Papers

We study the optimisation landscape of the maximum mean discrepancy (MMD) over weighted atomic measures. For the one-dimensional Gaussian kernel, we show that, for any compactly supported target distribution, sufficiently many particles ensure that every local minimum has arbitrarily small MMD. Moreover, any stationary point whose MMD remains bounded away from zero is unstable under an unweighted gradient flow over the particle locations. These results hold both when the weights are constrained to the simplex, and when they are unconstrained but optimised at each step. When the target is itself finitely supported, we further strengthen this result: beyond an explicit particle threshold, there are no spurious local minima, and every spurious stationary point is unstable. To our knowledge, these are the first finite-particle overparameterisation guarantees of their kind for weighted MMD optimisation.

In sectors such as energy, aquaculture and chemical processing, the effectiveness of both mechanistic and data-driven models is frequently limited by poor-quality training data, which obscures underlying physical laws and degrades predictive accuracy. Such quality issues commonly manifest as anomalies in the dataset, ranging from sensor faults to process upsets and incorrect operator inputs. While traditional rule-based filtering can remove a subset of the anomalies, it is unable to leverage contextual information stored as unstructured text, such as engineer comments and operational logs that explain under what conditions the data was collected. We propose a text-informed data filtering method that leverages large language models (LLMs) to integrate unstructured operational data with tabular sensor data. The method consists of two steps: (1) contextual anomaly detection by reasoning over both data modalities, and (2) generation of a binary mask that is applied to filter the training dataset for model development. To investigate the potential benefit, a steady-state soft-sensor regression model, namely a virtual flow meter that estimates well flow rates, is used as the downstream evaluation target. Experiments on a synthetic oil and gas dataset show improved model performance compared to traditional rule-based filtering. Ablation studies show that textual operator comments can provide filtering-relevant signals beyond sensor data alone, with the multimodal approach reducing downstream MAPE from 7.1% to 5.4% relative to sensor-only filtering. This work demonstrates that integrating unstructured operational knowledge with sensor data through LLMs can improve data quality for industrial machine learning applications, offering a scalable alternative to manual data curation.

Existing placement studies for pinching-antenna systems (PASS) usually assume that antenna movement is instantaneous and free. In practice, actuators move at finite speeds and must pause radiation while repositioning. This makes performance depend on consecutive antenna positions, turning placement into an edge-weighted longest-path problem on a temporal trellis. We also show that an obstacle blocks a wider range of antenna positions as it approaches the user, causing sudden performance drops that make multi-step lookahead essential, and that once obstacle height is included every blockage event shadows at least a fixed minimum length of waveguide. We give an exact offline solution with complexity O(TMS) and bound the per-slot loss of a W-step receding-horizon policy by O(1/W). Measurements show that the required lookahead window is set by shadow width rather than by waveguide length. Across 20 random seeds, the exact policy delivers 17.7% higher max-min throughput than per-slot optimization under blockage, and matches it when no blockers are present.

We study the data-driven approximation of control-affine systems, where the drift and each control vector field exhibit a low-dimensional structure. While the ANOVA (ANalysis Of VAriance) decomposition is tailored to low-dimensional structures, a straightforward implementation does not exploit control affinity, rendering the approximation computationally expensive. We propose a coupled least squares approach for given state-control samples, whose ansatz is affine linear in the control by construction and can be realized by a preprocessing of the ANOVA index sets. Based on a Marcinkiewicz–Zygmund inequality for the resulting ansatz space, we prove a probabilistic $L_2$-error bound with explicitly given constants. Separate error control on drift and input vector fields renders the approach flexible to, e.\,g., parameter changes. Finally, we numerically illustrate the identification of the coupling structure via global sensitivity indices on discretizations of a nonlinear advection-diffusion equation, the shallow water equations and the viscous Burgers equation.

Recovering 3D human motion from sparse wearable sensors is challenging due to the inherent noise, occlusion, and instability of real-world tracking signals. Existing approaches are restricted to fixed sensor configurations and, when built on diffusion models, are computationally prohibitive for real-time deployment. We propose OmniPoser, a universal framework that recovers full-body SMPL motion from a flexible combination of heterogeneous inputs, such as 3D keypoints, sparse IMU rotations, head-mounted 6-DoF poses, or any subset thereof, within a single model. OmniPoser introduces a cross-modal masking mechanism paired with a Masked Conditioning Encoder, enabling one architecture to incorporate diverse sensor configurations through a masked design. A decoupled dual-stream generative backbone generates complete motion via Conditional Flow Matching, requiring only a single ODE integration step at inference. Across three settings and seven benchmarks, including in-the-wild Nymeria and Ego-Exo4D datasets, OmniPoser performs on par with state-of-the-art methods while running at over 700 FPS. The project page is available at https://jamesbridgewater51.github.io/omniposer-eccv/.

