Geophysics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 9 November
Multiparameter elastic full-waveform inversion (EFWI) provides a powerful means of reconstructing subsurface elastic properties, but its practical application remains challenging because of strong nonlinearity, parameter crosstalk, and the different sensitivities of seismic data to individual elastic parameters. We develop a robust implicit neural representation (INR) framework for multiparameter EFWI, using continuous neural representations to parameterize the subsurface model and provide an effective model-space constraint during inversion. Within this framework, we systematically investigate different elastic parameterizations and demonstrate that parameterization plays a critical role in controlling parameter coupling and inversion stability. Among the tested formulations, combining P-wave velocity (Vp), P-wave impedance (Ip), and the Vs/Vp ratio provides the most robust overall performance, yielding improved parameter separation and more reliable reconstruction of elastic properties. We evaluate the proposed approach on the Marmousi and Overthrust models under increasingly challenging inversion conditions and further apply it to the field Volve ocean-bottom-node dataset. Both synthetic and field-data results demonstrate that the INR-based formulation, together with the Vp-Ip-Vs/Vp parameterization, improves inversion stability, suppresses parameter cross-talk and inversion artifacts, and produces geologically coherent multiparameter models. These results highlight the importance of jointly considering model representation and physical parameterization in multiparameter EFWI and demonstrate the potential of robust INR-based inversion for complex synthetic and field seismic applications.
Geometric Topology
Infinitude of integral Hitchin representations of Coxeter polygons
We prove that there are infinitely many Hitchin $\operatorname{SL}^\pm(3, \mathbb Z)$-representations of a Coxeter $n$-gon($n \geq 9$) group whose boundary parameters are fixed. We construct infinitely many Hitchin $\operatorname{SL}^\pm(3, \mathbb Z)$-representations of a specific Coxeter $6$-gon or $8$-gon corresponding to positive integral points of a generalized Markoff equation.
Read the full abstract
Artificial Intelligence
Can LLM Agents Organize Their Own Memory?
In most LLM agents, weights are fixed after deployment, so new information must be retained in external memory. Existing approaches often rely on predefined records and fixed organizational rules, which cannot readily accommodate evolving tasks and interactions. This raises the question: can LLM agents organize their own memory using ordinary documents? We introduce self-organized memory (SoMem), a framework that stores memory in linked documents. The agent decides what to summarize, how to group related information, and when to revise existing topics, and every summary it produces stays linked to the original interactions as evidence. The runtime validates updates and automatically derives temporal, relational, and recent-memory views from the organized content. At query time, the agent chooses what to search and read, following document links and inspecting original evidence as needed. On LoCoMo, BEAM-100K, and LongMemEval-S, SoMem outperforms the strongest baseline by 2.66, 3.10, and 2.00 percentage points, respectively. The constructed memory can be reused across models, and additional reasoning effort can further improve answer quality. These findings demonstrate the feasibility of agents organizing and using their own persistent memory through ordinary documents. Code is available at: https://github.com/Qi202/SoMem.
Read the full abstract
Statistical Methodology
A Semiparametric Bayesian Generalized Linear Mixed Model with a Nonparametric Count Distribution for Seizure Data
Functional count data arise when recurrent events are recorded repeatedly over time and scientific interest centers on how event rates evolve over time and across experimental conditions. Standard mixed effects models for counts have interpretable mean structures but require a parametric conditional distribution, which can be restrictive in the presence of overdispersion, unusual zero frequencies and heavy tails. To address these challenges, we develop the functional count DPGLM (fcDPGLM), a semiparametric Bayesian mixed model that combines GLM-style regression with a nonparametric count distribution. Parametric condition-specific mean trajectories and subject-specific random effects determine the conditional mean exactly, allowing the interpretation of contrasts as usual, while remaining details of the count distribution are learned nonparametrically from the data. Poisson and negative-binomial mixed models are recovered as limiting cases. A gamma completely random measure representation allows for efficient posterior simulation. In simulations mimicking a rat seizure study, fcDPGLM estimated mean contrasts as accurately as a correctly specified negative-binomial mixed model, and more accurately otherwise, and recovered zero and exceedance probabilities that the parametric models misrepresent. In the seizure study, seizure rates rise and then decline in the early period after induced status epilepticus but remain high and nearly flat in the chronic period, and rats differ substantially. Poisson and negative-binomial mixed models estimate similar mean trajectories and period contrasts, but only fcDPGLM matches the observed zero frequency, variance and high-burden days.
