Papers

Decentralised federated learning removes the aggregation server but makes collaboration dependent on transient peer availability. In mobile and intermittently connected systems, evaluating a promising peer consumes contact time and may cause the exchange opportunity itself to vanish, so that the evidence a learner gathers about a peer is perishable: it decays because links expire and because peer models drift while old measurements age. This paper develops a self-contained theory of optimal stopping for the resulting peer-selection problem. We formalise a receiver's within-contact decision as a finite-horizon Markov optimal-stopping problem with costly information acquisition and a future-arrival outside option, and prove that it admits an optimal policy characterised by a reservation value (Snell-envelope structure). Around this formulation we prove: (i) stage-uniform, drift-aware concentration and a maximin certification rule that is correct with high probability together with a finite-sample identification bound; (ii) a mobility-aware value of-information stopping rule and comparative statics showing that higher link hazard lowers the value of continued probing and enlarges the stopping region; (iii) a closed-form value of waiting under marked-Poisson contact arrivals, together with a search-theoretic reservation value whose comparative statics we characterise; and (iv) a myopic-optimality theorem establishing that, in sufficiently volatile (monotone) mobility regimes, the one-step confidence-safe rule is a sound surrogate for the optimal policy and never stops prematurely. We instantiate the theory as PROSE (Perishable-evidence Reservation-value Optimal Stopping for Exchange), a lightweight, fully local policy, and delineate the static contact and drift-free limits in which classical sequential decision problems are recovered. The development is entirely analytical.

A physics-informed latent neural operator framework is proposed for three-dimensional compressor cascade flow prediction. The proposed method combines a Transolver-based latent encoder, which extracts compact global latent representations of flow conditions from point-cloud CFD data, with a coordinate-based PINNs decoder to reconstruct continuous flow fields while incorporating physics-informed residual constraints during training. The results show that the proposed framework can accurately reconstruct pressure and velocity distributions of complex three-dimensional cascade flows and maintains good prediction capability under previously unseen operating conditions. The present method demonstrates promising potential for efficient CFD surrogate modelling and aerodynamic prediction in turbomachinery applications.

Scheduling clients for model training is critical in federated learning due to both data and system heterogeneity. Most previous works focus on the quality of the scheduled clients to achieve faster convergence, shorter wall-clock convergence time, or better average model performance. They rarely consider the diversity of clients, which is important to counter heterogeneity and improve performance for the worst-off clients. In this work, we advocate the use of determinantal point processes (DPPs) to model and enhance the diversity in client scheduling. We first design the kernel matrices of DPPs using gradient information and quality scores, which inherently enables a flexible quality-diversity trade-off. Applying fast MAP inference over DPPs, we propose Adaptive Determinantal Client Scheduling (ADCS) in FL. We further quantify the gradient approximation error of ADCS and develop convergence analysis for general biased client selection in FL with non-convex loss functions. We conduct comparative numerical experiments showing that ADCS outperforms state-of-the-art client scheduling algorithms, including both quality-based and diversity-based ones.

The edge complexity of a weighted graph is the smallest ratio of the Fourier $\ell^1$ and $\ell^2$ norms of its adjacency matrix over all vertex labelings. We study the difficulty of finding this minimum by relating it to a graph symmetry. Adding a universal vertex with sufficiently large incident weight produces an explicit Fourier $\ell^1$ lower bound. We show that equality holds exactly when the source graph has a fixed-point-free involutory automorphism. Two-sided estimates compare the excess above this bound with the squared Frobenius distance to the nearest weighted graph having such a symmetry. For sources of constant weighted degree, these estimates determine the exact leading term as the added weight tends to infinity. A stronger separation for simple source graphs proves that additive $\frac{1}{256N^{\frac{7}{2}}}$ approximation of weighted edge complexity is NP-hard, even on connected graphs of odd order $N$ with at most two distinct positive integer weights, each at most $N^2$. We also prove that recognizing a simple graph with a real Fourier labeling is NP-complete. A seven-vertex example shows that every minimizing labeling can have nonreal Fourier coefficients even when real Fourier labelings exist. An exact rational certificate for this example is included in the appendix.

