Papers

We present Tulva, a practical threshold FHE scheme which equips a base FHE scheme with an asynchronous threshold decryption mechanism in a modular way. Our scheme simultaneously (1) is compatible with virtually any base FHE scheme and leaves its ciphertexts unchanged apart from an appended random tag, (2) supports a wide range of threshold profiles $(t,K)$ and ring degrees $\geq 256$ with 6-25 bits partial decryption per plaintext bit, and (3) achieves security in the presence of partial decryption queries on challenge ciphertexts based on the standard LWE assumption. We provide an open-source implementation, benchmark its threshold decryption performance and evaluate the end-to-end costs for typical privacy-preserving Ethereum Virtual Machine (EVM) use cases. For 16-out-of-512 threshold and a publicly preprocessed ciphertext, computing secret-dependent partial decryption and public share combination requires only 3.7 ms and 0.5 ms, respectively, on commodity hardware. Compared with the state-of-the-art synchronous scheme [Zyskind et al., CCS'25], we replace their ciphertext-independent yet interactive preparation of secret-correlated randomness with a public, ciphertext-dependent 0.5 s preprocessing step, eliminate interaction in the online phase while maintaining similar decryption time on the same machine, and support 32$\times$ more total users.

Let theta be an ordinary character or a Brauer character of a finite group N. We show that theta extends to every overgroup $G=G_\theta\trianglerighteq N$ if and only if Z(N) <= Ker(theta) and the deflated character of Inn(N) extends to Aut(N)_theta. This equivalence is compatible with fields of values. A similar investigation for extensions from subnormal subgroups leads to a new characteristic subgroup of N.

AI weather prediction (AIWP) models have emerged as a key complement to physics-based numerical weather prediction (NWP) models, and now often outperform them on aggregate skill. Yet, several studies report that physics-based models still forecast extreme events better, in particular record-breaking extremes, based on the evaluation of deterministic forecasts. Here we evaluate heat extreme forecasts from probabilistic AIWP models over global land and show that they clearly outperform the ECMWF IFS ensemble, both raw and post-processed, including for record-breaking events, in all regions and at all lead times considered. AIWP models better forecast the occurrence and intensity of extremes, are better calibrated in the tail and miss fewer records. Station observations confirm these results. We argue that many reported AIWP limitations stem from unsuitable verification choices, specifically comparing deterministic forecasts that target different quantities. Instead, a fair evaluation of extremes necessitates probabilistic verification.

Today's agentic systems are typically designed, built, and deployed by single developers with full control over all participating agents. While this simplifies development, it limits the potential scale of agentic systems. We study multi-developer agentic systems (MDAs) which enable developers to collaboratively build agentic systems to push frontier capability beyond what any individual developer could achieve alone. We construct empirical and theoretical models of developers to compare MDAs with single-developer agentic systems (SDAs). On GAIA, a general assistant benchmark, a population of specialized developers whose individual agents do not exceed 35% accuracy reaches 70% when their agents are combined in an MDA. This MDA performs 12 percentage points above the median general-purpose SDA built with several times each specialist's development budget. Our theoretical models show how MDAs can incentivize specialization and encourage more efficient use of aggregated developer resources. Put together, our results make a preliminary case for MDAs as a new scaling axis by enabling developers with diverse capabilities and resources to advance agentic AI collaboratively.

In this work, an aSi and cSi heterojunction solar cell with an intrinsic layer is simulated using AFORS-HET software. The study focuses on the role of the electronic density of states DOS of the amorphous emitter, p type aSiH, in governing the performance of the HIT cell and the electronic transfer across the heterointerface. It is found that the DOS of the p type aSiH emitter can cause an unfavorable recombination and trapping effect, which becomes more severe and difficult to avoid when the bulk dangling-bond density in the a-Si emitter approaches or exceeds the doping concentration. An increase in deep trap density beyond a critical value causes a drop in the built in voltage and shifts the depletion layer into p type aSi. This situation leads to a sharp decline in HIT cell performance, namely the open circuit voltage and the fill factor, and to the appearance of an S shaped IV characteristic. The simulation results suggest that the emitter activation energy needs to be optimized to achieve high efficiency. Under optimal conditions, holes accumulate near the heterojunction interface, leading to the formation of an inversion layer. This paper presents a physical mechanism for recombination processes within the emitter and concludes with the optimized parameters required to achieve a HIT cell efficiency comparable to Sanyo record of 23 percent on a thin silicon wafer.