Read the full abstract
Neurons and Cognition
Language processing in the autistic brain: A meta-analysis of neuroimaging studies across modalities and age
Language is central to social communication, which is altered in autism, and language abilities vary widely among autistic individuals. How the autistic brain processes language is central to understanding autistic communication. Yet neuroimaging studies report inconsistent findings on how autistic brain activity is altered during language processing. We synthesized 45 fMRI and PET studies (753 autistic and 777 non-autistic participants) with seed-based d mapping with permutation of subject images (SDM-PSI). We estimated autistic and non-autistic activation and their difference with meta-regressions adjusted for each study's language modality and mean age. Autistic and non-autistic brains activated a similar left-lateralized language network. Autistic brains showed reduced activity throughout the left perisylvian cortex, from the inferior frontal gyrus to the mid superior temporal gyrus and the posterior superior temporal sulcus, and increased activity in the right temporal pole. Spoken language showed a larger autistic reduction than written language in the left mid superior temporal gyrus. Adults, but not children and adolescents, showed increased autistic activity in the right superior temporal gyrus and cuneus. Autistic language processing thus relies on the typical left-lateralized network but shows reduced leftward lateralization.
Read the full abstract
Computer Vision
Text Alone Predicts Which Classes a Segmentation Readout Keeps
An open-vocabulary segmentation readout silently loses a class deployed beyond its training categories. It still emits masks with high confidence, but the class never wins a single pixel. The training loss falls anyway, and mIoU can even rise while classes die. We show that which classes collapse is written in the class-name list alone. Before any image is seen, we rank each class's collapse risk from its text similarity to the training vocabulary alone. The ranking holds at AUROC 0.69-0.96 across nine model families, twelve backbones and eight pretraining recipes. Two further families, measured only after the predictions were written down, land in the same band. What the list writes, it can rewrite. With identical images and a class's counterpart moved in or out of the training list, retraining flips that class's gain over the frozen baseline by 21 IoU points; swapping in unrelated names changes nothing. We find a vocabulary-distance mechanism behind the collapse. Collapsed classes lose every vote to near-name neighbors, but in the frozen features they survive. Turning down the readout's amplitude, or stopping training early, brings them back, up to +16 points, without retraining; trained with its amplitude held down, the readout never collapses.
Read the full abstract
Complex Variables
Smooth limits of compact Kahler manifolds with vanishing first Chern class
Let $\pi\colon X\to\Delta$ be a proper holomorphic submersion from a smooth complex manifold onto the unit disc. Suppose that for every $t\in\Delta^*$ the fiber $X_t$ is a compact Kahler manifold whose real first Chern class vanishes. We prove that the central fiber X_0 belongs to Fujiki's class $\cC$, that is, X_0 is bimeromorphic to a compact Kahler manifold.
Read the full abstract
Biological Physics
Crowding limits relative bacterial surface accumulation with density-induced effective heating
The accumulation of microswimmers near walls has been studied for decades. While hydrodynamics and single-cell kinematics are well-explored, the role of cell-cell interactions remains understudied. However, in realistic, crowded environments such as the human gut, these interactions can be just as crucial. Here, we demonstrate that they actively weaken bacterial surface accumulation. Analysis of entrapped cell tracks reveals that crowding enhances orientational fluctuations, which increase the effective rotational diffusion and facilitate wall escape. A minimal model shows that this collective effect is equivalent to raising the system's effective temperature with density.
Read the full abstract
Statistical Methodology
Characterizing Underrepresentation when Generalizing Causal Survival Estimates
Randomized trial findings are routinely generalized to broader target populations to inform decisions in medicine and public policy. When the trial sample is misaligned with the target population, generalized effect estimates can yield suboptimal decisions. We characterize the target subgroups that are underrepresented in the trial cohort in time-to-event settings. Here, underrepresentation arises in two distinct ways: baseline underrepresentation, when a subgroup's covariate profile is rare in the trial relative to the target, and time-varying underrepresentation, when differential censoring erodes a subgroup over follow-up. Existing diagnostics either conflate these mechanisms or apply only to the baseline case. We show that the asymptotic variance of the efficient generalizability estimator admits a decomposition in which only two terms can diverge as representation weakens: a sampling-weight term capturing baseline misalignment and a censoring-weight term capturing differential loss to follow-up. Underrepresented subgroups are therefore those with imprecise generalized estimates, and the larger of the two terms gives the responsible channel. We use this decomposition in TRACE (Temporal Rashomon Analysis of CEnsored data), which combines a self-normalized variance objective with a tree-based search to localize underrepresented subgroups, attribute each to baseline mismatch or censoring, and find the follow-up intervals where censoring drives the variance inflation. We solve the joint problem with a parametric dynamic program nested inside a Rashomon ensemble of near-optimal trees. In synthetic benchmarks, TRACE recovers the planted underrepresented regions. In a case study generalizing the National Job Training Partnership Act (JTPA) Study to U.S. JTPA-eligible adults, restricting to reliably generalizable subgroups narrows the 95% confidence interval for the refined estimand by 60%.
