Papers

MathNet-Retrieve asks a retriever to find, for a math problem, a document stating the same problem. An LLM under one fixed prompt writes each gold document and its near-miss distractors; LLM judges filter them. We call this procedure the "recipe", training on pairs built the same way "recipe-matching", and ask how much score it buys beyond the ability the benchmark claims to test. Two models from one base, matched in rows and settings, differ only in the training file: pairs written under the benchmark's published prompt by another vendor's LLM and judge, or computer-algebra-verified pairs with no LLM anywhere. The first leads by 45 R@1 points on the easy tier. By a non-LLM paraphrase control, half to two thirds of that gap comes from the pairs being LLM-written at all: LLM rewrites under two unrelated prompts, with the verified model's negatives, recover 30 and 22 of the 45 points; back-translations with the same negatives recover almost none. The remaining 15 to 25 points appear only under the benchmark's own prompt and vanish on real duplicates no generator wrote, the same problem in two languages. The hard tier rewards the recipe's pair structure, a deep rewrite against a minimal-edit near-miss: LLM rewrites alone score zero on it, attaching negatives unlocks it, and every negative that does so costs cross-language points; the sets scoring highest on it separate near-misses no LLM wrote worse than LLM rewrites with verified negatives. MELD also moves when a model trains on pairs built its way, without losing retention; on SABER-Math the registered attack fails, and the one gain, from its LLM-written summaries, is small but holds at a matched budget. Only on MathNet-Retrieve could we pin an inversion, benchmark score up and real retention down, to one edit of a training file. We release the generator-free duplicate evaluations, the near-miss test and three trained models.

Using diffuse-interface theory, we uncover unexpected phase-change phenomena that are absent from widely used models of evaporation and condensation. In the inviscid limit, the same adjacent thermodynamic states admit stationary evaporation and condensation with equal and opposite mass fluxes, even when the vapor pressure lies below saturation. Viscosity breaks this reversibility but gives rise to multiple stationary condensation rates for the same vapor state. Lattice Boltzmann simulations dynamically recover these states, showing that these phenomena are dynamically realized by the diffuse-interface equations rather than arising solely as mathematical solutions of the reduced interfacial relations. Our results reveal previously unexplored behavior in a well-established theory of liquid–vapor interfaces.

Using Gaia DR3 photometric and astrometric data, we identified the likely members of the open clusters Collinder 110 and NGC 2477, and derived their astrophysical and dynamical parameters. We obtained reddening values of $E(B-V) = 0.43 \pm 0.05$ and $0.33 \pm 0.04$, and ages of $1.82 \pm 0.10$ Gyr and $1.05 \pm 0.10$ Gyr for Collinder 110 and NGC 2477, respectively. Within the uncertainties, the distances derived from the color-magnitude diagrams ($2.09 \pm 0.06$ and $1.38 \pm 0.04$ kpc) are consistent with those obtained from Gaia DR3 parallaxes ($2.08 \pm 0.31$ and $1.45 \pm 0.11$ kpc). Both clusters exhibit large core and limiting radii. The core radius ($R_c$) of Collinder 110 does not follow the expected decreasing trend with increasing evolutionary parameter ($\log\tau_2$). In addition to losing part of its low-mass stellar content, its relatively large $R_c$ may reflect primordial conditions and/or the presence of a possible stellar-mass black hole. The relatively large $R_c/R_h$ ratios, together with their advanced dynamical evolution ($\log\tau_2 = 1.24$ and $1.34$), indicate that significant two-body relaxation and mass segregation have occurred within their central regions. The small values of $R_h/R_j$ ($0.32$ and $0.16$) suggest that low-mass stars have dynamically redistributed within the comparatively large Jacobi radii under the combined influence of internal and external processes. However, their large $R_t/R_j$ ratios ($1.60$ and $1.23$) imply that both clusters are tidally affected and have likely lost part of their low-mass stellar populations to the Galactic field. Despite these effects, their locations in the third Galactic quadrant, where giant molecular clouds are relatively scarce, may explain their survival to their present ages of $1.82$ and $1.05$ Gyr, respectively.