Typed decision models (TDMs) are increasingly used in applications such as request routing and model selection, making reliable decisions critical to system performance. Fallback options accommodate categories outside regular choices, enabling deferral to human experts or more capable models. While recent works have examined adversarial vulnerabilities in TDMs, their ability to follow fallback rules in non-adversarial settings remains underexplored. We introduce the Paired Fallback Robustness (PFR) Test to evaluate two aspects of fallback reliability: whether models correctly switch between regular and fallback options, and whether they make consistent decisions under logically equivalent definitions of the same rule. We evaluate open-source decision models on three benchmarks, with controlled comparisons across architectures and parameter counts. We find that correct recognition does not reliably transfer to correct fallback options, and errors persist even when the correct category is explicitly supplied.Moreover, logically equivalent formulations of the same fallback rule can also produce substantial disagreement in model decisions. These findings reveal a crucial reliability gap in current TDMs: fallback mechanisms designed to improve reliability can instead introduce errors. It highlights faithful execution of fallback rules and consistency under equivalent definitions as essential requirements for reliable typed decision-making.

As Large Language Models (LLMs) continue to scale, their energy consumption during inference has become a critical concern. While Autoregressive (AR) models are the industry standard, Diffusion Language Models (DLLMs) have emerged as a promising alternative capable of parallel token generation. In this paper, we perform a comparative energy benchmark of diffusion models (LLaDA-8B, Dream-7B) against AR counterparts (Llama-3.1-8B, Qwen-2.5-7B). We analyze the trade-offs between model architecture, KV-cache optimization, and Joules-per-token efficiency. Our results show that at longer sequences (200–600 tokens), diffusion models achieve 2.5–2.7 J/token, approximately 2$\times$ more efficient than AR models with KV-cache (5.3–5.7 J/token). KV-cache provides substantial benefits for AR models at longer sequences (50–70% energy reduction).

In this paper, we consider the endpoint Sobolev regularity of Radon transforms associated with real-analytic functions in the plane. We establish all of the failure of endpoint regularity for real-analytic phases. On the one hand, the critical endpoint regularity $L^p(\mathbb{R}^2)\rightarrow L^p_s(\mathbb{R}^2)$, corresponding to extreme points of the reduced Newton polyhedron of phase function, is true if only and only if $p=2$. For generic extreme points, this failure of endpoint regularity generalizes a result of Christ \cite{christ}. On the other hand, we find that the endpoint Sobolev regularity also fails on a bounded segment of the reduced Newton polyhedron when the bisectrix $y=x$ intersects one of unbounded faces at a relative interior point. In particular, our results imply that the Radon transform is not bounded from $L^p$ into $L^p_{1/n}$ with $p=n/(n-1)$ or $p=n$, where the phase functions are real-valued homogeneous polynomials $S(x,y)=a_1x^{n-1}y+a_2x^{n-2}y^2+\cdots+a_{n-1}xy^{n-1}$ with $n\geq 3$ and $a_1a_{n-1}\neq0$. This answers a problem of Phong and Stein in 1994. Moreover, the main results of this paper also give examples which show failure of critical Sobolev regularity for degenerate Fourier integral operators satisfying finite type conditions formulated in \cite{Seeger93,seeger2}. Based on earlier results and our proof of several remaining cases, we conclude that all of other critical endpoint Sobolev estimates are true for Radon transforms associated with real-analytic phases in the plane. Therefore, all failures of endpoint regularity estimates are clearly determined.

Policy mirror descent (PMD) refers to a class of policy optimization methods associated with different mirror maps. For a broad class of decomposable Legendre mirror maps encompassing many standard choices, Li and Wei [2026] show that the PMD policy sequence converges to an optimal policy from any admissible initialization under any positive constant step size, even when the optimal policy is not unique. In this note, we provide a counterexample with assistance from OpenAI's GPT-5.6 model, showing that PMD with a non-decomposable Legendre mirror map need not converge in policy.