Read the full abstract
General Relativity and Quantum Cosmology
Deep-Learning-Assisted Bayesian Inference for Extreme-Mass-Ratio Inspirals
Extreme-mass-ratio inspirals (EMRIs) pose a major challenge to Bayesian parameter inference because of their complex, highly multimodal likelihood surfaces. Locating narrow high-likelihood regions is particularly difficult without knowledge of the true source parameters, which is unavailable in real observations. We develop a deep-learning-assisted Bayesian inference framework that combines neural-network-guided parameter-space localization, particle swarm optimization, and Markov chain Monte Carlo (MCMC) sampling. Applied to simulated six-month EMRI signals, the framework recovers posterior distributions in close agreement with reference MCMC analyses initialized near the true parameter values, without requiring such knowledge for initialization. Once the neural network has been trained, the complete inference pipeline takes approximately 10 hours per signal.
Read the full abstract
Machine Learning
An Evaluation of Tabular Foundation Models as Training-Free World Models
World models are widely considered to be a potential key component of model-based reinforcement learning and robot planning, but typically require task-specific training on substantial interaction data. We investigate whether a pretrained tabular foundation model (TabFM) can serve as a training-free dynamics model for world modeling by framing next-state prediction as in-context regression over transition pairs, and where such a model helps and where it breaks. Across synthetic dynamical systems and robot locomotion benchmarks, TabFM is more accurate than LSTM and MLP models trained from scratch on the same limited transitions in almost all settings, and among regressors fitted to the same data it is the most accurate on locomotion, while a Gaussian process is more accurate on smooth low-dimensional systems. However, models trained on the full dataset remain substantially better on most systems. When rolled out autoregressively over multiple steps, TabFM's accuracy depends strongly on how much context it is given. With a short context, its rollouts quickly degrade to the level of a trivial baseline, whereas with a long context they improve substantially and, on one locomotion task, exceed a trained LSTM. Inference is also about three orders of magnitude slower than an LSTM. Overall, in their current state, we find that training-free priors are useful as predictors when data are scarce and enough context is available, but they are not yet a substitute for trained world models in planning.
Read the full abstract
High Energy Physics (Theory)
Proof of the QNEC, without assumptions and under a weight
We complete a proof of the quantum null energy condition (QNEC) in the mathematical framework of Half-Sided Translations and Modular Inclusions (HST/HSMI) removing an assumption that applied to previous proofs. Here we establish a theorem, which when translated to a physical setting implies that the relative entropy comparing the state of fields outside a stationary black hole with a Hartle-Hawking (HH) state is always a convex function of the horizon cut. Previous proofs only applied when a purification with finite null energy was available. We also obtain a QNEC generalization in which the HH Hilbert space is replaced with one based on a weight which is plausibly of relevance to evaporating black holes in flat space.
Read the full abstract
Differential Geometry
Conical Singularities of Toda Solutions with One Finite-Energy Component and Classification on the Punctured Plane
We study Toda systems associated with arbitrary complex simple Lie algebras. For solutions on a punctured disc, we prove that finite energy of at least one component near the puncture forces every component to have a conical singularity there. On the punctured plane, we give a geometric formulation and proof of the Lin–Nie–Wei classification of solutions with prescribed cone exponents at the origin and finite total energy in at least one component. We express the parameter space as the complex centralizer of the monodromy modulo a maximal compact subgroup. We also recover the formulas of Karmakar–Lin–Nie–Wei for the cone exponents at infinity and the total energies of the components.