Contact-rich manipulation depends on force sensing that is hard to infer from visual signals alone, both for collecting demonstrations and for training policies. Force-annotated data, however, remains hard to obtain at scale: real-robot collection ties every demonstration to physical hardware, physics simulators report contact forces that deviate systematically from real measurements, and learned world simulators, though scalable and realistic, are vision-only, so operators feel nothing during data collection and the data carries no force/torque (F/T) labels. We present HapticWorld, an interactive world simulator that predicts joint torque together with observations and renders it back to the operator in real time, closing the haptic loop between a human and a learned world model. Across three contact-rich tasks, torque feedback raises data collection throughput by 1.6 times on average. Policies trained on HapticWorld-generated demonstrations succeed in 54/60 real-world trials, approaching the 56/60 upper bound of real-world data, and far exceeding the 19/60 success rate of the vision-only baseline. Moreover, the success rates measured inside HapticWorld closely match real-world evaluation, demonstrating that HapticWorld can serve as a stand-alone F/T-conditioned policy evaluation platform.

We present the JWST NIRSpec 1–5 micron spectra of the Uranian rings and four ringmoons: Puck, Perdita, Portia, and Juliet. We find that the compositions of the rings and moons strongly resemble water-type Kuiper belt objects, showing deep 3.0 micron O–H and water ice features as well as the CO$_2$ band at 4.27 micron. The volatiles show a radial trend of decreasing volatile abundance at closer radial distance to the planet, consistent with recent estimates of the densities of the innermost moons. Given the observed compositional signatures, we suggest two ways to reconcile the history of the small inner moons with that of the regular ones. The first is that the Uranus-tilting collision produced a disk from an impactor formed out of material similar to that which comprises water-rich Kuiper belt objects, and the ringmoons are small undifferentiated remnants of that process. Alternatively, the rings and ringmoons postdate the formation of the regular moons, and formed from the later tidal disruption of a several hundred kilometer sized undifferentiated water-type Kuiper belt object.

Large language models (LLMs) are enticing in their promise to support data visualization (Vis) through faster and simpler workflows for data prep, analysis, and visualization creation. Yet LLMs are notoriously error-prone and not built for data visualization tasks. Few studies have explored LLM adoption among Vis practitioners. To fill this gap, we conducted semi-structured interviews with members of the Data Visualization Society, a global community of data visualization designers. Our findings show that Vis designers actively use LLMs for both creative and technical aspects of the visualization process. A new visualization workflow is emerging, a process we call vibe analysis, analogous to vibe coding. Some key challenges raised by participants parallel those of vibe coding, while others are Vis-specific, like gaps in Vis knowledge and chart verification. This work opens up opportunities for research combining LLM-mediated work with Vis tools that incorporate data visualization guidance, constraints, and best practices.

Agentic AI envisions ecosystems of intelligent agents collaboratively solving complex tasks with minimal human intervention. In such ecosystems, each agent possesses specialized expertise, making effective expert selection central to overall system performance. While most current approaches assume a small number of well-documented models, real-world expertise is far more diverse and cannot be adequately captured through static metadata or hand-written descriptions. We anticipate a future with millions of specialized language models (LLMs), each excelling in different domains or problem types. Rather than relying on predefined capability statements, we propose a retrieval-based paradigm in which an assistant agent infers expertise dynamically by examining models' observable behavior. Upon receiving a user query, the assistant ranks candidate LLMs based on demonstrated competence, enabling efficient and adaptive expert selection. The TREC Million LLM Track operationalizes this paradigm by shifting the retrieval target from documents to expert LLMs. Participants are given a discovery set consisting of queries, answers, and log-probabilities from more than one thousand LLMs and are challenged to infer meaningful expertise representations for each model. Given an unseen test query, systems must then rank the LLMs according to their expected performance, providing the first large-scale benchmark for expertise retrieval in agentic AI.

