAM consistently outperforms existing methods in both zero-shot and few-shot medical image segmentation tasks, using only simple text queries. The code will be publicly available upon acceptance.

Mamba-based models have emerged as a promising paradigm for medical image segmentation due to their linear-complexity state-space modeling. However, their effectiveness is still limited by the mismatch between anatomical geometry and tissue-specific semantics, as well as by spatially entangled diagnostic cues. To address these limitations, we propose GeoSemba, a Mamba-based segmentation framework that jointly models cross-level geometric-semantic interactions and cross-dimensional spatial-channel dependencies within a single scan. To this end, GeoSemba comprises two complementary components. The Semantic-guided State Refiner (SSR) derives semantically discriminative region representatives and leverages geometry-conditioned inter-region dependencies to enable coherent semantic propagation across structurally related regions. The Cross-dimensional Affinity Refiner (CAR) adopts a coarse-to-fine strategy of macro-perception and micro-focus to selectively enhance informative spatial-channel interactions while suppressing weak and noisy correlations. Extensive experiments on benchmark datasets spanning six medical imaging modalities show that GeoSemba consistently delivers superior segmentation accuracy while maintaining high computational efficiency.

Semi-supervised learning addresses label scarcity and high annotation costs in medical image segmentation by exploiting the latent information in unlabeled data to enhance model performance. Traditional discriminative segmentation relies on segmentation masks, neglecting feature-level distribution constraints. This limits robust semantic representation learning and adaptive modeling of unlabeled data in scenarios with few labels. To address these limitations, we propose SemiGDA, a novel Generative Dual-distribution Alignment framework for semi-supervised medical image segmentation. Our SemiGDA overcomes the reliance of discriminative methods on large labeled datasets by aligning feature and semantic distributions to boost semantic learning and scene adaptability. Specifically, we propose a Dual-distribution Alignment Module (DAM), which employs two structurally distinct encoders to model image and mask feature distributions. It enforces their alignment in the latent space via distributional constraints, establishing structured feature consistency. Moreover, we design a Consistency-Driven Skip Adapter (CDSA) strategy, which introduces dual skip adapters (Image and Mask) to fuse multi-scale features via skip connections. Using a consistency loss, CDSA enhances cross-branch semantic alignment and reinforces fine-grained semantic consistency. Experimental results on diverse medical datasets show that our method outperforms other state-of-the-art semi-supervised segmentation methods. Code is released at: https://github.com/taozh2017/SemiGDA.

We introduce VoxTell, a vision-language model for text-prompted volumetric medical image segmentation. It maps free-form descriptions, from single words to full clinical sentences, to 3D masks. Trained on 62K+ CT, MRI, and PET volumes spanning 1K+ anatomical and pathological classes, VoxTell uses multi-stage vision-language fusion across decoder layers to align textual and visual features at multiple scales. It achieves state-of-the-art zero-shot performance across modalities on unseen datasets, excelling on familiar concepts while generalizing to related unseen classes. Extensive experiments further demonstrate strong cross-modality transfer, robustness to linguistic variations and clinical language, as well as accurate instance-specific segmentation from real-world text. Code is available at: https://github.com/MIC-DKFZ/VoxTell

Medical image segmentation supports clinical workflows by precisely delineating anatomical structures and lesions. However, medical image datasets medical image datasets suffer from acquisition noise and annotation ambiguity, causing pervasive data uncertainty that substantially undermines model robustness. Existing research focuses primarily on model architectural improvements and predictive reliability estimation, while systematic exploration of the intrinsic data uncertainty remains insufficient. To address this gap, this work proposes leveraging the universal representation capabilities of visual foundation models to estimate inherent data uncertainty. Specifically, we analyze the feature diversity of the model's decoded representations and quantify their singular value energy to define the semantic perception scale for each class, thereby measuring sample difficulty and aleatoric uncertainty. Based on this foundation, we design two uncertainty-driven application strategies: (1) the aleatoric uncertainty-aware data filtering mechanism to eliminate potentially noisy samples and enhance model learning quality; (2) the dynamic uncertainty-aware optimization strategy that adaptively adjusts class-specific loss weights during training based on the semantic perception scale, combined with a label denoising mechanism to improve training stability. Experimental results on five public datasets encompassing CT and MRI modalities and involving multi-organ and tumor segmentation tasks demonstrate that our method achieves significant and robust performance improvements across various mainstream network architectures, revealing the broad application potential of aleatoric uncertainty in medical image understanding and segmentation tasks. The code is available.