Read the full abstract
Atmospheric and Oceanic Physics
Ensemble Covariance Propagation with Machine Learning Weather Prediction Models in a Realistic Data Assimilation System
Integrating machine learning weather prediction (MLWP) models into classical data assimilation (DA) workflows requires that they accurately evolve ensemble covariances. We evaluate this capability in a two-way coupled DA system, where an ensemble variational (EnVar) component evolves the mean and an ensemble Kalman filter (EnKF) evolves the covariances. The system is cycled over one month and assimilates a wide range of real-world observations, including microwave radiances. Using the MLWP model GraphCast for the EnKF and the physics-based MPAS model for the EnVar yields ensemble spread and covariances comparable to a reference configuration in which MPAS evolves both components. The resulting EnVar mean achieves consistently lower RMSE against GFS analyses, except near the model top, where spurious correlations degrade performance. These results demonstrate that MLWP models can propagate covariance information in a realistic DA cycling system and provide a proof of concept for their integration into future operational ensemble systems.
Read the full abstract
General Relativity and Quantum Cosmology
Efficiency enhancement in the Unruh quantum Otto engine
We investigate the efficiency enhancement of an Unruh quantum Otto engine (UQOE) composed of two initially unentangled Unruh-DeWitt (UDW) detectors. The two detectors have different accelerations and interact with a massless scalar field in the Minkowski vacuum. We choose a diagonal free Hamiltonian for the UQOE and define its work and heat with respect to the proper time of one of the detectors. Consequently, the coherence terms generated through the interaction with the scalar field do not directly contribute to the work and heat of the UQOE. Instead, the efficiency enhancement arises from the relative proper-time factor, which assigns different weights to the population changes of the two detectors. We then identify cyclic operating points that satisfy the conditions of initial-state passivity, cycle closure, and energy conservation. Their efficiency is higher than that of a conventional UQOE, and the state of the UQOE after the heat-absorption stage is separable. Finally, we study the effect of the separation between the two detectors on the engine efficiency. During the interaction of the two detectors with the field, increasing their separation reduces the generated entanglement; however, this separation enters the nonlocal coherence terms, thereby restricting the range of parameter settings compatible with cycle closure and influencing the achievable efficiency enhancement. Our results demonstrate that the UQOE can achieve enhanced efficiency even when its state after the heat-absorption stroke is separable, showing that entanglement generation is not a necessary condition for efficiency enhancement.
Read the full abstract
Machine Learning
LBSNet: Lightweight Brain State Modeling via Learnable Spectral Prototypes for EEG-Based Fatigue Detection
EEG-based fatigue detection requires modeling spectral information across bands and its temporal variation. Recent methods combine multi-band modeling with adaptive temporal processing, yet how spectral structure across bands should influence the contribution of each temporal interval remains insufficiently explored. We propose LBSNet, a lightweight network that uses prototype-based brain-state modeling to link local spectral structure with temporal aggregation. LBSNet extracts compact multi-band features through band-specific temporal convolution, shared spatial projections, and cross-band recalibration. Within each window, relative spectral activity is encoded as a normalized descriptor and softly matched to learnable spectral prototypes. Prototype assignments and learned prototype importance jointly determine temporal weights for power aggregation, allowing the spectral pattern of each window to guide its contribution to the prediction. Subject-independent experiments on SEED-VIG and SADT show that LBSNet outperforms seven baselines while maintaining low computational cost. Ablation and visualization analyses support prototype-guided temporal weighting, while the learned prototypes reveal a gradual redistribution of relative cross-band spectral states with increasing fatigue. The code is publicly available at https://github.com/zy0929/LBSNet.
Read the full abstract
Exactly Solvable and Integrable Systems
On a S-Integrable Nonlinear Evolution Equation with an Additional Algebraic Constraint
We introduce a matrix nonlinear evolution equation that obeys an additional cubic constraint. That matrix partial differential equation in two independent variables is integrable in the sense of inverse scattering transform. It has a zero curvature representation with a Lax pair being a linear bundle in pole gauge and is a generalization of Heisenberg ferromagnet equation related to the special linear algebra of arbitrary rank. We prove that the scattering operator is gauge equivalent to a linear bundle Lax operator with a complex diagonal $\lambda$-leading term. We derive a few reductions e.g. a Cartan reduction, and an orthogonal reduction of the nonlinear evolution equation under consideration.