In this work, we revisit the classic Yu-Shiba-Rusinov (YSR) problem of magnetic impurities in $s$-wave superconductors and uncover a quantum geometric origin of the YSR bound-state structure. The key geometric quantity is the momentum-space average of Bloch band projector over low-energy electrons, whose matrix rank, positive eigenvalues, and corresponding eigenvectors directly control the number, energies, and symmetry representations of the YSR states, respectively. As a result, a superconductor with nontrivial normal-state quantum geometry can host more bound states than its trivial counterpart, even if they share the same electronic dispersion. We apply our theory to monolayer $1H$-NbSe$_2$ and find that the geometric texture of its single superconducting band enforces three symmetry-distinct YSR channels for a magnetic point impurity at Nb or Se sites. Our work opens a route to probing normal-state wavefunction geometry through local YSR spectroscopy.

We study the empirical eigenvalue distribution of a standardized up Laplacian of the Linial-Meshulam model $Y_{q+1}(n, p)$. We first give a definition of a Gaussian–semicircle law: $\mathcal{N}(0, \sigma^2)\boxplus \operatorname{SC}(s\sigma^2)$, and give a combinatorial formula of its moments in terms of pairing partitions. In addition to that, we also prove that the limiting empirical eigenvalue distribution of this standardized up Laplacian follows a Gaussian–semicircle law, in the sense of almost surely weak convergence.

We introduce ROTE (RollOut Testing of Exact memorization), a benchmarking protocol for evaluating symbolic memorization of neural architectures. We study memorization and the extension of symbolic rules in neural sequence models by using sequences whose complexity is regulated by Lempel–Ziv–Welch (LZW) compression. Under ROTE, each architecture is trained as the same finite-context conditional predictor and is evaluated using teacher-forced one-step prediction as well as closed-loop rollout on the withheld symbols. Following a shared prediction-and-rollout evaluation routine, the benchmark evaluates gated recurrent, minimal recurrent, attention-based, and hybrid recurrent-attention models with their native computational characteristics preserved. Beyond standard predictive metrics, the benchmark reports normalized string distances, training time, memory usage, and parameter count across an LZW-complexity sweep. The study establishes a connection between the complexity of algorithmic sequences and the memorization capacity of neural architectures, revealing the trade-offs involving memorization quality, rollout stability, and computational expense. Our software and reproducible experimental code can be obtained from https://github.com/nla-group/rote.

Semantics for systems with names and data often depends on finitely many distinguished values. The theory of finitely supported structures treats this dependence through invariance under permutations fixing a finite context. We develop this approach to semantics over arbitrary permutation groups and arbitrary infinite sets of atoms. The central difficulty is that finitely supported predicate spaces need not be complete lattices. We prove that, for predicates valued in a complete lattice with trivial atom action, every monotone finitely supported transformer nevertheless has least and greatest fixed points. These lie in the complete lattice determined by the transformer's context and coincide with the fixed points of every compatible monotone ambient extension equivariant under its stabilizer. Support-transfer bounds track dependencies through semantic constructions, while uniform finiteness yields finite convergence. For Boolean predicates, $m$ context-stabilizer orbits on the carrier suffice for convergence after at most $m$ iterations, even when the supported predicate lattice is orbit-infinite. We apply these results to automata, operational and modal semantics, abstract interpretation, and resource rewriting. Ultrahomogeneous atom structures in finite relational signatures yield finite cell representations, illustrated by an authorization monitor. The resulting account separates semantic existence, finite convergence, and effective computation.

