Papers

Researchers increasingly use generative AI, particularly large language models (LLMs), to automate tasks across the research pipeline. We study the reliability of these tools at the reading, classification, and synthesis of large bodies of academic literature. We frame LLM-assisted literature reviews as a measurement problem, treating models as measurement systems and tracing how their errors affect downstream conclusions. As a test case, we use three different implementations of ChatGPT to identify and extract metadata from economics papers that use rainfall as an instrumental variable. We benchmark each implementation against a subset of human-labeled evaluation data, and then deploy those implementations to extract metadata from the full corpus. The LLMs perform well on binary classification, but performance deteriorates as tasks demand greater contextual interpretation. More importantly, how much researchers can rely on model outputs depends not only on the complexity of the reading task but also on the type of claims the data is asked to support. The same amount of measurement error substantially affects paper-level claims while having little effect on broader claims about the literature. Measurement error in LLM-generated data is thus most consequential at precisely the level of detail that constitutes an LLM's principal value added over human reviewers. We conclude that standard model performance metrics are informative about the quality of generated data but do not by themselves establish the credibility of downstream inference. Researchers must also evaluate whether substantive claims are robust to the measurement system used to generate the underlying data.

As the primary structural component connecting the engine to the wing or fuselage, an aircraft pylon requires optimized structural efficiency; this paper provides topology optimization of multimaterial pylons using the generalized shape function (gSF) approach. The gSF method uses $n$ natural-coordinate design variables per element to provide optimized designs up to $2^n$ distinct material phases while promoting close to discrete material layouts in conjunction with the density and formulated Heaviside projection filters. Pylon structural compliance is minimized subject to volume constraints. Exploiting the geometric features of a typical pylon structure, multimaterial evolution is performed on a corresponding 2D design domain representing the midplane, with up to 14 candidate materials. The optimized two-dimensional layout is then extruded to achieve the corresponding three-dimensional optimized pylon structure. The Method of Moving Asymptotes is employed to achieve the final design variables. The resulting convergence histories exhibit smooth and stable objective minimization. The results highlight the capability of the multimaterial topology optimization framework to effectively optimized aircraft pylons with multiple candidate materials, without requiring a considerable expansion of the design variable set.

We investigate a nonrelativistic approximation to general relativity that does not require the metric coefficients to be weak perturbations of the Minkowski metric. Instead, we assume a differential hierarchy in which the contributions to the curvature containing second derivatives of the metric dominate over those quadratic in its first derivatives, while the metric-dependent coefficients multiplying the retained terms are kept fully nonlinear. Since this separation is not covariant, we formulate the approximation in harmonic coordinates, where it corresponds to retaining the principal part of the reduced Einstein equations. We show that the truncated Einstein tensor satisfies the contracted Bianchi identities up to the order neglected in the derivative hierarchy, ensuring compatibility with energy–momentum conservation at the retained order. The standard Newtonian equations are recovered in the weak-field regime, whereas outside it the dynamics retains a nonlinear dependence on the lapse and spatial metric. In particular, null geodesics probe the spatial geometry at leading order, suggesting gravitational lensing as a natural setting in which observable departures from the standard weak-field description may arise. We illustrate the construction with a static spatially isotropic configuration and discuss the scope and limitations of the approximation.

We consider the global dynamics of finite-energy solutions to the $2$-equivariant harmonic map heat flow from $\bbR^2$ to $\bbS^2$ and $\dot H^1$-bounded radial solutions to the energy-critical nonlinear heat equation in dimension $6$. Building upon soliton resolution, we obtain a classification of their multi-bubble dynamics. First, we prove that all such solutions are global. For the harmonic map heat flow, the number and sign of the bubbles are determined by the energy class of the initial data, while for the nonlinear heat equation the bubble signs necessarily alternate. In both models, the largest scale is determined, up to a positive multiplicative constant, by the spatial tail of the initial datum. The remaining scales concentrate according to exponential and iterated-exponential laws determined by the largest scale. Finally, global bubble trees with an arbitrary number of bubbles exist for both models. For the harmonic map heat flow, this follows directly from the classification; for the nonlinear heat equation, we construct alternating bubble trees with any prescribed radial $\dot H^1$ tail, thereby realizing the scale laws arising in the classification.