Shannon entropy is a standard measure of ciphertext randomness, evaluated either globally over the byte histogram of the entire ciphertext or locally as the average over non-overlapping blocks of ?xed size. While the global statistic captures only the overall byte distribution, the local statistic is also sensitive to non-randomness that is con?ned to parts of the ciphertext. Both measures are commonly assessed against reference values derived under the assumption of independent uniform bytes. However, STL encodes most numerical data as single-precision IEEE-754 ?oats, and format validity requires them to be ?nite, causing nonuniform byte probabilities and dependence between the two bytes of each exponent pair. We derive the exact mean and variance of entropy under the valid-STL (VSTL) distribution, which assigns equal probability to every admissible sequence of triangle records, and provide second-order expansions that keep the global reference values practical to compute at large ?le sizes. Using these expansions, we establish a two-sided hypothesis test for the global entropy statistic, whose limiting value is lowered from 8 to 7.99991479609 bits. The format constraints also change the local entropy reference: applying the capacity energy criterion to triangle records selects a block size of 1950 bytes instead of 1936 bytes, with an exact reference mean of 7.903106732975 rather than 7.902469316647 bits. Although small, these shifts are su?cient to place the VSTL reference means outside the conventional 95% acceptance intervals at practical ?le sizes, making unadjusted entropy tests prone to falsely rejecting ideal VSTL ciphertexts. Monte-Carlo simulations con?rm all reference values within sampling error. We therefore recommend using the format-speci?c reference values derived here to evaluate STL ciphers.

The interplay between inversion symmetry breaking and superconducting pairing mechanisms is a long-standing puzzle of solid-state physics. Ternary platinum phosphide SrPt$_3$P is an excellent case study in this regard, since it shares the distorted anti-perovskite crystal structure of non-centrosymmetric superconductor CePt$_3$Si but preserves the inversion symmetry center, enabling a direct comparison between the two compounds. Despite intense scrutiny, the superconducting pairing strength, gap symmetry, and magnitude of SrPt$_3$P remain an open question. Here, we study the gap structure of high-quality SrPt$_3$P single crystals with $T_c = 8.6$ K by using point-contact Andreev-reflection spectroscopy. The spectra systematically showed symmetric conductance maxima and a zero-bias minimum, as is typical of fully gapped superconductors, and dips at finite voltage $V_\mathrm{dip}$ due to the local transition of the contact region to the resistive state. In contacts with high resistance, where the dips fall sufficiently far from the gap features, the conductance curves can be well fitted with a single isotropic $s$-wave gap. The temperature dependence $\Delta(T)$ of the gap amplitude extracted from the fit is in good agreement with the one predicted by the BCS theory, with a zero-temperature mean value $\Delta(0)=1.88\pm0.08$ meV, and a mean gap ratio $2\Delta/k_\mathrm{B}T_c = 5.2\pm0.2$.

First defined by Bouchitté and Todinca, Potential Maximal Cliques (PMC) are a well studied notion in graph theory due to its numerous applications. The PMCs of an undirected graph $G$ are defined as the maximal cliques of the minimal triangulations of $G$. Their number can be exponential in the size of the graph. PMCs were crucial in the design of a polynomial-time algorithm which solves Maximum Weight Independent Set on $P_6$-free graphs (Grzesik, Klimošová, Pilipczuk, Pilipczuk, 2020). They also play a central role in the proposal of a polynomial-time algorithm to solve Largest Induced Subgraph of treewidth $\le t$ (Fomin, Todinca, Villanger, 2013). Last but not least, certainly one of the main application of PMCs is the existence of an exact dynamic-programming algorithm to determine the treewidth of a graph (Fomin, Kratsch, Todinca, Villanger, 2008) which uses its list of PMCs as an input. In fact, the enumeration of PMCs stands as the bottleneck for the computation of treewidth with this method: the PMC-enumeration step takes output-quadratic time (Bouchitté, Todinca, 2002) while the dynamic-programming one takes a time linear in the number of PMCs. This two-step procedure was proven to be among the most performing approaches in practice to determine treewidth. Answering a longstanding open question, we propose an algorithm to enumerate PMCs in output-linear time, i.e. a running time linear in the number of PMCs. This offers nice perspectives not only from the theoretical viewpoint - better running time for the PMC-based methods determining treewidth, for Minimum Fill-In and Largest Induced Subgraph of treewidth $\le t$ - but also from the practical side, with the possibility to propose top-performance implementation.