On-policy self-distillation provides dense supervision for reasoning models via a privileged self-teacher. However, this supervision reflects local token preferences rather than the expected return of an unguided student continuation. Recent methods combine this signal with GRPO. They use GRPO for update directions and the teacher for signal scaling. However, we demonstrate a more effective integration within PPO. Specifically, we identify complementary positional dynamics: privileged information enhances a critic's value-prediction accuracy toward the response's end, whereas self-teacher guidance is strong initially, drops mid-generation, and rebounds finally. These insights motivate Value-Augmented Self-Distillation Policy Optimization (VASPO), wherein a self-teacher and critic evaluate identical privileged information at student-generated prefixes. The critic computes token-level PPO advantages from predicted continuation returns, while the self-teacher modulates selected positive advantages via bounded probability ratios. Across six mathematical reasoning benchmarks, VASPO improves mean accuracy over PPO by 2.59 and 1.81 percentage points on Qwen3-4B and OLMo3-7B variants, respectively, and achieves the highest mean accuracy among evaluated methods on Knights and Knaves. Our code is available at https://github.com/cii030/VASPO.

The Petz recovery map provides a near optimal decoder for approximate quantum error correction, but its inverse square-root normalization complicates circuit implementation. We use the polar structure of the Petz recovery isometry to implement recovery through a single quantum singular value transformation (QSVT). We give explicit block-encoding circuits and an extended recovery isometry that yields approximate Petz recovery without postselection. We bound the diamond distance between the approximate and ideal Petz encoding-recovery channels in terms of a spectrum-weighted loss, which provides a channel specific objective for optimizing the QSVT angles. Uniform polynomial approximation yields an asymptotic error bound that decays exponentially with the number of QSVT iterations, at a rate controlled by the smallest nonzero scaled singular value. Applications to Haar-random encoding and the Hayden-Preskill protocol illustrate the tradeoff between circuit depth and recovery accuracy and place existing scrambling decoders within a common Petz recovery framework.

Evaluations of classifiers (AI or human or agents) often assume a batch setting in which every agent produces labels on every example in a test set, which can be expensive or impractical in many contexts. This motivates online policies that adaptively select agents to query as examples arrive and update policies as outcomes are observed. We study the problem of selecting, in an online stream, the agent that best balances (i) predictive performance and (ii) subgroup fairness of a sensitive attribute, and that allows for delayed observations of true class labels. We formulate this as a Bayesian multi-armed bandit problem using Thompson Sampling, explicitly balancing accuracy and fairness. We conduct empirical evaluations on real-world datasets spanning problems involving tabular classifiers, large language models, and human labelers, demonstrating substantial reductions in regret relative to baselines. Our theoretical contributions include guarantees on the performance of the selected agent that incorporate both the accuracy and fairness aspects of the problem.

For a fixed matrix-polynomial trajectory in $\operatorname{SL}_3(\mathbb R)/\operatorname{SL}_3(\mathbb Z)$, we give a necessary and sufficient intrinsic Diophantine condition for its time averages to equidistribute at a power rate on its homogeneous orbit closure. The condition uses a lattice in the space of polynomial functions on the normalized orbit-closure group, each admitting a representative of total degree at most two in the matrix entries. These polynomial tests have a common degree bound of two, independent of the curve degree and optimal in general. The proof begins with a classification of the possible orbit-closure groups and derives lower bounds for polynomial coefficient vectors from the intrinsic condition. It combines quantitative nilmanifold equidistribution, unipotent tangent approximation, and effective flow estimates, with arithmetic separation and quantitative nondivergence controlling cusp excursions and periodic-orbit obstructions.