Long-horizon robotic search must resolve natural language against heterogeneous, incomplete, and often ambiguous evidence: textual information, prior maps, and observations arriving over time. The core challenge is to contextualize these streams and decide where to gather evidence before selecting a target. We present WORLDS: World-state Observation and Reasoning for Language-guided Discovery and Search, a framework that grounds reasoning in a persistent graph initialized from geospatial priors and updated by perception. Parallel Reasoners maintain competing candidate interpretations and request evidence to distinguish between them. We collect and process the requested observations with a multimodal Examiner, after which a Judge selects a grounded target or requests another pass. WORLDS achieves 51.8% navigation success across all 5,311 CityNav test episodes, the highest reported success rate, exceeding the previous published best by 15.7 percentage points under an OSM-only, high-resolution orthographic protocol. On 1,000 shared episodes, it achieves 50.0% versus 27.9% for the strongest adapted baseline using the same model, prior, sensing stack, and movement budget. Observation-based verification by the Examiner contributes 5.9 points of this success, and at a reduced reasoning-effort setting WORLDS still exceeds the adapted GeoNav baseline by 18.8 points while generating fewer tokens. We also demonstrate WORLDS on a quadrotor, which flies the generated sensing waypoints and grounds three language targets, including a vehicle absent from the map, from its onboard imagery.

In this work, we study a class of structured non-convex and non-smooth optimization problems arising in imaging applications, where the objective is the sum of three functions. We consider the Davis-Yin splitting method, a FISTA-type accelerated variant, along with a quasi-Newton line-search method and a plug-and-play (PnP) extension to solve the problem. We develop a unified convergence analysis based on the Davis-Yin envelope under mild assumptions, including the Kurdyka-Łojasiewicz property. Within this framework, we establish subsequential and global convergence of the iterates to stationary points, together with residual convergence rates. We demonstrate the performance of the proposed methods on image restoration and low-rank matrix completion problems. For low-rank matrix completion, we perform experiments on both synthetic data and public datasets to evaluate recovery accuracy and efficiency. In imaging tasks, including image deblurring, we compare the convergence behaviour and reconstruction quality of several algorithmic variants using peak signal-to-noise ratio (PSNR). The results show that FISTA-acceleration and line-search methods improve convergence, while PnP denoisers enhance image quality, and the proposed methods remain competitive on matrix completion tasks.

Given a rational matrix $A$, is every sufficiently large power $A^n$ entrywise nonnegative? We prove that this problem is decidable in deterministic polynomial time. This is in contrast with deciding eventual nonnegativity of a single prescribed entry sequence $(A^n)_{ij}$, which amounts to the Ultimate Positivity Problem for linear recurrence sequences, whose decidability is open. The previous decidability procedure (D'Costa, Ouaknine and Worrell, STACS 2024) splits the matrix powers into residue classes modulo a torsion exponent $D$ whose value can be exponential in the input size. We show that it is sufficient to check a single progression $A^{Dk+1}$: the exponents $n$ with $A^n \geq 0$ are closed under addition, and two consecutive exponents of the progression are coprime, so they generate every sufficiently large exponent. Along the progression, eigenvalues are grouped by their common $D$th power, and the required coefficients for each group are computed without forming $A^D$, any $\lambda^D$, or a splitting field: each summand is evaluated in its own root field $\mathbb{Q}(\lambda)$, and Galois equivariance keeps the degree and height of every class sum polynomially bounded, enabling certified zero and sign tests. The same algorithm with one case rejected decides eventual positivity, and a variant decides whether the matrix has any nonnegative power at all.