Organizations may perceive disorder in their environment as a threat which can impede the attainment of desired goals. In such a situation, organizations may become single minded, obsessed, and rely on brute force to attempt to bend the external environment to their will. These efforts may be enabled by perceived power imbalances between the organization and other actors in the environment, but they may also be stymied by the complexity of other organizations in the institutional field. This paper examines such an example in Soviet Kazakhstan from 1929 to 1933, when the All Union Communist Party pursued a policy of sedentarization of the nomadic people. Forcibly migrating nomads to collective farms under a brutal policy of repression, the Soviet bureaucracy at all levels saw no policy as off-limits to achieving its goals but continued to run up against resistance from the complex organizational forms inherent in Kazakh nomadism. Using archival documents, testimonies from Kazakhs, and arrest records from this time period, I show just exactly how the Soviet apparatus pulled the devil from the machine and how, ultimately, the machines were doomed to failure. However, the human cost of forced migration continues to linger.

Hybrid quantum methods have received limited study for resource-constrained facial biometrics. We present a hybrid quantum-classical facial recognition pipeline designed to run on standard computing hardware. Images undergo gamma correction, contrast enhancement, and principal component analysis before their features are encoded into an eight-qubit variational quantum classifier. Classical image matching then performs recognition. In experiments with 50,000 images, comprising 25,000 faces from CelebA and 25,000 non-face images from CIFAR-10, the method outperformed the reported CPU-trained FaceNet baseline in accuracy and training efficiency. The pipeline was also evaluated on GPU and quantum hardware. In an attendance monitoring deployment at Mahindra University in collaboration with Lloyds Technology Centre, CPU inference took 0.2 to 0.5 seconds per person, and the system remained robust to the use of spectacles. These findings support the feasibility of deploying hybrid quantum methods for facial recognition on existing CPU hardware.

Reference-based multiple imputation is used in longitudinal clinical trials to assess sensitivity to assumptions about outcomes unobserved after intercurrent events. Most existing methods target continuous outcomes and use multivariate normal working models. Scheduled count outcomes require a model that preserves integer support, skewness, overdispersion, longitudinal dependence, and an exposure-based rate interpretation. We propose visit-specific negative binomial (NB) margins linked by a Gaussian copula. Covariate-adjusted log rate models define the margins, while the copula captures longitudinal dependence. For missing outcomes in the active arm after a prespecified event, termed the trigger, assigned arm continuation retains the fitted active arm marginal mean, jump to reference uses the corresponding reference arm marginal mean, and the intermediate rule interpolates between these means on the log rate scale. Combined with the copula, the resulting NB margins determine an imputation distribution conditional on the retained history. We account for discreteness through randomized probability integral transform augmentation and fit the model with a custom Metropolis-within-Gibbs sampler. In targeted simulations, the model recovered the generating marginal and dependence parameters with little bias. In simulations with incomplete data, treatment effect estimates were closest to the corresponding complete data estimates when the imputation rule matched the mechanism governing outcomes after the trigger. We illustrate the method using repeated incontinence episode counts from a published trial in overactive bladder. Estimated rate ratios comparing active treatment with placebo remained below 1 under all reference-based assumptions, with modest attenuation toward the null and the greatest separation between rules at the final visit.

Ensemble nitrogen-vacancy (NV) centers are widely used for quantum sensing and magnetometry, where performance depends on the accurate implementation of microwave pulse sequences. In these systems, inhomogeneous broadening and short coherence times limit reliable multi-pulse control, while hardware imperfections and off-resonant excitation distort implemented rotations, making direct calibration of arbitrary gates tedious. To address this, we implement a control framework with a ZXZXZ gate decomposition which uses fixed-angle, phase-parameterized pulses, enabling universal single-qubit control from a single calibrated $ \pi/2$ primitive ($X_{\pi/2}$) along with virtual Z operations. We compare a standard pulse with a composite pulse designed for high-power operation in a regime where the Rabi frequency is comparable to the hyperfine splitting, leading to non-uniform evolution across hyperfine levels. Using pre-determined combinations of $\pi/2$ pulse sequences, we observe that standard pulses rapidly deviate from ideal behavior under repeated application, while the composite pulse suppresses this error accumulation by more than 70$\%$ for longer sequences despite its longer duration. In contrast, randomized benchmarking yields similar decay factors for both implementations, highlighting its limited sensitivity to coherent control errors. These results establish a practical route to robust control in short-coherence ensemble systems and emphasize the importance of pulse-level characterization.