Graph-based analysis of electronic health records (EHRs) yields disease-trajectory catalogs, but structural support does not establish semantic novelty. We present a two-level framework to prioritize 99,527 branching motifs previously discovered by an upstream patient-disjoint, train-only MIMIC-IV v3.1 pipeline and frozen after validation-based catalog selection. The framework separates path-level and node-to-endpoint semantic sparsity from reliability-adjusted clinical priority. Three deterministic textualizations per motif or node-target pair were embedded with BAAI/bge-small-en-v1.5 and calibrated against 166,972 pre-2023 PubMed references using mean top-20 cosine distance. Positive recombination occurred in 8.63% of motifs and had negligible effect on global ranking, whereas specificity and support/GNN reliability terms changed clinical-priority ordering. Robustness analyses across neighbor counts, templates, encoders, generic-label taxonomies, score settings, PrimeKG, and corrected density-based RND showed that broad ordering was more stable than hard top-k membership. A semantic-dose analysis did not support a monotonic relation between score and forecasting benefit after endpoint and token-count adjustment (1.96 days per SD; 95% CI, -0.75 to 4.68). In the frozen catalog-q90 high-semantic subgroup, branch-enhanced forecasting reduced conditional event-time MAE by 20.1 days (95% CI, 13.3-27.0) and sMAPE by 2.37 percentage points (95% CI, 1.81-2.93), while discrimination and endpoint-ranking differences were inconclusive. Corrected RND showed low tail agreement with the primary score (Spearman rho = 0.430; q90 Jaccard = 0.125) despite a similar aggregate timing association. These findings support representation-aware, uncertainty-aware literature-relative prioritization of discovered trajectories rather than causal discovery or clinician-adjudicated novelty.

The muon anomalous magnetic moment, $a_\mu = (g-2)_\mu/2$, shows a sizeable discrepancy between Standard Model (SM) predictions and direct measurements, hinting at either New Physics or limitations in modeling low-energy strong interactions. As the uncertainty in the SM predictions currently exceeds the experimental uncertainty by a substantial margin, further improvements in theoretical precision are essential to clarify the origin of this discrepancy. The SM theoretical uncertainty is dominated by hadronic processes. The hadronic light-by-light (HLbL) scattering contribution relies on highly accurate data for hadron production via two-photon fusion. Operating at $\sqrt{s}=3.773\,\text{GeV}$ with over $20\,\text{fb}^{-1}$ of integrated luminosity, the BESIII experiment is ideally suited to probe these processes at small momentum transfers. By precisely measuring channels like $e^+e^-\to e^+ e^-\pi^0$ and $e^+e^-\to e^+ e^-\pi^0\pi^0$, BESIII provides the critical experimental input necessary to sharpen the SM prediction and clarify the true nature of the $a_\mu$ anomaly.

Edge deletion problems ask for the fewest edges whose deletion gives a graph with a specified property. We study two such problems on block graphs. First making the graph claw-free, and second making it a line graph. Since every line graph is claw-free, the second problem is at least as difficult as the first. We show that on block graphs the two optimum values are always equal, and both are attained by the line graph of a forest. The main step is a completion argument, in which starting from any claw-free solution, we put back edges to get some nice structure inside each block still maintaining the claw-free property. This reduces both edge-deletion problems to the same problem of selecting vertices within the blocks of the input graph. We solve it by dynamic programming in $O(n+m)$ time, which also produces an optimal deletion set and a forest whose line graph is what remains after the completion argument. We also extend the method to $K_{1,r+1}$-free edge deletion for every fixed $r$.

Generative super-resolution (SR) can recover sharp detail in computed tomography (CT) images from severely downsampled observations, but the missing high-frequency content is not uniquely determined by the low-resolution input. Consequently, visually plausible reconstructions may contain anatomically unsupported vascular detail, such as missing, spurious, or incorrectly connected vessels. To overcome this challenge, we introduce VascuSR, an iterative anatomy-aware diffusion framework that combines image reconstruction with explicit 3D vascular guidance. First, to provide richer anatomical guidance, we use an explicit 3D vascular representation that captures structural information beyond individual slices. This representation also simplifies vascular refinement by allowing a lightweight 3D generative refiner to focus on vascular structure rather than complex CT appearance. Second, to guide slice-wise SR, a lightweight anatomy controller injects the refined structure into the diffusion decoder, while iterative feedback updates this guidance from each new SR volume. On re-annotated MSD Task08 and HVM datasets, VascuSR outperforms interpolation and learning-based SR baselines in segmentation accuracy (Dice and clDice) and 3D vascular structure (connectivity and mean centerline distance), while maintaining image quality comparable to its SR backbone. Compared with learning-based baselines, VascuSR reduces class-averaged connected-component counts by over 20% on both datasets and mean centerline distance by over 10% on MSD.