Read the full abstract
Machine Learning
CRAFT-GBM: Selecting Best Features for Gradient-Boosted Machine Learning
Feature subsampling is used to accelerate gradient-boosted trees, yet standard approaches fix the feature count in advance and largely ignore the split-gain evidence accumulated during training. We introduce CRAFT-GBM (Confidence-Ranked Adaptive Feature Tracking), a native LightGBM extension that uses this evidence to adapt which—and, by default, how many—features are searched each boosting round. CRAFT-GBM transforms root-split gains into bounded observations and constructs joint time-uniform bounds valid under adaptive, non-stationary training histories. In its default endogenous mode, after initialization, CRAFT-GBM retains exactly the features that cannot yet be ruled out as best, letting the selected feature-set size $K_t$ emerge from training history rather than a pre-specified fraction. With probability at least $1-\delta$, every feature with the largest expected gain is retained jointly across all rounds after initialization. When a feature fraction $f$ is specified instead, CRAFT-GBM ranks features by their interval point estimates and provides an approximate top-$K$ guarantee for feature budget $K_f$. We implement CRAFT-GBM within LightGBM while preserving its native feature-bundling representation, and evaluate it across 14 public and synthetic dataset families and four noise-augmented settings. In the fully endogenous setting, CRAFT-GBM reduces per-fit training time by 39.74% with 1.04% lower test loss in the geometric mean across all 18 RQ1 experiments. Under pre-specified feature fraction $f$, Exogenous CRAFT-GBM also yields aggregate training-time reductions up to 13.52 percentage points larger than random feature subsampling, with no increase in aggregate test loss relative to full-feature LightGBM. Accumulated training evidence can thus support adaptive feature search with statistical guarantees while retaining the practical advantages of native gradient boosting.
Read the full abstract
Artificial Intelligence
CHORD: Multi-resolution Compression for Multimodal Time Series
Wearables, mobile devices, and IoT platforms generate multimodal time-series data continuously and at scale, making efficient storage and transfer a primary concern. The same recordings are reused after collection by many downstream models, often at different temporal resolutions, so a compressed representation should be small, useful across resolutions, and produced at a bit rate known in advance. Prior work meets parts of this: symbolic and error-bounded compressors reconstruct accurately but fix resolution at compression time, learned codecs target one signal domain, and discrete tokenizers reconstruct without accounting for rate. We introduce CHORD, a coarse-to-fine compressor for heterogeneous multimodal time series. A residual cascade orders each payload from broad to fine temporal scales and synthesizes corrections through learned shapelet-inspired atoms. A shared pathway supplies a reusable synthesis basis across modalities, while private pathways retain capacity for sensor-specific morphology. Finite scalar quantization fixes the bit width of every modality and scale, so payload size is known before encoding, and scale dropout trains each coarse-to-fine prefix to be a valid reconstruction operating point. Across seven wearable-sensing and forecasting benchmarks, CHORD stays within 0.62 points of uncompressed classification accuracy on average at the 8x target, the smallest gap of any compressed method, with all resolutions decoded from one payload.
Read the full abstract
Computer Vision
LO-World: Unified BEV Latent Modeling for Joint 4D Occupancy and LiDAR Forecasting
LiDAR point clouds and 3D semantic occupancy represent two distinct modalities capturing fine-grained metric geometry and dense volumetric semantics, respectively, holding natural potential for mutual complementarity. However, existing world models typically forecast these modalities independently or cascade them heuristically, missing the opportunity for reciprocal cross-modal regularization. To address this, we propose LO-World, a unified 4D world model that forecasts both modalities within a shared BEV latent space. This single-stream formulation enables dense semantics and metric geometry to mutually reinforce each other, while eliminating multi-pipeline redundancy. Specifically, LO-World couples an efficient hybrid range-voxel autoencoder with a flow-matching generator that caches historical context to reduce iterative sampling latency. Extensive experiments demonstrate that LO-World outperforms prior state-of-the-art baselines, improving occupancy forecasting IoU by 10.9% and reducing LiDAR Chamfer Distance by 30.1%. Meanwhile, it sustains a real-time throughput exceeding $25$~FPS on a single RTX 4090. Our code is available at https://github.com/lzzzzzm/LO-World.
Read the full abstract
Logic in Computer Science
Properties of Public Announcements as Operators: Commutativity, Monotonicity, Idempotence, Fixed Points, and Stabilization
For each formula $\varphi$, we introduce the public announcement operator $U_\varphi$, which maps every model to its updated model under the public announcement of $\varphi$. We then study its algebraic and dynamic properties with a particular focus on commutativity, monotonicity, idempotence, fixed points, and stabilization. Among many results, we show the following. First, monotonicity of $U_\varphi$ is characterized by a family of preservation conditions and equivalence to an existential formula. Second, there exists a formula that achieves every nonzero ordinal as a stabilization rank—the first stage at which a model stabilizes—of some model. Finally and unexpectedly, extinction of every model under iterated announcements of $\varphi$ is equivalent to common unknowability, being always informative when true, and a fixed point view of the Moore sentence. Over S5, those are further equivalent to common unbelievability and being always informative. Other results include characterizations of commutativity and strong commutativity, idempotence as success, properties of fixed points, and finite and $\omega$-stabilization. For analysis, we use PAL (state elimination) and BPAL (arrow elimination), clarifying their subtle differences.