We propose DeltaStreamer, a novel agentic framework for streaming video understanding inspired by how humans observe, remember, and investigate. Streaming video understanding requires a system to retain evidence before knowing which details future queries will need. Existing approaches often obscure event transitions and omit relevant details when compressing video history, and cannot recover missing visual evidence. DeltaStreamer addresses these limitations by combining change-aware memory with query-driven planning and targeted evidence gathering with replay. Before a query arrives, it continuously observes the stream and explicitly records visual changes in structured memory, preserving how events unfold rather than only what is present. When a query arrives, a query-driven planner identifies the required evidence and guides memory retrieval and targeted replay. By revisiting historical frames, the agent can recover visual details omitted during observation and ground its answer in evidence beyond the retained descriptions. This connects continuous observation with active investigation tailored to each question. Experiments on OVO-Bench and StreamingBench demonstrate improvements over representative streaming video understanding methods, with overall scores of 66.07% and 59.53%, respectively. Project page: https://bbbbchan.github.io/DeltaStreamer/

We present OpenTWBench, an open evaluation suite for Taiwan-domain knowledge in large language models, built around two benchmarks. tw-exam-bench contains 255,854 four-option questions across 148 academic subjects, from fifteen years of Taiwan's national examinations published as open data by the Ministry of Examination; answer keys agree across 31,312 duplicated question groups at 99.91%, with all 27 disagreements accounted for. A reviewed second pass withdrew 2,180 items whose stems depend on a figure the text does not carry; all scores use the remaining 253,674. tw-formosa-bench contains 1,696 difficulty-curated questions about everyday and contemporary Taiwan - the Taiwan that licensing examinations never ask about - generated from a community-curated corpus under a discipline in which no model writes a ground truth, and curated against a frozen seven-model pool so that the pool scored at the passing margin (46.9-74.6%, median ~54%); a leave-one-out experiment shows the selection preserves model ranking (near-zero bias). Within six weeks the frontier had cleared that band: of 108 models now on the public board, 22 score above its ceiling, the best 91.6%. We also document how much of a score comes from the apparatus rather than the model. On identical questions, switching answer extraction from pattern matching to an explicit \boxed{} answer moves accuracy from 20.8% to 93.9%; a single placeholder character in a system prompt moved one model from last (58.2%) to first (83.9%) on the same benchmark; doubling the thinking budget from 8k to 16k tokens bought 0.2% while shuffling which questions failed; and a three-judge validation panel proved to be one judge in disguise, two members approving 92-99% of everything shown. We release all datasets, per-item difficulty data, curation manifests and a leaderboard at https://opentwbench.ai

We address inverse rendering with 3D Gaussian Splatting (3DGS) by introducing explicit visibility modeling for relighting as well as material and light decomposition. Our approach augments the standard rasterized 3DGS pipeline with a sparse visibility oracle that estimates transmittance-weighted occlusion for Gaussian primitives from sampled lighting directions. This oracle is implemented using a point-based ray tracer accelerated by a bounding volume hierarchy (BVH) and queried sparsely within the inverse-rendering objectives, while the original 3DGS loop is retained for training efficiency, interactive visualization, and tooling compatibility. Shading follows a physically based rendering (PBR) formulation with a principled microfacet bidirectional reflectance distribution function (BRDF) and low-resolution high-dynamic-range (HDR) environment maps. We evaluate our method on the TensoIR synthetic dataset against baselines based on neural radiance fields (NeRFs), tensor representations, and 3DGS. The results show competitive performance and improvements over several earlier baselines for novel-view synthesis (NVS), albedo reconstruction, and relighting. Ablations indicate that explicit visibility modeling is an important contributor to relighting and material-illumination disentanglement, while pure PBR supervision can favor appearance fitting in NVS. Overall, the findings suggest that sparse, point-based visibility queries effectively complement rasterized splatting for inverse rendering while preserving the practical advantages of the 3DGS ecosystem.

Amorphous alkali oxychlorides are fast ion conductors, but their atomic structure remains unresolved. Total scattering sums all partial pair correlations into one curve that has no unique inversion, so the structure must be modeled. Melt-quench molecular dynamics builds a model from the interatomic potential alone, and reverse Monte Carlo (RMC) builds one from the measurement alone. The two have rarely been compared with all else held fixed. We benchmark both routes and two hybrid schemes on amorphous \ce{Li_{0.9}NbO_{0.9}Cl_{4.1}} against one neutron pair distribution function with one foundation potential. The hybrid schemes put the energy either in the acceptance rule (inline) or in short relaxations between blocks of refinement (alternating). Each refinement starts from a melt-quench glass or from a random packing of isolated units. The glass has the lowest energy and places the first-shell peaks correctly, but carries almost none of the oscillation beyond 6~Å (goodness of fit $R^{2} = 0.29$). Pure RMC fits the measurement to within its counting noise ($R^{2} = 0.99$), but at more than $1.3$~eV/atom above the glass, and niobium loses about half of its first-shell chlorine. Started from the glass, both hybrid schemes reach $R^{2} = 0.92$ within $0.1$~eV/atom of its energy and keep its coordination chemistry. Structures from the two inputs fit almost equally well and differ in energy by a tenth of the RMC penalty, yet they differ in whether their polyhedra are connected. The input is therefore part of the method. We recommend quenching a glass with a foundation potential and refining it against the measurement with a hybrid scheme. The resulting structure carries both the local chemistry and the medium-range order, and it is the starting point for modeling ion transport in these conductors.