We study disjoint solution pairs $(u,D)$ to the minimal surface equation in $\mathbb{R}^n$, with $u=0$ on $\partial D.$ A conjecture of W. Meeks states that at most two such pairs can exist in $\mathbb{R}^n.$ We confirm the conjecture in two complementary cases. For $n\in\{2,3,4,5,6\},$ we prove it under a superlinearity assumption. In dimension $n=2,$ we establish the result under a suitable sublinearity assumption. We also derive estimates on the asymptotic angle of domains supporting nontrivial solutions for $n\geq 3.$

No: neither through the bosenovae of gravitational atoms formed around stars and compact objects, nor through those of boson stars formed in the galactic dark-matter halo. We assume that all the dark matter consists of ultralight bosons with attractive self-interactions, which can be captured in the gravitational potential of astrophysical objects or condense into self-gravitating configurations. When these systems collapse in a bosenova, part of their mass is turned into relativistic radiation that escapes the galaxy. We bound the mass that a single object can eject and sum it over realistic models of the Milky Way and of other types of galaxies. For gravitational atoms, with realistic stellar populations and stellar evolution, the stars of the Milky Way deplete at most $\epsilon^{\rm GA}_\star\simeq3\times10^{-9}$ of its halo, and the pump is equally negligible in other galaxies. The central black hole, Sgr A$^*$, could reach $\epsilon^{\rm GA}\sim10^{-3}$, but only for a boson mass disfavoured by ultrafaint dwarf galaxies and beyond the validity of the perturbative treatment, where only a geometric bound on the dark matter flux is available. For boson stars feeding on the rest of the halo, these would radiate at most $\sim3\times10^{-5}$ of the Milky-Way halo and $0.1\%$ of a DF2-like one.

We model selected 6.7 GHz methanol flare profiles in G358.93-0.03 using the one-dimensional Maxwell-Bloch framework. The 2019 non-periodic flare, associated with an accretion burst, produced complex single-dish light curves characterized by sharp rises, multiple peaks, secondary structure, and extended decays. For six representative velocity channels, we model each light curve as a sum of scaled intensity responses from effective emitting systems driven by phenomenological pump pulses of fixed width but with component-dependent background rates, amplitudes, and arrival times. The inferred initial inverted column densities correspond to characteristic superradiance timescales $T_{R,0} \simeq 0.3$-$1.4~{\rm d}$, which are shorter than the adopted dephasing and relaxation timescales, $T_2 = 6.9~{\rm d}$ and $T_1 = 144.4~{\rm d}$. The fitted components therefore lie in the fast-transient superradiance regime. Existent VLBI observations motivate the possibility that a single-dish velocity channel blends emission from multiple unresolved flaring regions, although these observations are not used as direct fitting constraints. Although the decomposition is not unique, it provides a physically motivated description of the complex flare morphology within a unified Maxwell-Bloch framework.

How galaxies stop forming stars is a complex and central problem in astrophysics. Numerous physical mechanisms have been proposed, yet a simple, unified framework connecting these mechanisms remains lacking. Here, we derive a step-function quenching response from the fundamental principle of gas mass conservation, governed by the competition between gas-supply and gas-depletion timescales. This response also enables the gas-supply timescale to be inferred from the observed quenched fraction. Applying this method to approximately 350,000 galaxies, we find a characteristic gas-supply timescale of ~1 Gyr, reveal distinct inside-out and outside-in quenching signatures associated with supermassive black holes and dark matter halos, respectively, and identify galaxies with intermediate stellar surface mass densities as occupying a critical state for quenching and rejuvenation. Our work establishes a unified physical framework in which a simple step-function response captures how diverse physical processes regulate galaxy quenching and the resumption of star formation through their effects on the competition between gas-supply and gas-depletion timescales.

Long-horizon precise manipulation produces mixed-quality data for post-training vision-language-action (VLA) models: even successful demonstrations and intervention-assisted rollouts may contain detours, hesitations, failed attempts, and imperfect recoveries. The key challenge is to identify which action chunks actually advance the task, rather than uniformly cloning or discarding imperfect trajectories. We propose Graph-Reweighted Behavior Cloning (GR-BC), an offline iterative post-training framework that builds a task-conditioned goal-reachability graph over demonstrations and iteratively collected rollouts, including both successful and failed episodes with human interventions. Reverse graph search computes minimal remaining steps to task success within the dataset support, and each action chunk is weighted by how much its execution reduces this potential. This consequence-based weighting reuses recoverable segments from failed trajectories, leverages interventions as recovery paths, and down-weights locally regressive chunks even inside successful trajectories. New rollouts are incorporated by appending them to the graph and rerunning search, without training a reward model, action-conditioned critic, or progress predictor. Across four precise manipulation tasks and three VLA backbones, GR-BC improves single-round supervised fine-tuning and reaches 91% success after a few offline iterations, without trajectory comparisons or progress labels. More details and videos are available at: https://gr-bc.github.io