In-context learning (ICL) enables large vision-language models (LVLMs) to perform tasks by following patterns from in-context examples, yet its potential to amplify societal biases remains underexplored. We systematically investigate how ICL influences gender bias in LVLMs through VL-BICLE, an evaluation framework comprising six ICL settings, three tasks, and four datasets. Our experiments on six LVLMs reveal that gendered ICL demonstrations act as a directional force, shifting model bias toward the demonstrated gender through a cross-gender mechanism that disproportionately degrades performance on the opposite gender. This effect appears in image captioning and pronoun prediction but not in visual question answering, indicating that gendered ICL influences bias only when the task output involves gendered language. Similarity-based retrieval methods inherit the training pool's gender imbalance and offer no debiasing advantage, while standard quality metrics remain blind to these bias shifts. To mitigate this bias, we replace real in-context images with synthetic ones from stable diffusion models while keeping captions unchanged. This simple intervention reduces gender bias without degrading caption quality.

Automated assessment of the Critical View of Safety (CVS) in laparoscopic cholecystectomy requires both recognition of the three CVS criteria and anatomical grounding in small, rare, and often occluded hepatocystic structures. Learning-based methods differ in the anatomical information they use, from image-level classification to detection, segmentation, or graph-based reasoning, yet grounding the safety-critical anatomy remains the main bottleneck. We propose CasCVS-Net, a staged multi-task cascade that jointly performs object detection, semantic segmentation, and CVS assessment, trained on the Endoscapes dataset. The model couples the tasks through predicted anatomy: predicted boxes guide segmentation, and predicted masks provide region-level features for CVS classification, so CVS assessment at inference uses only model predictions rather than ground-truth annotations. To reduce optimisation instability in this coupled setting, training progresses from detection to detection-segmentation and then to the full three-task cascade, followed by task-wise fine-tuning. Evaluation on the public unseen test set shows that CasCVS-Net improves over matched single-task baselines on all three tasks, achieving 32.0 detection mAP, 46.8 semantic mIoU, 15.3 rare-anatomy mIoU, and 67.2 CVS mAP. It outperforms the state-of-the-art LG-CVS and SV2LSTG by 6.3% and 4.5% relative CVS mAP, respectively, corresponding to 4.0 and 2.9 mAP points. These results show that staged task coupling through predicted boxes and masks improves anatomical grounding for CVS assessment, particularly for rare hepatocystic structures.

We study the sensitivity of the associated Higgs production process $pp \rightarrow ZH$ to anomalous Higgs couplings within the Effective Field Theory framework. Event samples are generated using $ \text{MadGraph5\_aMC@NLO}$ at $\sqrt{s}=13$ TeV using the UFO model HPO\_ewk\_prod\_NLO, and analyzed using a dedicated Python analysis of HepMC events. We investigate several kinematic observables and derive constraints on the EFT parameter $\epsilon_{ZZ}$ using a $\chi^2$ statistical analysis. A 95% confidence level interval $-0.026 < \epsilon_{ZZ} < 0.383 $ is obtained from the simulated data.

A tabular foundation model must discover which distinctions matter for each new table without updating its parameters. We develop in-situ representation refinement: support labels guide changes to the episode's representations, improving the information available to later queries. A regularized leave-one-out objective yields a support correction and its query extension. The leading term separates attention-based reading from state-dependent scaling, motivating RefineICL: an attention-gated, FFN-free contextual stack with selected low-rank feature interaction and typed memory. A direct intervention tests the role of evolving support states: removing one intermediate support update while preserving the block's query output increases final query cross-entropy in all 72 tested episodes. RefineICL-L24 reaches 0.93836 OVR-AUC and 0.87173 accuracy on AMLB29. A benchmark-informed continuation reaches 1644.8 Elo on the 38-dataset TabArena snapshot, 31.4 Elo above TabPFN-3 under the same evaluation. It also improves all four reported metrics over TabPFN-v3 on both TabZilla views. In a matched 100K-update depth grid, an expanded FFN gives no consistent validation benefit and uses 60.2% more peak inference memory at L8. These results connect learning within a forward pass to representation refinement and show how this view guides a competitive, memory-efficient model.