Strongly paired medical image translation contains a substantial component that can be predicted directly from the source. The current reality is that pure deterministic prediction can smooth away fine detail, while generative models can recover detail but may also introduce unnecessary or potentially harmful variation. We propose PredRA, a fast framework that uses the deterministic prediction as a stable reference and further extracts additional deterministic components from the residual during generative refinement, thereby improving fidelity while maintaining perceptual quality. The goal is simple: we view the entire residual-based generative process as an optimization problem and derive a practical solution that recovers useful residual detail through controlled refinement while keeping the prediction close to the paired target. Mechanism studies further show that useful residual information follows structured patterns, but its usefulness is difficult to estimate reliably at the voxel level. We therefore globally control how much of the residual refinement is added to the deterministic prediction, thereby reducing the accumulation of unnecessary uncertainty. PredRA therefore combines deterministic component extraction with controlled stochastic refinement for fast, fidelity-preserving, and perceptually strong medical image translation. We validate the approach across multiple real-world datasets, showing that controlled refinement improves fidelity to the paired target compared with full residual refinement while retaining much of the perceptual benefit of generative modeling. PredRA achieves competitive or superior performance to substantially larger state-of-the-art models with 1.4-11.9x fewer total parameters and 3.1-26.2x fewer trainable parameters, while its 32-step flow sampler requires 31.25x fewer sampling steps than the matched 1000-step DDPM.

Medical report generation has made significant progress with the rise of modern vision-language models and the growing availability of large-scale medical datasets. However, hallucinations remain a major challenge, largely due to the limitations of supervised fine-tuning (SFT), which prioritizes lexical similarity to reference reports rather than clinical correctness. While reinforcement learning has shown strong performance in domains with verifiable rewards such as mathematics and code generation, its application to open-ended medical tasks remains limited. Existing work focuses on evaluating final answers, overlooking the model's reasoning, despite evidence that flawed reasoning can degrade overall performance. In this study, we propose a framework that verifies the model's reasoning process by integrating anatomical regions, bounding boxes, and region-level textual descriptions. We design spatial and factual reward mechanisms to ensure that the model's reasoning is both visually grounded and factually accurate. Starting from Qwen3-VL-8B-Instruct as our base model, we adapt it to the medical domain using supervised fine-tuning, introduce reasoning capability through a cold-start SFT stage, and refine it with reinforcement learning. We find that RL provides performance gains beyond those achievable through SFT alone, and that jointly verifying both reasoning steps and final outputs yields larger improvements than verifying either in isolation. We further identify multiple modes of reward hacking in the RL stage. Finally, the model's structured think traces enhance interpretability, making its outputs easier to audit for clinical use.

We study the divisibilities of Tamagawa numbers $ c(E) $ of elliptic curves $ E $ over global function fields $ K $ in terms of their torsion subgroups $ E(K)_{\operatorname{tors}} $. In particular, for a non-isotrivial elliptic curve $ E / k(t) $, where $ k $ is a finite field of characteristic greater than $ 3 $, we prove that $ |E(k(t))_{\operatorname{tors}}|^2 $ divides $ c(E) $, except possibly in four exceptional torsion families. More specifically, we give a complete characterisation of divisibilities for $ c(E) $ in each torsion family, and provide explicit examples to prove that they are best possible. Over a general global function field, we also prove that a rational point of prime order $ 5 \le N \le 101 $ on $ E / K $ forces $ N^2 $ to divide $ c(E) $, which motivates our result. Finally, we formulate a conjecture on the leading coefficient of the $ L $-function of $ E / K $, motivated by the integrality of Birch–Swinnerton-Dyer quotients.