This paper studies resolvent expansions at all four spectral thresholds for the discrete Schrödinger operator $H=-\Delta+V $ on $\Z^3$, where $\Delta$ denotes the discrete Laplacian on $\mathbb{Z}^3$ and $V(n)$ is a real-valued decaying potential of $\mathbb{Z}^3$. At the thresholds 0, 4, 8, and 12, we first give a complete classification of the threshold types and derive the corresponding resolvent expansions in weighted $\ell^2$-spaces, including the non-generic cases of resonances and eigenvalues. As a result, we establish the Jensen-Kato type time-decay estimates of the evolution $e^{itH} P_{ac}(H)$ to discrete Schrödinger equations with thresholds singularities on $\Z^3$, with decay $O(t^{-3/2})$ when all thresholds are regular and $O(t^{-1/2})$ otherwise. These results provide a three-dimensional lattice analogue of the Jensen–Kato classical analysis at zero energy threshold on $\mathbb{R}^3$.

Memory-efficient subspace optimization has emerged as a practical approach to reducing the memory cost of LLM training while preserving optimization performance. Previous algorithms typically rely on the leading singular subspace of stochastic gradients, yet the gradient noise may dominate the subspace in later training stages and can prevent convergence. To overcome this limitation, we adopt a residual correction framework, which maintains the optimizer states in the subspace and adds the orthogonal residual gradient to the parameter update. We first integrate this scheme into momentum SGD to develop SMRC, which attains an $O(T^{-1/4})$ convergence rate for a fixed orthonormal basis, matching that of full-space momentum SGD. Our analysis introduces a subspace virtual iterate to resolve the coupling between the momentum and the residual correction. Furthermore, we extend SMRC to a multi-stage variant which periodically recomputes bases to track the evolving dominant subspace and retains the same rate. Since each basis refresh breaks the consistency of the virtual iterates, we devise a boundary correction approach that aligns them across stages. Beyond momentum SGD, we incorporate our framework into Muon and propose PMRC, which applies the matrix sign to the projected momentum. Through a weighted momentum tracking analysis that accounts for the residual gradient, we establish an $O(T^{-1/4})$ convergence rate for PMRC, which matches the rate of full-space Muon. We also design a multi-stage PMRC with periodically refreshed bases, which maintains a convergence rate of the same order. Extensive experimental results confirm the effectiveness and efficiency of our algorithms.

Long-horizon tasks expose a weakness of monolithic tool-use loops: one agent must decompose, execute, and verify a task within a single trajectory. We introduce LAGO, which treats specialized language agents as reusable temporally extended options. A meta-agent retrieves options from a persistent library or spawns new ones, dispatches sub-tasks, verifies results, and admits successful novel options. The library grows across tasks without updating model weights. On 500 SWE-bench Verified tasks, frozen Qwen3.5-35B-A3B in both roles resolves 58.6% under single-attempt evaluation with the official Docker evaluator. On a fixed 150-task subset, we re-run mini-SWE-agent, Terminus 2, and LAGO with the same policy, temperature, task IDs, and evaluator; their resolve rates are 49.3%, 57.3%, and 60.3%, respectively. Cross-model experiments provide evidence consistent with a role for decomposition quality. A 50-task component ablation finds four fewer resolved tasks without decomposition, two fewer with shuffled reused content, and six fewer without the library. These results support reusable agent-options as a practical abstraction for task streams, while leaving component-level causality task- and model-dependent.