Aggregate accuracy summarizes how often a model is correct, but not which judgments remain correct when requests change. We evaluate Jev and nine language models on ContractNLI, with repeated requests that preserve the contract and target while varying visible hypotheses, requested answers, and output order. A four-model follow-up adds an equal-call repetition control and coverage of the remaining test contracts. Corrections and regressions can cancel, leaving unchanged accuracy despite changed decisions; consistent answers can also preserve errors. Majority voting over recorded responses yields small, uncertain gains, with repeated wrong answers accounting for much of the observed error. Gold-class diagnostics further reveal different error profiles behind aggregate model rankings. Resource comparisons add distinct trade-offs: Jev is cheapest and fastest for the joint baseline, while hosted language models achieve higher accuracy. Together, these findings connect decision-level evaluation to whether additional calls improve correctness and what the resulting decisions cost.

Fixed roadside cameras repeatedly observe a stable scene overlaid by sparse moving traffic. Temporal memory can recover weak observations, but reusing moving evidence at stale locations can corrupt occupancy predictions. Motion compensation addresses displacement, while reliance on the resulting history remains a separate learning problem. We introduce RoadOcc, which learns soft routing among fixed-coordinate history (Persist), velocity-addressed history (Transport), and current evidence (Refresh). Motion state and class-consistent historical support supervise these source choices. Dynamic-aware cross-attention (DCA) updates candidate locations, multi-scale voxel velocity estimation (VVE) constructs transport addresses from multi-scale current–history correspondence, and velocity-guided dynamic sparse fusion (VDSF) combines routed evidence under fixed sparse-token budgets. On InfraOcc, RoadOcc reaches 65.29 mIoU and 32.37 dynamic mIoU, gains of 4.44 and 4.71 over STCOcc. Controlled address experiments show that VVE raises dynamic mIoU by 0.87 over fixed-coordinate reading. Across three seeds, supervised P/T/R adds 1.40 dynamic points over motion-corrected retrieval, while removing Refresh costs 0.32 points. Results from two transfer models, Occ3D-nuScenes, and longer intervals provide additional support. Code will be released.

Ground-based astronomical observations frequently contain streaks produced by artificial satellites, space debris, and potentially Near-Earth Objects (NEOs). While machine-learning models can reliably detect these features, their practical adoption in observatory operations is often limited by the lack of integrated tools for visual inspection, validation, workflow management, and structured data storage. This paper presents the StreakMind Workbench, a framework that applies MLOps practices to bridge research-oriented machine-learning pipelines with routine observatory operations. Rather than introducing new detection algorithms, the Workbench addresses a software-engineering challenge in astronomical computing: maintaining a single authoritative source of scientific processing code while providing astronomers with an operational environment for workflow execution and result inspection. Integrated with the reference StreakMind AI model of Carrillo et al. (2026) and implemented in Python using PyQt5, the Workbench supports the complete workflow from FITS ingestion to database storage, including inference, result inspection, database exploration, training management, and Minor Planet Center formatted observations. Validation on 273 images from La Sagra Observatory demonstrates successful end-to-end workflows while maintaining consistency with the underlying StreakMind scientific code. The platform facilitates operational use of a research ML pipeline in meter-class observatories and moderate-scale campaigns, supporting Space Situational Awareness and planetary defence.

Understanding articulated objects is fundamental for robotic interaction, requiring accurate rigid-part discovery and the recovery of their kinematic relations. Existing approaches often treat articulation as a by-product of reconstructed geometry or recover it through per-instance optimization. We instead build on the hypothesis that articulation is directly observable from persistent motion: points on the same rigid part move coherently, while relative motion between parts reveals their kinematic constraints. We present Track2Art, a motion-centric framework for recovering structured articulated objects from RGB-D interaction videos. Track2Art lifts tracked image points into persistent 3D trajectories and combines pretrained tracking features, visual descriptors, and explicit trajectory geometry. These representations are grouped into a variable number of rigid-part hypotheses and subsequently used to recover directed kinematic relations, joint types, and joint geometry through rotation-equivariant learned–analytic reasoning. On the aligned 20-object PartNet-Mobility suite, Track2Art achieves 0.695 Point IoU and 0.410 end-to-end J@20, while requiring neither ground-truth part counts nor test-time optimization.