Greenhouse gas emissions estimates can be derived using inverse methods by combining atmospheric concentration observations with chemical transport models. The latter traditionally use physics-driven simulators such as Lagrangian Particle Dispersion Models (LPDMs), which are expensive to run and do not scale well to modern satellites' high resolution data. Previously we developed a performant atmospheric transport emulator that approximates LPDM outputs ("footprints") over South America ~1,000X faster than the UK Met Office's LPDM. Expanding towards global emulation is not straightforward, as atmospheric transport is regionally heterogeneous. This paper evaluates spatial transferability capabilities of models across four world regions: South America, East Asia, South Asia, North Africa using both region-specific and multi-region models, and leave-one-region-out experiments. Regional differences are characterised in the context of input variable and output footprint distributions. This work builds intuition in cross-region generalisation and transfer learning, aiding regional performance towards efficient global emissions estimates.

We study the complexity of finding $(\delta,\epsilon)$-Goldstein stationary points of nonsmooth nonconvex Lipschitz functions. By now, it is known that randomized first-order algorithms can solve this task with a dimension-free oracle complexity [Zhang et al., 2020], whereas deterministic algorithms cannot, as their complexity must scale at least linearly with the dimension $d$ [Jordan et al., 2023, Tian and So, 2024]. This leaves open whether deterministic algorithms can nevertheless solve the problem with oracle complexity polynomial in $d$. We answer this question negatively by proving a lower bound of order $(1/\epsilon)^{\Omega(d)}$ for deterministic algorithm, closing the exponential gap between the previously known lower and upper bounds and resolving an open problem posed by Jordan et al. [2023]. We further discuss several extensions and implications of this result to weaker stationarity notions, finding a descent direction and deterministic smoothing. Overall, our results establish an exponential computational advantage in nonsmooth nonconvex optimization offered by randomization.

A fundamentally challenging question in K-12 education is about the effects of taking more advanced or challenging classes. It is particularly complex because students (and/or their parents) choose whether to enroll in these classes, making causal analysis challenging. In this paper, we begin to tackle this question by taking advantage of a novel dataset from a public school system in the US. This dataset records students' course enrollment decisions, prior academic histories, demographics, and subsequent outcomes around the time of a district-wide change that introduced optional open-enrollment advanced middle-school courses in subject areas. This is a rich observational dataset, but enrollment in advanced classes is driven by student characteristics and choices rather than random assignment. This creates a core identification challenge: the same factors that influence enrollment in advanced courses are also predictive of academic outcomes. As a result, simple comparisons between enrolled and non-enrolled students are confounded, and naive estimates may reflect underlying differences in student ability, motivation, or support rather than the impact of coursework itself. Our analysis shows that enrolling in advanced English courses has a net positive but modest effect on student achievement outcomes. However, these benefits are unevenly distributed: some students with relatively large predicted gains ("middle achievers" in prior years) are less likely to enroll than others. Some other groups (e.g. Black students and those with lower socio-economic status) also demonstrate significantly lower propensity to enroll. This gap between predicted benefit and observed enrollment illustrates how careful data analysis can extract actionable insights from large observational datasets, including identifying students who appear well-positioned to benefit but do not select into advanced options.

The high-frequency spectral roll-off of continuous-time ARMA (CARMA) models can magnify the effect of narrow-band disturbances when aliasing is weak, making standard maximum-likelihood Whittle estimation sensitive to affected ordinates. We show that a logarithmic transformation $r_k = \log \rho_k$ converts the Whittle scale problem into a Gumbel location problem, connecting robust spectral estimation to the classical minimax theory of Huber and Rieder. A piecewise centering correction—closed-form for small $b$, implicit closed-form for the practitioner range—preserves Fisher consistency for any clipping level without numerical optimisation. Clipping the Gumbel score symmetrically and transforming back yields a two-sided clipped Gumbel score (the standard Rieder–Hampel bounded-influence form) whose normalised influence curve is proved locally asymptotically minimax under shrinking gross-error contamination. The efficiency loss is quantified by a single scalar $K(b)$: at $b = 1.5$, only $16\%$ nominal asymptotic variance overhead. In the stated AR(2) Monte Carlo design, the sample-size trends are compatible with the asymptotic rate, and at $b=1.5$ the bias reductions are about $40\%$, $90\%$, and $96\%$ for $a_1$, $a_2$, and $\lambda$, respectively.