The formation and motion of small polarons are ubiquitous in energy materials and influence photoinduced processes. Polaron transport is governed by thermally activated lattice dynamics and rare hopping events, requiring spatial and temporal scales only attainable with machine-learning molecular dynamics, which enables extensive sampling of atomic trajectories and polaron configurations over nanoseconds. Moreover, the Born-Oppenheimer approximation underlying standard ab initio molecular dynamics cannot describe the nonadiabatic coupling between electronic states mediated by nuclear motion, which controls energy exchange and excited-state relaxation. Here, we develop a simulation framework that combines machine-learning-assisted polaron dynamics with nonadiabatic molecular dynamics to bridge these complementary aspects of polaron physics, and determine how different polaron-hopping mechanisms affect excited-state dynamics. The enlarged trajectory space allows us to provide the first direct dynamical confirmation of the adiabatic or nonadiabatic classification of polaron hopping regimes originally proposed on the basis of static Marcus-theory arguments, and to connect individual hopping events directly to nonadiabatic electronic transitions. In bulk TiO$_2$, nonadiabatic hoppings occur preferentially when the energy of the polaron-induced gap state approaches the conduction-band manifold, promoting both electronic de-excitation and, via a Landau-Zener mechanism, excitation from the ground state. In MgO, the predominantly adiabatic character of hole-polaron hopping suppresses these effects, leading to longer charge-carrier lifetimes and an absence of thermally activated excitations. At the rutile TiO$_2$(110) surface, hopping events are rarer than in bulk because the excess electron is strongly stabilized in the subsurface region, leading to correspondingly longer de-excitation times.

Simulating defects of every dimension together is expensive, and the savings of adaptive methods are rarely tested at equal accuracy. We present a variational framework in which fields, dislocation lines and defect densities share one free energy, one dissipation law and exact ledgers of mass and Burgers vector, so that a region can change its representation without gaining or losing defects. On this basis, an allocator switches the dislocations themselves between discrete particles, where they are sparse, and a continuous density, where they are dense. The switch is made block by block from local counts, conserves the Burgers vector exactly and is controlled by a single threshold. For $5\times10^5$ dislocations, the allocator reproduces the discrete model to within 1% in the sparse region at 14-37 times lower cost, depending on the machine; the advantage grows to 81 times at $10^6$ dislocations, and a threshold calibrated on one ensemble meets its error target on an independent one. For an idealized array of dislocation walls in copper, the allocator is 37.5 times cheaper than particle dislocation dynamics and, at almost the same cost, six times more accurate than continuum dislocation dynamics between the walls. In a three-dimensional copper model coupling grain-boundary cavitation with damage, ligaments left at nominal coalescence delay creep rupture by 25-48% (post hoc sweep). By contrast, grids refined at interfaces gain less than twofold once the active region exceeds about 30% of the domain, so representation switching, rather than grid refinement, is the effective form of adaptivity for defect-rich microstructures.

GPU-accelerated first-order methods are increasingly effective for solving large-scale linear programs (LPs), yet LP preprocessing remains predominantly CPU-based and can become costly. A key challenge is that dependencies among presolve reductions and frequent modifications to sparse LP data make efficient GPU parallelization difficult. To address this challenge, we develop GPU-Preprocessor, a GPU-native LP preprocessing framework that integrates folding and presolve. To efficiently implement folding on GPUs, we adapt a randomized partition refinement algorithm to LP folding, using sparse matrix-vector products and parallel sorting to construct stable row and column partitions. For presolve, we introduce a new piecewise-linear reformulation rule that exploits underlying convex piecewise-linear structure to obtain sparser LP representations. We also develop a framework that combines phase scheduling with a Scan-Plan-Apply design to coordinate presolve rules and reduce sparse-matrix reconstruction overhead through batched updates. Numerical experiments on three LP benchmark sets show that GPU-Preprocessor achieves greater average problem-size reductions than PSLP and Gurobi Presolve, with SGM10 preprocessing-time speedups of up to $43.9\times$ and $677.7\times$, respectively. Here, SGM10 denotes the shifted geometric mean with a 10-second shift. On the largest tested quadratic assignment problem LP relaxation, these speedups reach $126.3\times$ and $2{,}207.9\times$. When integrated with the GPU-accelerated LP solver HPR-LP, GPU-Preprocessor achieves SGM10 speedups of $2.5\times$ to $5.0\times$ in combined preprocessing and solving time over HPR-LP without preprocessing. The code is available at https://github.com/PolyU-IOR/GPU-Preprocessor.