In The Art of Computer Programming, Knuth benchmarked 10 encoding techniques for computing the chromatic number of the queen's graph: the minimum number of colors needed to color the squares of an $n \times n$ chessboard so that no two squares sharing a row, column, or diagonal receive the same color. In this paper, we extend his analysis much further by comparing thousands of encodings for the same problem, which allows us to identify additional factors that are important for solver performance. We obtain 1584 encodings for this problem by varying (a) the constraints that encode which color is assigned to each cell, (b) the constraints that forbid the same color appearing in a row, column, or diagonal line, and (c) the symmetry-breaking constraints. We find that the three most impactful encoding factors are (i) the choice of symmetry-breaking constraints, (ii) enabling so-called clique hints, and (iii) enforcing that each cell is assigned exactly one color through blocked clauses. Furthermore, while Knuth proposed clique hints as an advantage of the order encoding, we show in fact that they can be effectively employed for the one-hot encoding as well.

We prove that there exists $A\subseteq \mathbb{N}$ such that \[ R_A(N)=\frac{\pi}{4}N+ O\!\left(N^{1/4}\sqrt{\log N}\right), \] where $R_A(N)=\#\{(a,b)\in A^2:a+b\le N\}$. This improves the record $O(N^{1/4}\log N)$ obtained by Ruzsa in 1997.

We report a direct observation of the electronic structure of monolayer SnTe on graphene by angle-resolved photoemission spectroscopy. By combining low-energy electron diffraction, momentum microscopy, core-level spectroscopy, and first-principles calculations, we show that the observed monolayer-limit electronic structure is consistent with a puckered even-layer biatomic-layer building block rather than a nonpuckered square-flat structure. The observed band dispersion and its polarization dependence are reproduced by a puckered structure with a substrate-renormalized buckling amplitude. We further show that graphene actively interacts with ultrathin SnTe by selecting its in-plane orientational texture, renormalizing the buckling amplitude, and modifying the local interfacial electrostatic environment in the monolayer limit. In comparison, multilayer SnTe exhibits a more ringlike momentum-space distribution, interlayer-split valence bands, and a stronger p-type character, revealing the onset of thickness evolution from an interface-stabilized puckered ultrathin limit toward a more bulklike state.

4D radar complements dense image semantics with long-range geometry and radial motion, but existing radar–camera detectors largely solve where to align the modalities while leaving whether a piece of evidence supports an evolving object hypothesis implicit. An image token may describe an occluder, a nearby radar return may belong to another object, and a pose-aligned memory slot may carry incompatible motion. We formulate hypothesis-conditioned evidence grounding, which separates candidate access from evidence use: semantic, geometric, or temporal evidence is filtered or conditioned by the evolving 3D state before updating the corresponding query. \sgdetpp{} instantiates this principle through Anchor-Grounded Semantic Retrieval (AGR), which conditions deformable image retrieval on pooled anchor-consistent radar support; Geometry-Consistent Anchor Refinement (GCR), which attentively aggregates individual associated returns; and Doppler-Verified Correspondence (DVC), which replaces history only when current radial motion contradicts it. \sgdetpp{} improves the strongest compared method by 3.82 mAP and 6.82 ODS on OmniHD-Scenes and by 6.82 mAP and 9.22 NDS on ManTruckScenes, while also leading the listed methods in the TJ4DRadSet test comparison. Mechanism-targeted evaluations show that AGR improves strict AP in every projected-occlusion bin, the yaw-aligned box gate raises target-return purity from 29.95% to 58.87%, and DVC preserves 96.11% of motion-consistent history while retaining 75.90% contradiction recall. Code will be released.