We study a fixed-feature solver, referred to as Spectral-ELM, in which Chebyshev–Gauss–Lobatto (CGL) differentiation matrices are applied to the nodal values of an extreme learning machine (ELM) trial function. Curved domains are treated using a mapped multi-patch formulation. Each curved quadrilateral patch is represented as the image of a reference square, physical derivatives are computed from the associated metric terms, and adjacent patches are coupled by enforcing continuity of the solution and its normal flux. Local QR orthogonalization is used to reduce near-linear dependence among the discrete features. The numerical study includes comparisons with direct CGL collocation, an analytically differentiated ELM, a global Random Feature Method, and a TransNet sampling strategy. It also includes a matched KdV test, a small three-dimensional example, and a Poisson problem on a curved domain containing a hole. For a smooth elliptic problem on a square, direct CGL collocation gives the smallest error. When applied to the same fixed-feature trial space, spectral and analytic differentiation produce comparable errors. For the curved-domain problem, mapped direct CGL collocation and mapped Spectral-ELM achieve similar accuracy with comparable numbers of unknowns. The present formulation is particularly well suited to smooth PDEs in one to three dimensions, where tensor-product grids enable accurate high-order discretizations, while sparse-grid and dimension-adaptive strategies offer promising directions for extension to higher-dimensional problems.

Large Language Models (LLMs) perform well on medical examinations and question-answering benchmarks, but remain unreliable on medical calculation tasks that require exact numerical outputs. These calculations support high-stakes decisions such as medication dosing, organ-function assessment, and prognostic scoring, for which even small errors can have serious clinical consequences. We introduce MedCode, a framework that improves medical calculation by training LLMs to generate embedded executable code. Given a clinical context, the model identifies the relevant calculator, extracts its input variables, and produces a script that delegates arithmetic operations to a deterministic interpreter. Executing the script returns the calculated value together with an explanation and the appropriate unit. We construct supervised fine-tuning (SFT) and preference datasets from the MedCalc benchmark and additionally curate a dataset for calculation tasks in Intensive Care Unit (ICU) scenarios. We further propose weighted Direct Preference Optimization (wDPO), which adaptively emphasizes preference pairs that are difficult for the model to distinguish. Experiments with LLaMA3-8B, Qwen2.5-7B, and Mistral-7B show absolute accuracy gains of 20–30 percentage points, demonstrating the effectiveness of embedded code generation for medical calculation.

By definition, a dual reflection group is a finite group which grades an Artin–Schelter regular algebra $A$ such that the identity component $A_e$ is again regular. In this paper, we completely classify the nonabelian dual reflection groups for regular algebras of dimension $3$: we first classify all possible nonabelian gradings which refine the natural $\mathbb{N}$-grading of the algebra, and then determine which of these gradings yield dual reflection groups. We show that none of the cubic algebras admit dual reflection groups, while for quadratic algebras, the dual reflection groups arise either from Ore extensions of two-dimensional regular algebras, or from the Sklyanin algebras $S_{q,0,1}$. In the Sklyanin case, the dual reflection groups form a new infinite family of groups $\Delta_n$ of order $27n^3$, answering a question of Goetz, Kirkman, Moore, and Vashaw. We also extend the four-dimensional examples of the same authors to two new infinite families of dual reflection groups. Finally, we show that an AS regular algebra admitting a dual reflection group need not have binomial relations, answering another of their questions. However, we conjecture that the relations must be binomial when $A_e$ contains no elements of degree $1$, and we prove this for nonabelian gradings in dimension $3$.