Robot-free collection with handheld interfaces enables scalable acquisition of human demonstrations for policy learning. However, existing handheld interfaces often provide limited intuitive force feedback, making contact-rich demonstration collection challenging for human operators. To address this, we present FOMI, a force-feedback manipulation interface with a scissor-shaped mechanism, which transmits object interaction forces directly to the operator's hands through actuation links without force sensors. This intuitive force feedback supports real-time action adjustment during robot-free collection of contact-rich demonstrations. Around FOMI, we build a demonstration collection system that integrates sensor calibration, data acquisition, and synchronization for robot-free collection. By further mapping pose changes and gripper opening of FOMI to robot end-effector poses and gripper commands, the system also supports teleoperation. Across four contact-rich tasks, FOMI improves collection efficiency by approximately 29% and 18% over the best baselines in robot-free collection and teleoperation, respectively. Demonstrations collected with FOMI in both modes enable policies to perform challenging contact-rich manipulation tasks with high success rates.

The aim of this study is to explore methods for how to identify new quasars with different properties with regard to reddening, redshift, the presence of broad absorption lines from outflows, or other rare properties affecting the broad-band colours. We use Gaia-based astrometric pre-selection to study only sources that have zero parallax and zero proper motion within errors. We subsequently match these sources with information from wide-area photometric surveys and build colour-colour diagrams. Based on these diagrams we select targets for follow-up spectroscopy. In the subsequent analysis we also apply observations from SPHEREx to firmly establish redshifts. We secure spectroscopic observations for 24 sources. Of these, three are found to be stars and the remaining 21 are classified as quasars spanning a redshift range from z=0.139 to z=3.732. The majority of the sources located away from the normal quasar locus in colour-colour diagrams are found to be broad-absorption-line quasars. A single source, the lowest redshift source at z=0.139, is reminiscent of the so-called Little Red Dots found at high redshifts with JWST. Using the simple procedure followed here it is possible to locate previously undiscovered quasars with interesting properties, in particular Broad Absorption Line quasars and quasars with unusual extinction.

Let $M$ be a rank-one symmetric space of compact or noncompact type and real dimension $m=kn$, where $k\in\{1,2,4,8\}$ is the real dimension of the underlying division algebra. Meng and Wang proved sharp isoperimetric inequalities for reciprocal sums of the first $p$ or $\ell$ nonzero Neumann eigenvalues, with indices strictly below $m$ in the nonconstant-curvature cases. We keep the whole $m$-dimensional transplanted first eigenspace of an equimeasurable geodesic ball and treat its mass and stiffness forms as a coupled generalized eigenvalue problem. For every bounded connected smooth domain in a noncompact rank-one symmetricspace we prove the full sharp estimate \[ \sum_{j=1}^{m}\frac{1}{\mu_j(\Omega)} \ge \frac{m}{\mu_1(B)}, \] where $B$ is a geodesic ball with $|B|=|\Omega|$; equality characterizes geodesic balls. The Hopf-direction contribution is encoded by a positive, unital, trace-preserving projection on symmetric matrices, and the finite-dimensional closure is supplied by a positive-projection Jensen inequality. In compact type, under the localization hypothesis of Meng–Wang, the same full estimate is obtained whenever $\cot^2 R\ge k-1$, where $R$ is the radius of the comparison ball, as well as in the isotropic endpoint cases and for Hopf-balanced angular second moment. Consequently the complete $m$-term inequality holds for all complex and quaternionic projective spaces under the original localization assumption, and for the Cayley plane when $R\le \arctan(1/\sqrt7)$. In the remaining admissible Cayley interval we exhibit a finite-dimensional obstruction showing that the affine tangent-matrix bounds alone cannot yield the full trace estimate. This obstruction concerns the proof mechanism and is not a counterexample to the unresolved spectral inequality.

Enterprise group chats record project progress, decisions, task ownership, and other organizational information valuable to enterprise agents. Reliable retrieval is challenging because evidence spans interleaved threads, channels, and time, with meaning and authority conditioned on the speaker. Neither RAG's independent chunk retrieval nor memory systems' summary-based consolidation can reliably recover such distributed evidence. We present EGCForest, a novel multi-view hierarchical memory framework for enterprise group chats. EGCForest organizes an immutable message log into three complementary forests: a Channel forest for temporal workflows, an Actor forest for speaker-conditioned context, and an Artifact forest for deliverable lifecycles. Cross-view connections are enabled by a Placement Map and sparse structural bridges. Built on this memory backbone, EGCForest retrieves and assembles evidence by jointly leveraging message-level relevance, hierarchical context from forest summaries, and cross-view links, recovering distributed evidence. Experiments on three enterprise-chat benchmarks show that EGCForest consistently outperforms RAG and memory-system baselines, improving answer accuracy by up to 20.85 points over MemOS while supporting efficient online retrieval.