We use quantum simulation to study how visible neutrino decay modifies the vacuum CP asymmetry in muon-to-electron oscillations within a $3+1$ neutrino scenario. The decay channel preserves CP symmetry, with CP violation arising from phases in the mixing matrix. We track the energy redistribution from visible decay and distinguish how parent attenuation and daughter regeneration modify the CP asymmetry. For the reference benchmark, the regenerated contribution is $4.85\times10^{-4}$, approximately $46\%$ of the total decay-induced shift. The full circuit for an effective two-energy model agrees with independent classical evolution. On an IBM quantum processor, we prepare the daughter state conditioned on decay and restore its physical normalization using fixed source and decay probabilities. After correction for measurement errors, the flavor-probability and coherence measurements both agree with the prediction within one standard error. Null controls test the suppression of the regenerated signal when daughter coherence or the relevant CP-sensitive mixing factor vanishes.

Small language models (SLMs) are increasingly paired with knowledge graphs (KGs), yet end-to-end KG question answering conflates graph access, search, navigation, reasoning, and answer generation. This coupling makes it difficult both to determine whether an SLM can faithfully execute the reasoning path implied by a question and to attribute failures to navigation rather than to other stages of the pipeline. We isolate this capability by employing the THESEUS navigation and traceability framework and using frozen, off-the-shelf SLMs as local action policies. At each hop, the environment exposes the legal outgoing graph actions, and the model selects one executable graph action and decides whether to stop, without task-specific parameter updates, model-controlled beam search, or free-form answer generation. This controlled setting allows us to evaluate terminal-answer accuracy with Hits@1 together with path fidelity, using Path Edit Distance (PED) as the primary trajectory metric. Across the Kinship and MQuAKE-ST KGQAs, similarly sized local models differ substantially in answer accuracy and path fidelity, with the two metrics sometimes favoring different models. This model-dependent behavior also extends to prompting, as a single demonstrated trajectory can improve or degrade navigation depending on the model. These results motivate evaluating SLM graph reasoning beyond endpoint accuracy alone.

The reproducibility of organ morphology and the extent to which computational models can predict morphogenesis remain difficult to quantify, particularly for organs with complex networks of fluid-filled lumina. Here, we combine Topological Data Analysis (TDA), biophysical simulation, and Bayesian inference to study lumen morphogenesis in pancreatic organoids. Lumen formation is governed by physical processes that are challenging to measure directly, including cell proliferation and luminal osmotic pressure. We simulate organoid development using a phase-field model and address the inverse problem of inferring these parameters from either time-lapse images or single morphological snapshots. Since lumen architectures vary substantially in size, structure, and connectivity, conventional geometric descriptors provide only a partial representation of their morphology. We therefore represent each organoid using SampEuler, a topological descriptor derived from the Euler Characteristic Transform (ECT). We first show that SampEuler captures morphological information encoded by established morphometrics. We then perform parameter inference using an approximate Bayesian computation (ABC) rejection framework with the SampEuler Wasserstein distance. Using synthetic organoids with known ground-truth parameters, our approach accurately recovers the osmotic pressure and the proliferation rate while revealing a compensatory trade-off between the two processes. Applied to experimental data from ten pancreatic organoids, the inferred posterior distributions are consistent with biological expectations. Together, these results establish a non-destructive, image-based pipeline for estimating otherwise inaccessible physical parameters governing lumen formation and highlight the potential of topological representations for linking complex biological morphology to mechanistic models.

Reinforcement learning (RL) achieves strong performance in sequential decision-making but remains brittle under dynamic uncertainty and distributional shifts. Robust Adversarial Reinforcement Learning (RARL) improves robustness via worst-case perturbations, but existing approaches frequently suffer from unstable optimization and degraded value estimation. In particular, overly aggressive adversaries can drive the agent toward uninformative failure states, while adversarial perturbations amplify disagreement between double critics and introduce biased value targets. We propose a unified framework, RACER (Risk-sensitive robust Adversarial critic ConsistEncy-regularized Reinforcement learning), that revisits adversarial RL from a risk-sensitive perspective. First, we introduce a state-dependent adversarial objective that adaptively regulates perturbation strength, suppressing harmful disturbances while preserving informative exploration. Second, we propose critic consistency regularization to reduce disagreement between Q-value estimators and stabilize learning. Comprehensive experiments on challenging continuous control benchmarks demonstrate that RACER consistently improves performance, robustness, and training stability over strong robust RL baselines.