Enterprise email agents must combine information retrieval, structured state changes, temporal reasoning, and multi-step coordination. Recent agent benchmarks include productivity tasks, but few center on typed email workflows in a self-contained environment. We introduce EmailBench, a benchmark of 206 email and productivity scenarios across 16 task categories. The benchmark couples a typed email API specification with provider-neutral naming, a deterministic synthetic Enron-inspired corpus, and a scenario suite whose topic selection was informed by aggregate task-intent telemetry from an interactive prototype. Its hybrid evaluation protocol combines 258 executable static assertions with 211 LLM rubrics. We evaluate eight LM configurations on a fixed single-user corpus. The best-performing configuration passes only 33.5% of scenarios despite 99.7% of its tool calls completing without an observed API failure, with pass rates varying substantially across task categories. This gap shows that valid tool execution is not equivalent to task completion. EmailBench provides a self-contained environment for end-to-end email-agent evaluation, with broader tool coverage, multi-persona testing, and repeated-run evaluation as future work areas.

We present an end-to-end QP-based policy framework that enables systematic policy construction with minimal domain-specific design, while preserving the transparency and interpretability of model-based control. The proposed policy representation supports both domain-randomized model-based auto-tuning, where policy parameters are optimized over distributions of disturbances and model variations, and black-box policy construction, where the problem is formulated in terms of a (possibly) unknown model without requiring explicit notions of states, inputs, or the underlying system dynamics. We establish connections to existing policy representations and control paradigms, including robust control, multilayer perceptrons, and robot learning, and interpret our end-to-end QP policies as a common abstraction of these approaches. We validate the resulting framework in both simulation and hardware, demonstrating a broad range of applications spanning robustness, automatic policy tuning, and control under unknown system dynamics.

Vision-Language-Action (VLA) models have shown strong performance on robotic manipulation, but they often struggle to generalize to unseen tasks, configurations, and long-horizon settings. A key challenge is that VLAs overfit to training scenes and fail to follow novel language instructions. Off-the-shelf vision-language models (VLMs) often provide stronger generalization, but cannot directly control robot actions. To combine the common sense of VLMs with VLA control, we propose Guided Trace VLA (GT-VLA), a steerable framework that accepts guidance from an external generalist VLM through trace-conditioned action generation. GT-VLA uses a generalist model to identify semantic guidance for the current skill, converts this guidance into a 2D visual trace, and conditions its action policy on the resulting trace-rendered observation. This design separates semantic target acquisition, trace generation, and low-level action execution, allowing high-level guidance to propagate to robot actions. GT-VLA uses a Mixture-of-Experts architecture with skill-specific trace and action modules for robust execution. We evaluate GT-VLA on LIBERO and a physical robot platform, showing improved generalization over recent VLA baselines in both settings. The code and additional supplemental materials are available on our project website at https://ivaniz.github.io/gt-vla/.

AI agents can now formalize entire textbooks and major theorems in proof assistants such as Lean, but current efforts are typically centralized: a single team runs all agents and bears the full computational cost. We introduce Choir, an open protocol for distributed formalization. Choir decomposes a project into tasks that can be completed by independent contributors, each running their own agent with their own LLM subscription, while coordinating entirely through the project's GitHub repository. To support open participation, every contribution is checked by a deterministic gate before merge. Choir supports Lean 4, Isabelle, and Rocq, and is open source and modular, allowing projects to replace individual components or extend the protocol.

Selecting a data encoding is a central and poorly tooled decision in quantum machine learning. The feature map fixes the geometry of the Hilbert space, the expressibility of quantum kernels, and whether the circuit can run on near-term hardware. This paper presents Quantum Encoding Agents, an open-source system that turns encoding selection into a natural-language interaction. Given a dataset and an optional task description, it profiles the data, applies a hardware-aware policy using the gate-error threshold p* approximately 10^{-3}, returns a copyable Qiskit circuit with a justification in Portuguese or English, and scores the quantum kernel by Kernel-Target Alignment (KTA). When a numerical matrix is available, it estimates the correlation fractal dimension D_2 and selects original columns with FD-ASE, using q* = max(2, ceil(D_2)) as a qubit budget so that angle and IQP maps are not proposed at a width where the fidelity kernel has collapsed. The service implements seven encoding families: amplitude, angle, dense angle, IQP, basis, data re-uploading, and custom feature map. On benchmark datasets, KTA separates these encodings under realistic NISQ constraints.