Euclid data is going to lead to the discovery of approximately 10^5 new strong gravitational lenses, thereby increasing the number of known lenses by a factor of 30. Such an impressive jump has the potential to revolutionise the field of strong lensing, by covering an ever-increasing range in the parameter space occupied by the lenses and lensed sources, and by uncovering rare and precious strong lens systems. The first quick data release of Euclid has already revealed hundreds of lenses, giving us a taste of what the final Euclid strong lens sample might look like. These lenses include a variety of objects acting as deflectors, from isolated galaxies to clusters, from early-type galaxies to edge-on disks. As many as four double-source plane lenses, in which one foreground object lenses two galaxies at two different redshifts, have already been discovered. As more data are acquired, it might become possible to detect the strong lensing signature of dwarf galaxy satellites, and to find lenses with an unusually large number of images. In this paper, we try to anticipate the kind of discoveries that these data will bring along.

Frontier web agents using closed models have high inference and computational costs. Smaller open models can be an alternative, but to be performant, they require training supervision from a human or teacher model. Token-level teacher distillation is not an option for closed teachers as it needs access to the model's logits. The standard approach is to imitate teacher demonstrations offline, but it trains the student on states it has never visited, and in a style it does not write in. We introduce WebKD, where the teacher is the editor of the student's failures. Given a failed episode, the teacher finds the first point of mistake, and rewrites the student's reasoning traces, harness states, and actions from that point onward. The editing reveals that the teacher's corrections can be more surprising to the student than the student's own text. WebKD instead aims to edit minimally. In the data space, the teacher writes each correction in the student's style. In the objective space, the reasoning is weighted by its contribution to the action. A role-asymmetric KL anchor allows the corrected action to move freely while keeping the reasoning near the student's prior. On the TimeWarp benchmark, WebKD students gain about 5.5-10.7 pp over behavior cloning, outcome-reward RL, and teacher editing baselines. The performance transfers to unseen versions and to the live web in the OnlineMind2Web benchmark. WebKD opens a new way of distilling black-box models to improve the capabilities of open-weight web agents.

Understanding and controlling heat flow in low-dimensional magnetic materials is central to thermoelectrics and spin caloritronics. One of the challenges is the absence of sensitive probes of small, field-dependent modulations in thermal transport. Here we demonstrate transverse Seebeck thermometry using exfoliated TaIrTe$_4$ as a probe of thermal conductivity modulated by an external magnetic field on a van der Waals magnet, CrPS$_4$. With a temperature gradient oriented at $45^\circ$ to its principal crystallographic axes, TaIrTe$_4$ exhibits a transverse thermopower exceeding $200~\mathrm{\mu V\,K^{-1}}$ for 900 nm wide and 30 $\mu$m long electrodes, enabling precise detection of variations in the local temperature gradients. Exploiting this with CrPS$_4$ in a nonlocal geometry, we resolve distinct field dependent thermal conductivity changes associated with both the spin-flip and spin-flop magnetic transitions which we attribute to variations in the magnon$-$phonon coupling in CrPS$_4$. At 5 K, the transverse Seebeck response is modulated by more than 8$\%$ by an external magnetic field, demonstrating strong magnetic control of heat transport in CrPS$_4$. This all van der Waals platform offers high sensitivity to modulations of thermal conductivity and a broad adaptability to a wide range of exfoliated magnetic systems.

Reward model overoptimization in RLHF is usually described as reward hacking. We recast it as rent-seeking: the policy, a firm, earns reward from a regulator, the proxy reward model, that true quality does not back. The lens turns overoptimization into measurable quantities, namely rent, a capture index and deadweight loss, and into testable predictions from the economics of regulation. In a fixed, version-controlled testbed with three model families, accuracy-matched proxy reward models and two gold reward models, RL turns 41% to 66% of its proxy gain into rent while gold reward plateaus. Rent follows each regulator's prices. Neither proxy verifies facts, but one pays for confident specifics and the other for displays of diligence, so RL policies learn to fabricate details about nonexistent entities (97% to 100% of probes) or to deflect factual questions (67% to 98%). Kinship between policy and regulator adds a fixed premium of about 0.1 standard deviations but gives the policy no extra leverage under optimization. At a fixed expected fine, frequent mild audits beat rare severe ones, and random audits are as efficient per audit as targeted ones. Bringing the bunching estimator of tax economics to RL training, we also find a first sign that policies learn to stay just below a targeted audit's threshold.