Read the full abstract
Machine Learning
Late but Valuable: Why Deadlines Change the Measured Value of LLM Reasoning
Large language models (LLMs) increasingly plan and run experiments in automated laboratories. These experiments keep running while the model reasons and run on schedules with hard deadlines. If an LLM-generated answer arrives late, the system executes a safe fallback action instead, such as rejecting a sample. This creates a systems problem that standard serving metrics do not capture. An answer that matches the fallback adds no value, and answers that differ from the fallback may require more reasoning time. Reporting answer quality and latency separately can therefore overstate the value of reasoning. We test this on synthetic sample decisions, where the best choice is known exactly and a runtime enforces the deadline. We compare one LLM's fast mode, which answers directly, with its thinking mode, which reasons first. In a prespecified replay of 1,200 new decisions, thinking answers that missed a deadline were worth about five times as much as on-time ones. Multiplying the on-time rate by the mean answer value predicted about twice the value that thinking delivered. This prediction overstated thinking's advantage over fast mode by 0.0241 per decision, and by 0.0191 in a repeat on 300 new decisions. Both 95% confidence intervals excluded zero. In a single-GPU Qwen3-8B experiment using simultaneous request bursts, raising the number of requests from 8 to 56 reduced the share of thinking answers that met a 45 s deadline from 79.2% to 1.4%, although total output tripled. Evaluations of LLM agents for laboratories should therefore measure each answer's value and arrival time together and score the action that actually runs.
Read the full abstract
Statistics Theory
Coverage and shape of conformal highest-density sets: A minimax theory
We study conditional coverage and geometric efficiency of conformal highest-density sets. The excess volume of such a set over the conditional highest-density set, the smallest set with the required conditional coverage, splits into a coverage error and an excess mass. We show that, among procedures whose volume is accurate up to a constant, the coverage error has the minimax rate of nonparametric regression on the covariate, and that the excess mass has the squared minimax rate of conditional density estimation. The usual split conformal construction calibrates these sets with a single threshold, which corrects the bias of the density estimate only on average over the covariate. We prove that its coverage error can then be of the order of the slower density rate. We therefore propose two-stage calibration, which estimates the coverage of the super-level sets of a density estimate by a regression on the covariate and uses this estimated coverage as the conformity score. Since these sets are fixed at each covariate value, their coverage is as smooth as the conditional density, and, under mild regularity conditions, the coverage error attains the regression rate even when the density estimate is inconsistent. The shape of the set still depends on the accuracy of the density estimate: sup-norm accuracy controls it at the minimax excess-mass rate, whereas accuracy in an integrated norm, even of minimax order, does not. With a local polynomial density estimate, two-stage calibration attains both minimax rates up to logarithmic factors. Simulations and an application to bike-sharing demand illustrate the results.
Read the full abstract
Analysis of PDEs
Normalized solutions for Schrödinger-Maxwell equation with Sobolev critical and mass subcritical perturbation
In this paper, we study the Schrödinger-Maxwell equation with critical growth and mass subcritical perturbation \begin{align*} \begin{cases} \displaystyle - \Delta u+\lambda u+ \left(|x|^{-1}\ast| u|^2\right) u =|u|^{4}u +|u|^{p-2}u\quad\quad \text{in} \ \mathbb{ R}^3, \displaystyle \int_{\mathbb{R}^3}|u|^2dx=a^2, \end{cases} \end{align*} where $a>0$ is prescribed, $3<p<\frac{10}{3}$, $\ast$ denotes the convolution and $\lambda\in \mathbb{R}$ appears as a Lagrange multiplier. When the nonlocal convolution term is absent, the above equation can be reduced to the Schrödinger equation. In this case, the previous works showed that it admits a local minimizer (ground-state) and a mountain-pass type solution. By contrast, we surprisingly find that the above equation has no solutions with negative energy provided $a>0$ is sufficiently small. More precisely, we rigorously prove that the above equation possesses a mountain-pass type solution with positive energy for sufficiently small $a>0$, which reveals an essential difference in solution structure between the Schrödinger-Maxwell equation and the classical Schrödinger equation in the parameter range $3<p<\frac{10}{3}$. The main difficulty encountered in this work is the loss of compactness arising from the Sobolev critical term. Furthermore, we explore the dynamical properties of the corresponding time-dependent problem and establish the strong instability of its standing wave solutions. It is worth pointing out that this work refines and extends the existing literature by covering the unexplored parameter regime $p\in(3,\frac{10}{3})$.