Speculative decoding accelerates autoregressive generation by drafting multiple tokens per step, yet feature-based drafters rely on target hidden features whose contextual emphasis may not align with the needs of subsequent draft positions. Hybrid linear attention (HLA) models maintain compact recurrent states that accumulate historical associations, offering an efficient source of complementary context for speculative drafting. We introduce ReaDy, the first speculative decoding method, to the best of our knowledge, to leverage HLA recurrent states for position-specific context retrieval during drafting. ReaDy equips each draft position with a learned query to retrieve supplementary context directly from the target's committed recurrent states. A lightweight state reader integrates this context into draft representations through residual updates, while low-rank state reads and fused operators keep additional computation and I/O overhead low. Building on DSpark's parallel backbone and sequential prediction head, ReaDy employs a lightweight two-stage adaptation strategy, with reader capacity, state sources, and residual placement examined through ablation studies. Across seven benchmarks spanning mathematics, code generation, and dialogue, ReaDy achieves an average $3.16\times$ decoding speedup over autoregressive decoding with Qwen3.8-27B on a single NVIDIA B200 GPU. It outperforms DSpark on every benchmark, improving throughput by 7.97% and acceptance length by 8.50% on average. Our code is available at https://github.com/StartLuxLabs/ReaDy.

Longitudinal electronic health records (EHRs) can reveal disease-progression structure, while useful forecasting requires leakage-controlled discovery, censor-aware outcomes, and inspectable trajectory evidence. We developed a graph-enhanced framework in MIMIC-IV v3.1 that splits patients before graph construction, learns train-only consecutive-admission transitions, and freezes endpoint-specific branching-motif policies selected on validation data. The final catalog contained 99,527 motifs, and a time-aware Transformer retrieved at most 24 patient-compatible motifs per endpoint through endpoint-specific cross-attention. Across five training seeds, micro-AUPRC increased from 0.9330 to 0.9357, while event-time MAE decreased from 676.9 to 657.7 days (reduction 19.2 days; 95% paired patient-cluster bootstrap CI, 13.8-24.8). Timing effects were heterogeneous: MAE worsened for events within 730 days but improved by 58.9 days for events occurring after 730 days. In an exploratory internal temporal split, branch conditioning reduced conditional event-time MAE by 26.9 days (95% CI, 23.0-30.8) without evidence of better micro-AUPRC. Corrected survival evaluation over the supported 0-2,555-day interval did not establish macro C-index or integrated Brier score superiority over a matched history-only discrete-hazard model. A directed, degree-matched GNN repair substantially improved unseen-edge discovery, yet replacing GNN-derived scores within fixed motif identities did not improve downstream forecasting. These results position branching motifs as structured trajectory context whose main value is long-range timing support rather than uniformly stronger endpoint discrimination, survival prediction, or causal explanation.

Low-light video processing remains an underexplored problem in computer vision, primarily due to the scarcity of high-quality ground-truth data. Existing public datasets exhibit several limitations, including restricted motion diversity, limited scene coverage, weak degradations, reliance on auxiliary modalities such as event data, and misalignment between low-light inputs and ground truth. Furthermore, current datasets typically address either low-light enhancement or scene understanding, whereas these tasks should be considered jointly. We introduce OLiVES, a benchmark dataset for outdoor low-light videos with dynamic scenes, designed to support both low-light video enhancement (LLVE) and video object segmentation (VOS). The dataset addresses key content gaps by incorporating a diverse range of natural textures, animals, occlusions, dynamic activities, and close-up scenes. To the best of our knowledge, OLiVES is the first outdoor low-light video dataset captured using professional cinema-grade cameras. It provides over 25,000 aligned normal- and low-light frame pairs, along with over 46,000 VOS annotations. We benchmark state-of-the-art LLVE and VOS methods on OLiVES and show that training on it improves performance on both tasks. The OLiVES project page is available at: https://anonymous-OLiVES.github.io/.