In medical image segmentation tasks, the domain gap caused by the difference in data collection between training and testing data seriously hinders the deployment of pre-trained models in clinical practice. Continual Test-Time Adaptation (CTTA) aims to enable pre-trained models to adapt to continuously changing unlabeled domains, providing an effective approach to solving this problem. However, existing CTTA methods often rely on unreliable supervisory signals, igniting a self-reinforcing cycle of error accumulation that culminates in catastrophic performance degradation. To overcome these challenges, we propose a CTTA via Semantic-Prompt-Enhanced Graph Clustering (SPEGC) for medical image segmentation. First, we design a semantic prompt feature enhancement mechanism that utilizes decoupled commonality and heterogeneity prompt pools to inject global contextual information into local features, alleviating their susceptibility to noise interference under domain shift. Second, based on these enhanced features, we design a differentiable graph clustering solver. This solver reframes global edge sparsification as an optimal transport problem, allowing it to distill a raw similarity matrix into a refined and high-order structural representation in an end-to-end manner. Finally, this robust structural representation is used to guide model adaptation, ensuring predictions are consistent at a cluster-level and dynamically adjusting decision boundaries. Extensive experiments demonstrate that SPEGC outperforms other state-of-the-art CTTA methods on two medical image segmentation benchmarks. The source code is available at https://github.com/Jwei-Z/SPEGC-for-MIS.

Prototype or region-attention modules have recently improved medical image segmentation but still suffer from two fundamental limitations: 1) they represent each semantic concept as a point or isotropic region, failing to capture the inherently anisotropic geometry of real feature distributions; and 2) many rely on non-differentiable clustering or one-way kernel weighting, which restricts their ability to form coherent region-level representations. We address these issues with the Anisotropic Differentiable Granular-Ball (AD-GBC) module, which generalizes prototypes into learnable geometric regions parameterized by a center and an anisotropic vector scale. AD-GBC aggregates local features into region-level semantics and redistributes the refined representation back to pixels in a fully differentiable manner, enabling geometry-aware refinement within modern UNet-style architectures. Two geometric regularizers, a Wasserstein-based diversity loss and a scale consistency loss, mitigate center collapse and encourage stable, well-formed region geometry.AD-GBC yields consistent improvements across four widely used medical segmentation benchmarks (BUSI, GlaS, CVC-ClinicDB, ISIC17) when integrated into two strong backbones (Rolling-UNet and U-KAN), demonstrating that the proposed geometric region formulation generalizes well across different imaging conditions. The code is available at https://github.com/SiaShen-dot/AD-GBC.

While diffusion has drawn considerable recent attention from the language modeling community, continuous diffusion has appeared less scalable than discrete approaches. To challenge this belief we revisit Plaid, a likelihood-based continuous diffusion language model (DLM), and construct RePlaid by aligning the architecture of Plaid with modern discrete DLMs. In this unified setting, we establish the first scaling law for continuous DLMs that rivals discrete DLMs: RePlaid exhibits a compute gap of only $20\times$ compared to autoregressive models, outperforms Duo while using fewer parameters, and outperforms MDLM in the over-trained regime. We benchmark RePlaid against recent continuous DLMs: on OpenWebText, RePlaid achieves a new state-of-the-art PPL bound of $22.1$ among continuous DLMs and superior generation quality. These results suggest that continuous diffusion, when trained via likelihood, is a highly competitive and scalable alternative to discrete DLMs. Moreover, we offer theoretical insights to understand the advantage of likelihood-based training. We show that optimizing the noise schedule to minimize the ELBO's variance naturally yields linear cross-entropy (information loss) over time. This evenly distributes denoising difficulty without any case-specific time reparameterization. In addition, we find that optimizing embeddings via likelihood creates structured geometries and drives the most significant likelihood gain.