Read the full abstract
Artificial Intelligence
Layer Roles Matter: Mitigating Modality Imbalance in Multimodal Emotion Reasoning via Adaptive Layer Profiling
Multimodal Emotion Reasoning (MER) requires perceiving and integrating emotion cues from multiple modalities. Recent advances increasingly leverage Multimodal Large Language Models (MLLMs) to enhance reasoning capacity. However, due to the inherent heterogeneity in conveying emotions across modalities, these models often suffer from modality imbalance and confront flawed emotion understanding. In this paper, we conduct an in-depth analysis on MER models and uncover two key empirical insights. (i) In MER, modality imbalance manifests two distinct failure modes: unimodal over-reliance and insufficient cross-modal utilization, which call for different intervention strategies. (ii) MLLM layers exhibit implicit functional hierarchies in multimodal emotion processing, playing separable roles in stabilizing unimodal representations and integrating cross-modal information. Motivated by these observations, we propose a Layer-Aware Modality Rebalancing (LAMR) framework by explicitly aligning intervention strategies with layer roles. Specifically, we first introduce two quantized metrics for adaptive layer role profiling. In each layer, Modality Entropy (ME) measures the dispersion of modality contributions, and Cross-modal Intensity (CMI) captures the strength of cross-modal interactions. Then, we respectively enhance modality equilibrium in stabilization layers and cross-modal propagation in integration layers. Extensive experiments on two comprehensive MER benchmarks demonstrate that LAMR effectively mitigates modality imbalance and improves emotion reasoning across diverse MLLM backbones, without additional training.
Read the full abstract
Computer Vision
RBA: A Read-out Basis Adapter for Parameter-Efficient Interactive OCTA Vessel Segmentation with SEEM
Retinal disease monitoring with optical coherence tomography angiography (OCTA) relies on segmentation of vessels and the foveal avascular zone (FAZ). Promptable models such as SEEM (Segment Everything Everywhere All at Once) allow click-based refinement, but without adaptation SEEM segments thin vessels poorly and needs many clicks. Fully fine-tuning its 342 million parameters on a few hundred images is costly and risks overfitting, and parameter-efficient methods such as low-rank adaptation (LoRA) usually adapt only the encoder. Yet SEEM reads every candidate mask from one feature map at quarter resolution, too coarse for vessels one to three pixels wide. This study introduces the Read-out Basis Adapter (RBA), whose read-out branch generates 32 image-specific basis maps at half resolution. Each candidate accesses them through its own vector and a zero-initialised projection, so the model starts exactly as SEEM. The encoder branch, a standard rank-1 LoRA, supplies the information needed to tell arteries from veins. RBA adds 0.47 million trainable parameters. Under simulated error-driven clicks, with the ground truth used only to place them, RBA outperforms LoRA, Conv-LoRA andAdaptFormer on OCTA-500 3mmlarge vessels. Averaged over three seeds, it reaches an intersection over union (IoU) of 85.16% and a Dice score of 91.93%, 2.90 and 1.72 percentage points above Conv-LoRA. On the 6 mm subset, one run gives margins of 4.35 and 2.76 points. The gain extends to arteries, veins and ROSE-O but not to the FAZ. In SEEM, adaptation capacity pays more at the mask read-out than in the encoder.