Balloon-borne radio detectors offer a highly effective approach to monitoring immense volumes of the Earth's atmosphere and ice for ultra-high-energy cosmic rays and neutrinos. Because these payloads measure radio emission from a single, highly elevated vantage point, reconstructing the air-shower energy using traditional methods is not possible. In this work, we present a comprehensive, simulation-based energy reconstruction framework tailored to the unique cosmic-ray event geometries observed by balloon payloads. For the first time, we adapt and apply this methodology to direct atmosphere-skimming cosmic-ray showers, accounting for their highly asymmetric development in rarefied air. Furthermore, we provide a major update to the energy reconstruction of reflected downward-going showers by incorporating spherical reflection and ray de-focusing mechanics directly into simulations. We rigorously validate the method across different event geometries, demonstrating a baseline energy resolution of $8-11\%$. Finally, we quantify the systematic effects of finite detector pointing resolution, unknown primary mass composition, and signal-to-noise ratio, establishing a robust framework for the analysis of cosmic-ray events in next-generation balloon missions such as PUEO and POEMMA-Balloon with Radio.

We prove Ebenfelt's SOS conjecture, which implies the Huang–Ji–Yin gap conjecture.

Cooperation emergence is a central problem in multi-agent systems because decentralized agents must coordinate while adapting to the changing behavior of others. Evolutionary game theory identifies strategically stable outcomes, but stability under a population adjustment dynamic need not imply that finite-sample learning agents can reach the same outcome through local reward feedback. We study this distinction in a transparent three-agent governance-motivated game involving a government, a platform firm, and users. We derive replicator dynamics for the fixed stage-game incentives, evaluate the cooperative evolutionary basin on a symmetric initial-condition grid, and compare it with learning-basin estimates for three decentralized value-based learners. The learning analysis uses independent Q-learning with $\varepsilon$-greedy action selection, scaled Boltzmann exploration, and SA–EA BQL under the same payoff environment and outcome criterion. The evolutionary basin has volume $V_E=1.00$ on the sampled grid. The empirical learning basin is $0.88$ for $\varepsilon$-IQL and $0.00$ for both scaled Boltzmann and SA–EA BQL. Diagnostic traces show that broader action diversity and nonzero value separation can coexist with failure to sustain the cooperative joint action in this fixed configuration. These results indicate that evolutionary stability and learning accessibility are distinct properties of a coupled game–learning system. The shared-bike setting is a motivating application; the broader contribution is a framework for comparing population-level stability with the finite-sample accessibility of cooperation under specified multi-agent learning dynamics.

We study the spectral stability of bright and dark solitons in the discrete nonlinear Schrödinger (DNLS) equation using multiprecision arithmetic. The eigenvalues governing stability are exponentially small in the lattice spacing and cannot be resolved with standard double precision. To address this, we develop a computational framework combining multiprecision arithmetic, an exact Jacobian for the stationary problem, and a squared-operator formulation for spectral analysis. This enables accurate resolution of exponentially small eigenvalues and direct comparison with exponential-asymptotic predictions. Our results show that onsite bright solitons are spectrally stable, whereas intersite bright solitons and both onsite and intersite dark solitons are unstable. Bright solitons require only a few eigenvalues and allow efficient large-scale computations, while dark solitons demand higher precision due to their proximity to the continuous spectrum. Simulations up to \(N=65{,}250\) grid points (31.7 GB RAM) highlight the necessity of multiprecision arithmetic for capturing beyond-all-orders spectral effects.

Quantum Rabi model (QRM) is a fundamental model in light-matter interactions (LMIs), whereas this finite-component system manifests infinite physics to be explored and a full understanding has not yet been reached after a hundred years since it was proposed. In the present work we create a geometric perspective for the QRM by mapping to a curved nanowire system, with curvature-induced spin-orbit coupling (SOC) manipulatable by LMI strength and frequency ratio, which is capable of revealing and perceiving the property universality and diversity from the ground state to the whole spectrum in a unified picture. Indeed, we unveil a super university in the adiabatic regime which manifests a universal scaling of spin texture ubiquitously valid for different couplings, different frequencies and different energy levels. The curve frame also facilitates our finding that a direction changeover of tangential spin swing emerges in the non-adiabatic spin texture in association with the quantum phase transition (QPT) in the QRM, which endows a geometric non-adiabaticity interpretation and connotation to the QPT. We also clarify that, among the curvature-induced effects, the SOC accounts for the spectral bifurcation and oscillations. Our method and analysis, readily generalized to other light-matter models, opens a dialogue between geometry and physics, thus paving a novel way to gain deeper insight into LMIs.