We extend Hintz's separation of variables for radiative gravitational and electromagnetic perturbations from Kerr-Newman to Kerr-Newman-de Sitter black holes. On subextremal backgrounds with nonzero rotation and charge, an explicit algebraic matrix transformation yields coupled two-component radial and angular ordinary differential systems linked by a common separation constant. The transformation is independent of the frequency and azimuthal number and applies to both signs of spin weight. For real frequencies, we establish a complete angular mode decomposition. We also translate regularity at the future black hole outer and cosmological horizons into ingoing and outgoing boundary conditions, respectively, for the separated radial systems. These results show that exact separation of the coupled radiative equations extends beyond asymptotically flat Kerr-Newman spacetime.

We prove that every mod $k$ integral cycle $T$ of sufficiently high codimension in an oriented Riemannian manifold $\mathcal{M}$ can be approximated in flat norm by a cycle which is a $\mathbb{Z}_k$-smooth submanifold $\Sigma$ of nearly the same area, up to a singular set of codimension 5. If in addition $T$ belongs to a mod $k$ homology class admitting an embedded $\mathbb{Z}_k$-smooth manifold representative, then $\Sigma$ can be taken fully regular, up to the codimension one stratum of classical singularities. This extends the smooth approximation theory to cycles with coefficients in $\mathbb{Z}/k\mathbb{Z}$ in stable regimes. Moreover, we provide examples showing the impossibility of locally approximating minimizing cycles by smooth or $\mathbb{Z}_k$-smooth submanifolds, even in homology classes with smooth or $\mathbb{Z}_k$-smooth representatives.

Widely used methods for identifying crystal structure in atomistic simulations are limited in practice to a handful of simple structures, leaving many important crystalline systems without automated classification techniques. Voronoi topology analysis offers a route to classifying arbitrary crystalline and defect environments, but realizing this potential first requires enumerating the Voronoi cell topologies that can arise under thermal perturbation of a structure of interest. We present an algorithm that deterministically enumerates all Voronoi cell topologies arising from generic infinitesimal perturbations of an arbitrary crystal structure, producing structural classifiers that extend automated identification to general crystalline systems. We demonstrate the approach on several structures of general interest.

Reinforcement learning with verifiable rewards (RLVR) trains language models from checked task outcomes, but assigning a terminal reward to individual reasoning tokens or actions remains difficult. Group-relative methods such as Group Relative Policy Optimization (GRPO) sample several attempts per prompt, but lose their reward-based learning signal when all attempts receive the same outcome. We propose Hindsight-Privileged RLVR (HP-RLVR), a single-rollout method that turns a completed trajectory's own future into token-level update weights. The generating policy snapshot scores each eligible token under its causal history and under a hindsight context containing the realized future and verifier feedback. A normalized Jensen–Shannon divergence between the two views weights the reward relative to a historical baseline, yielding bounded token credit for both successful and failed attempts, with no reference solution or learned critic. In paired Qwen3-8B runs over a shared $100$-step schedule, HP-RLVR trains $1.21\times$, $1.46\times$, and $2.30\times$ faster than GRPO on mathematics, code generation, and ALFWorld, using one eighth the rollouts per prompt. Accuracy is comparable overall, with HP-RLVR ahead on mathematics at all three sizes and GRPO ahead on code and ALFWorld. In ALFWorld, $43.2\%$ (4B) and $49.0\%$ (8B) of GRPO's gradient-norm records are zero, versus none for HP-RLVR.

We establish a smoothness criterion for Fano varieties with quotient singularities. As an application, we show that if $X\to B$ is a Lagrangian fibration from a compact hyper-Kähler manifold and $B$ has quotient singularities, then $B$ is a projective space.