Read the full abstract
Machine Learning
On the Global Convergence of Stochastic Adam for Physics-Informed Neural Networks
The Adam optimizer is widely used to train neural networks with highly nonlinear and nonconvex objective functions. Remarkably, Adam often drives the empirical training loss nearly to zero. This phenomenon has been partially explained by neural tangent kernel (NTK) analyses of gradient descent (GD) and stochastic gradient descent (SGD). However, analyzing Adam presents additional challenges. Its first-moment momentum couples stochastic gradients across iterations, while its adaptive second-moment preconditioner depends on the same stochastic gradient used in the update. In this paper, we establish linear convergence guarantees for Adam in training sufficiently wide two-layer neural networks with either full-batch or minibatch gradients. For full-batch Adam, we construct a Lyapunov function to control the effect of historical momentum. For minibatch Adam, our analysis builds on an exact virtual-iterate representation, a conditional analysis of the dependence between stochastic gradients and adaptive preconditioners, and uniform concentration bounds for the second-moment estimates. We prove geometric decay of the empirical loss for full-batch Adam and geometric decay in restricted expectation on a high-probability event for minibatch Adam. Beyond regression, we extend our convergence analysis to physics-informed neural networks (PINNs) for solving the Poisson equation, where differential operators and separate sampling of interior and boundary points introduce additional challenges. We establish analogous global convergence guarantees for both deterministic and stochastic Adam under explicit conditions on network width and Adam's parameters, characterizing the effects of minibatch size on convergence. These results help bridge the gap between Adam's empirical success and its theoretical understanding, providing sharper convergence guarantees for PINNs than existing analyses.
Read the full abstract
Signal Processing
Capture-Aware Predictive OFDM Reacquisition in Clock-Asynchronous Space-Air-Ground Networks
Reliable orthogonal frequency-division multiplexing (OFDM) reacquisition must validate an intended burst and return its delay and carrier frequency offset (CFO) correc- tions within capture tolerances. In clock-asynchronous space- air-ground integrated networks (SAGINs), shared geometry and timing uncertainty couples distributed observations, while public- preamble replay creates competing peaks. We propose Capture- Aware Predictive Reacquisition (CAPR), a Bayesian receiver that models competition under both hypotheses and marginalizes unknown spatial interference covariance and target complex amplitude before shared-state fusion. CAPR maximizes posterior probability mass within the capture rectangle and constrains continuous L-BFGS-B refinement through a mass-loss budget. We derive closed-form predictive likelihood reductions, an exact finite-cell action optimizer, a conditional approximation regret bound, and finite-sample false-acceptance calibration. Across 150 primary scenarios and five seeds, CAPR achieves 83.8% strict capture at 4.68% realized false acceptance, improving over Gaussian joint-MAP by 12.5 percentage points overall and 54.2 points across all 32 replay scenarios. Diffuse multipath, degraded assistance, and indistinguishable sources delineate its operational boundaries.
Read the full abstract
Statistical Methodology
Semiparametric causal mediation analysis in cluster-randomized trials with post-treatment confounding
Causal mediation analysis of cluster-randomized trials can reveal how interventions affect outcomes through individual and spillover pathways. Most existing methods target natural mediation effects under assumptions that exclude treatment-induced mediator–outcome confounding. We develop a semiparametric framework for interventional mediation effects that accommodates such confounding and within-cluster interference without requiring cross-world independence assumptions. We identify cluster-average and individual-average direct and indirect effects and decompose the indirect effect into individual and spillover components, distinguishing pathways through an individual's own mediator and those of other cluster members. These contrasts evaluate hypothetical interventions on the joint mediator distribution and can differ from treatment effects carried through individual causal pathways represented by natural effects, thereby providing an intervention-oriented perspective that complements natural effects. We derive efficient influence functions to construct estimators using cluster-level cross-fitting, characterize their multiple-robustness properties, and establish conditions for asymptotic normality and semiparametric efficiency. To facilitate implementation, we develop a lower-dimensional reparameterization that reduces the need for multivariate density estimation and integration and allows nuisance functions to be estimated using parametric models or machine learning. Simulation studies evaluate finite-sample performance, and an application to a cluster-randomized trial of cognitive behavioral therapy for chronic pain illustrates the proposed methods.
Read the full abstract
Machine Learning
Spectral Unlearning: Transformer Structure-Preserving Updates for Language Models
Machine unlearning in large language models (LLMs) aims to remove specific learned knowledge, capabilities, or behaviors while preserving model utility. Existing methods have mostly focused on the design of unlearning objectives, while largely overlooking the intrinsic matrix structure of transformer weights. In particular, standard optimizers operate on independent scalar coordinates, thereby ignoring this matrix structure. We propose spectral unlearning, an update framework that combines standard unlearning objectives with matrix-structured optimization. We show that the spectral update corresponds to the steepest-descent direction for transformers composed of linear layers with normalized inputs and outputs. Our approach is applied on top of existing unlearning methods and evaluated across complementary axes of memorization, privacy, and utility, summarized via their harmonic mean. Spectral unlearning consistently outperforms a broad range of baselines and closes most of the gap to a retain-trained reference. Code is available at https://github.com/unlearn-ml/spectral-unlearning/.