We give a new elementary proof of Sullivan's no-wandering-domain theorem for rational maps of degree at least two using hyperbolic area. We then develop this method further to prove that a transcendental entire function has no wandering Fatou components if its set of singular values is compact and all accumulation points of that set lie in the Fatou set. Neither proof uses quasiconformal techniques.

We consider the problem of numerical differentiation of functions defined on the half-line: the derivative $f^{(r)}$, $ r=1,2,\ldots$, is recovered from a finite set of perturbed Fourier–Laguerre coefficients of a function $f$, whose error is measured in $\ell_p$, while the accuracy of approximation is measured in the uniform metric with the weight $t^{\alpha}e^{-t}$. It is established that the Wiener class $W^{\mu}_{s}$ admits such a setting for $\mu>r+1-1/s$. A class of methods $\mathcal{S}^{\theta}$ is constructed that realize the optimal order of accuracy of numerical differentiation on $W^{\mu}_{s}$ and use the smallest amount of input information in order.

Automated grading of conjunctival photographs could reduce the cost and variability of trachoma prevalence surveys, but the relative value of modern pretrained visual representations, lightweight feature adaptation, and training-objective design has not been established under a common protocol. This study presents a controlled evaluation for binary classification of Trachomatous Inflammation-Follicular (TF) versus Normal using 1,546 images from the public UCSF/Lietman collection. Images are processed using the OPTED pipeline for zero-shot tarsal-conjunctiva segmentation, alignment, cropping, and standardization. We first compare six pretrained backbones using a common classification pipeline and then evaluate four lightweight adaptation mechanisms on DINOv2 ViT-B/14. Under stratified five-fold cross-validation, DINOv2 with Efficient Channel Attention (ECA) and focal-plus-center loss achieved 91.66 +/- 0.97% accuracy, 90.69 +/- 1.10% macro-F1, and 96.06 +/- 0.71% AUC. ECA introduces only five learnable parameters while matching the performance of substantially larger alternatives. Objective ablation further showed that ECA did not consistently improve plain DINOv2 across loss functions; the lowest-variance 91.66% accuracy was obtained with cross-entropy plus center loss. Overall, the fine-tuned DINOv2 representation provided most of the predictive performance, while ECA offered a highly parameter-efficient refinement whose effect depended on the training objective. The resulting workflow provides a reproducible benchmark for active trachoma image classification.

An approximate action of the pair groupoid of a metric space determines an action of its Lipschitz-thin groupoid on complete extended metric fibers, uniquely characterized by its local sewing estimate. We prove this assertion under a superlinear three-point defect estimate and a Lipschitz bound for finite products. As a result, we obtain a positive proof of Curry and Manchon's sewing conjecture [2], with the necessary path-scaling correction to its estimate. Further restriction to isotropy constructs the based holonomy anticipated in their Remark 4.15 as a representation of based Lipschitz loops modulo thin equivalence. Rectangular comparison estimates give full relative Lipschitz-homotopy descent, in particular under their strong knitting hypothesis and under any three point estimate of total order greater than two. We prove the corresponding flatness and basepoint-covariance statements. An area model shows that the order threshold is sharp. We also separate the exponential estimate from the general knitting assumptions and exhibit a compact disk metric for which strong knitting does not imply invariance under merely continuous homotopy. As an application, for controlled fields on a metric space, we prove that the classical rough integral descends to the thin groupoid and admits substitution with a common rough controller.