A subset of active galactic nuclei (AGN) manifests as compact radio sources with peaked radio spectra, commonly referred to as peaked-spectrum (PS) sources. Generally, PS sources are interpreted as young radio jets, but the early evolution of the jet is rarely observed and not fully understood. Here, we present VT J182357-234511 (VT J1823-2345), a radio transient in the Very Large Array Sky Survey (VLASS) that resembles a PS source. Notable properties of VT J1823-2345 include: (i) brightening by a factor of 3 between 1996 and 2021 at 1.4 GHz, (ii) peaked spectra with high frequency peaks at $\nu_{\mathrm{p,obs}}\gtrsim6\,\mathrm{GHz}$, (iii) a peaked 3 GHz light curve that has been fading since 2023, (iv) a resolved Gaussian (major-axis) size of 0.4 mas in 2024, and (v) a linear polarization fraction $p_{\mathrm{L}}\simeq0.73\%$ at 8.8 GHz with an extreme Faraday rotation measure $\mathrm{RM_{\mathrm{obs}}} \sim +12,000\,\mathrm{rad}\,\mathrm{m}^{-2}$. On the sky, VT J1823-2345 coincides with a supernova remnant and an F-type star, but no background galaxy is seen. Based on the multi-wavelength properties, we disfavor Galactic transients, explosive extragalactic transients, and blazar variability as plausible explanations. We argue that VT J1823-2345 is consistent with a "short-lived" parsec-scale PS source with a decades-old jet that does not grow into a large-scale radio galaxy. From comparing the full VLASS light curves, we speculate that VT J1823-2345 is similar to many VLASS nuclear transients reported in the literature, which may represent an emerging class of transient radio jets, perhaps related to episodic AGN activity or jetted tidal disruption events.

Pretrained representations are increasingly used to share and reuse information from medical images without exposing the original scans, but it is unclear how much patient-specific information these representations retain. We investigate this question for brain magnetic resonance imaging (MRI) by evaluating five frozen pretrained encoders spanning different architectures, pretraining objectives, and training domains on two longitudinal T1-weighted datasets, OASIS-2 and MIRIAD. Under an attacker model with access to released embeddings, the corresponding public encoder, and auxiliary MRI scans, we evaluate patient linkage and sensitive-attribute inference. Across the evaluated representations, we observe substantial subject-specific information: rank-1 patient linkage reaches up to 99% on MIRIAD and exceeds 90% on OASIS-2 for the highest-risk representation, despite the absence of an identification objective during encoder training. We further show that the released embeddings retain information about a demographic attribute, with recorded-sex prediction reaching balanced accuracy of 90.9% on OASIS-2 and 95.5% on MIRIAD. Overall, our results demonstrate that replacing raw brain MRI with learned embeddings does not, by itself, eliminate patient-linkage or attribute-inference risk, and is therefore insufficient on its own to establish anonymity.

We study large-k symmetric and antisymmetric Wilson loops in N=4 super-Yang-Mills theory. We represent their generating functions as Fredholm determinants built from the Hermite kernel weighted by a symbol function that encodes the representation. Using the integrable structure of the kernel, we systematically apply the Tracy-Widom approach to derive a closed system of integro-differential equations. The equations are then solved order by order in 1/N by a WKB ansatz, while keeping the symbol function general. We find a closed-form expression for the leading and subleading order contributions to the generating functions. Finally, specializing to the two representations, we perform the large-k saddle-point analysis and find the subleading contribution to the Wilson loop. Our results confirm the previously observed discrepancies with the dual D-brane description of giant Wilson loops.

We study the number of exact matrix-vector queries needed to sample a centered $d$-dimensional Gaussian with condition number at most $\kappa$ to total variation error $\varepsilon$. The best known algorithm uses $O(\min\{d,\sqrt{\kappa}\log(\frac{d}{\varepsilon})\})$ queries, while the known lower bounds do not grow as $\varepsilon\to0$ and match this bound only for $\kappa\le d^{1/5-\delta}$ at constant accuracy. We introduce a construction based on a hidden determinant sign and prove a lower bound of $\Omega\!\left(\min\left\{d,\sqrt{\kappa}\log\left(1+\frac{d}{\sqrt{\kappa}\,\varepsilon}\right)\right\}\right)$ uniformly for all $d\ge1$, $\kappa\ge4$, and $0<\varepsilon\le\tfrac{1}{16}$. This bound matches the known upper bound up to constant factors whenever $\kappa\le d^{2-\delta}$ for fixed $\delta\in(0,2)$, or $\varepsilon\le d^{-c}$ for fixed $c>0$. We also extend a recent fixed-spectrum construction with a hidden eigenbasis from $\kappa=2$ to $2\le\kappa\le d^{1-\delta}$, for any fixed $\delta\in(0,1)$. This gives an $\Omega_\delta(\sqrt{\kappa}\log d)$ lower bound at constant accuracy that holds even when the spectrum of the precision matrix is known to the algorithm.