Constrained general-sum dynamic games are a popular formulation for highly interactive multi-agent planning problems. In recent years, Generalized Nash Equilibrium (GNE) solvers have achieved real-time performance for small dynamic games. However, solution speed still remains a bottleneck, and controlling even a small number of agents (e.g., more than four) in a dynamic task remains elusive. In addition, Newton solvers that focus on the first-order conditions are vulnerable to non-Nash saddle points, limiting the usefulness of the provided solution. In this work, we propose a fast and versatile interior point solver for constrained dynamic games, along with a computationally efficient second-order correction that increases the probability of converging to a local GNE solution in constrained dynamic games. The performance of the proposed method is evaluated on numerical benchmarks and a physical experiment involving scaled race cars.

Modern LLM post-training composes supervised fine-tuning (SFT), reinforcement learning with verifiable rewards (RLVR), and on-policy distillation (OPD) into multi-stage pipelines, yet these stages are typically designed and evaluated in isolation. We show that this composition is consequential: a stage that improves the current model can make the next stage less effective. Through controlled experiments with Qwen3 models on math and science reasoning, we first characterize OPD across nine student-teacher pairs spanning 2x to 53x parameter ratios and show that OPD effectiveness depends on student-teacher compatibility rather than teacher scale alone. The surrounding stages of OPD reshape this compatibility in three ways: (1) A brief SFT warm-up improves subsequent OPD, while an RLVR-strengthened student regresses under distillation from the same teacher. (2) Adapting the teacher with RLVR raises downstream OPD accuracy in proportion to the capability it adds. Following these two interventions, we find that combining teacher adaptation and student warm-up alone raise average OPD accuracy from 29.2% to 43.8% (50% relative improvement) after the same number of distillation steps, with additional preparatory training. (3) At comparable accuracy, OPD leaves a stronger initialization for downstream RLVR than SFT, with a gap that widens as RL compute scales. Our results suggest that each post-training stage should be chosen not only for the capability it adds, but for the learning interface it creates for the next stage.

Here, we isolate how Cu concentration and Jahn-Teller distortions control the out-of-plane lattice parameter of pseudomorphic MgCoNiCuZnO-derived high-entropy oxide films on MgO. We prepare nine compositions spanning 0-20% Cu, verify target composition and phase by X-ray fluorescence and diffraction, and grow films by pulsed laser deposition across multiple substrate temperatures. At 500 {\deg}C, the lattice parameter increases nearly linearly with Cu content; at 300{\deg}C, it decreases nonlinearly, producing a 2.6% difference between temperature regimes for equimolar MgCoNiCuZnO at our growth conditions. We attribute this change in lattice parameter to temperature-dependent coupling between Cu-driven Jahn-Teller distortions and Co valence change. While prior work identifies Co valence as the primary driver of the MgCoNiCuZnO lattice response with temperature; our results establish Cu Jahn-Teller distortion as an equally important structural control parameter.

Humans rely on gaze, head movements, and visual cues to attend to speakers in noisy environments, yet auditory attention decoding (AAD) has been studied primarily using electroencephalography (EEG). We introduce the Multimodal Auditory-attention Egocentric Speech-TRacking Open (MAESTRO) corpus, the first AAD dataset to simultaneously record EEG, eye gaze, pupillometry, egocentric video, and head inertial measurement unit (IMU) data. MAESTRO includes four competing speakers and background noise across multiple signal-to-noise ratio (SNR) conditions, enabling attention decoding under realistic listening scenarios. Through a four-speaker attention decoding benchmark, we show that combining behavioral and physiological signals improves decoding performance over EEG-only approaches, enabling future advances in multimodal auditory attention decoding. These findings open the door to new applications, analyses, and methodological advances in multimodal AAD. The complete dataset is publicly available at https://huggingface.co/datasets/aspire-osu/maestro-eeg-dataset . The official code repository is available at https://github.com/ASPIRE-OSU/MAESTRO .