Let $M^n\subset\mathbb{S}^{n+1}(1)$ be a closed minimal hypersurface with constant $S=|h|^2$ and constant $f_3=\operatorname{tr} h^3$, where $h$ is the shape operator. We prove that $S>n$ implies $S\geq2n$, and that the equality images are precisely the minimal Cartan isoparametric hypersurfaces with three principal curvatures, which occur only in dimensions $n=3, 6, 12, 24$.

Under suitable group actions, it is known that $A_\infty$-structures descend to orbit categories. We study the converse problem of lifting an $A_\infty$-structure to a prescribed covering. We show that a compatible group grading determines a canonical lift, uniquely characterized by the strictness of the covering projection. As an application, we construct $A_\infty$-categories associated with admissible fractional Brauer graph algebras by lifting Brauer graph $A_\infty$-categories. Their geometric data augment the surface models of Brauer graph algebras with a Nakayama character encoding the covering. We prove that admissible fractional Brauer graph algebras with equivalent geometric data are derived equivalent. Furthermore, we establish a set of combinatorial derived invariants and prove their completeness in reduced genus zero and in reduced genus at least two when the defining ribbon graph is non-bipartite.

Vehicle-to-Everything (V2X) communications impose very demanding requirements in terms of latency and reliability, which must be met in scenarios where multiple services with diverse performance targets coexist. In such scenarios, traffic-intensive services compete for limited radio resources, complicating the fulfillment of V2X service demands. Within this context, Network Slicing (NS) emerges as a key factor that enables the creation of multiple slices and the allocation of resources among them to satisfy heterogeneous service requirements. In particular, this work addresses the Radio Access Network (RAN) slicing problem from the perspective of Physical Resource Block (PRB) partitioning under high traffic demand conditions. To this end, a reinforcement learning approach based on Proximal Policy Optimization (PPO) is proposed to determine PRB allocations that satisfy the strict latency and reliability requirements of V2X services, while improving resource utilization efficiency and minimizing performance degradation of enhanced Mobile BroadBand (eMBB) services. The proposed solution is evaluated through simulation-based experiments under various traffic loads and different V2X service requirements, demonstrating its ability to adapt resource partitioning to network conditions and service demands.

Private decentralized learning is affected by sampling noise, privacy noise, and decentralized bias under heterogeneous data. We propose Private Recursive Decentralized Optimization (PRDO). PRDO uses recursive estimation with same-batch gradient differences to reduce estimation errors caused by sampling and privacy noise, while its Exact Diffusion component corrects decentralized bias arising from data heterogeneity. Our analysis establishes a nonconvex convergence bound without assuming uniformly bounded data heterogeneity across nodes. It further gives a sufficient condition under which recursive gradient differences yield strictly lower query sensitivity than private Exact Diffusion, together with an example that rigorously satisfies this condition. Experiments show improved accuracy over the evaluated baselines.

Solutions based on large language models (LLMs) often rely on temperature sampling to improve accuracy and stability by aggregating multiple samples from the completion distribution. However, this memoryless approach is inherently suboptimal: because it lacks awareness of prior generations and their evaluations, it produces an increasing proportion of semantically duplicate answers as more samples are drawn, leading to diminishing returns. To address this limitation, we introduce FLEET, a novel method that integrates a memory mechanism into the generation process. FLEET represents each generation as a sparse trajectory through states whose entropy exceeds a predefined threshold and uses these trajectories to infer per-token utility scores that adjust the logits. Benchmark evaluations demonstrate that FLEET achieves the same accuracy as the repeated sampling baseline, with a 3x speedup, and substantially improves accuracy on complex coding tasks (LiveCodeBench Pass@32 increases from 59.9% to 66.2%) under the same budget. Furthermore, in the greedy-decoding configuration evaluated here, the approach is deterministic and uses a single calibration pass to derive its principal hyperparameters, requiring only minimal modifications to existing LLM pipelines.