Comparing audio-visual deepfake detectors requires coordinating dataset adaptation, temporal input representation, model interfaces and experimental conditions. We present DFD-Lab, a modular pipeline that separates these responsibilities while supporting shared training and evaluation workflows. We integrate three implementations: Xception-based maximum-logit fusion, ResNet with temporal LSTM fusion, and our AVFF reimplementation. Experiments cover external testing, degradation-based training augmentation and evaluation-time corruption. On a filtered subset of Deepfake-Eval-2024, models trained on FakeAVCeleb attain baseline AUROC values of 0.504, 0.538 and 0.458. JPEG50 training augmentation raises these to 0.691, 0.605 and 0.570, respectively, while all three accuracies decrease. These results illustrate why training interventions, evaluation corruptions and metric-dependent outcomes should remain distinct within a common pipeline. The contribution is the integration of audio-visual processing, interchangeable detectors and configurable experimental workflows, supported by empirical case studies. The findings highlight the challenge of cross-dataset detection and the complementary information provided by ranking and classification metrics.

We present a model of freeze-in dark matter (DM) and baryogenesis in which the new particles are a TeV-scale electroweak-doublet vector-like fermion (VLF) and two gauge-singlet Majorana fermions that constitute the DM, the heavier one with a mass in the tens-to-hundreds of keV range. Feeble Yukawa couplings of the VLF to the DM and the Standard Model Higgs doublet induce freeze-in production of the DM via VLF decays in the early universe. Due to a CP-violating phase in these couplings, DM propagation and rescattering induce opposite charge asymmetries in the VLF and SM sectors. Even in the absence of $B-L$ violation, the VLF asymmetry guarantees a baryon asymmetry once in-equilibrium SM interactions are taken into account. A relatively small conversion factor from VLF asymmetry to baryon asymmetry is balanced by a more direct asymmetry generation mechanism than in other minimal models of freeze-in baryogenesis. The observed DM energy density requires the heavier DM particle to remain well out of equilibrium throughout production. This allows us to map out the VLF mass-lifetime space using a novel perturbative calculation of the baryon asymmetry. We find that the model is already significantly constrained by LHC searches and structure formation probes, which will continue to test interesting parameter space in the coming years. The addition of VLF Yukawa couplings not involving the DM breaks $B-L$ and greatly expands the viable parameter space, while inducing potential signatures from dark matter decay and lepton flavor violation.

This article asks under what conditions artificial intelligence could warrant a rational attribution of consciousness. Linguistic ability, multimodality, memory, planning, action control, and humanoid embodiment are not sufficient evidence of phenomenal experience. Biological precursors such as excitability, homeostasis, neural networks, and hierarchical representation instead identify functions whose counterparts may be engineered. The paper compares conventional LLMs, hybrid h-LLMs, vision-language-action systems, and embodied agents with the Motivated Emotional Mind (MEM) architecture. MEM adds receptor-grounded representations, regulatory self-monitoring and interoception, affect, semblion-based associative memory, and re-entry into lower sensory and interoceptive maps. A related formulation defines a phenomenal state as a dynamic, integrated state of the whole embodied system in which sensor-grounded modal content is recurrently stabilized and coupled to a suprathreshold interoceptive representation of the system's regulatory state. The mathematical model of MEM published on arXiv specifies implementable relations among these components and makes the proposal more precise and testable. Comparative and causal tests in humans, invertebrates, and artificial systems could strengthen or weaken the MEM capabilities. MEM is therefore presented as a falsifiable research program requiring empirical validation and ethical caution.

Current privacy-preserving mechanisms (PPMs) in Complex Event Processing (CEP) systems are unnecessarily restrictive, reducing the utility of data received by data consumers. This article presents a novel approach to preserve privacy in CEP systems, improving the utility of detected event patterns by dynamically adapting the noise added to an unprotected data stream. We introduce a new guarantee named pattern-level differential privacy (DP), which enables us to apply and compare the strength of PPMs at the pattern level. We propose new pattern-level PPMs yielding pattern-level DP and analyze different trust settings of these PPMs and their requirements for context knowledge in the CEP system, e.g., the deployed queries. Our evaluation is based on three datasets (two real-world, one synthetic) and shows that the proposed PPMs increase data utility while preserving the same privacy level as the state-of-the-art PPMs. We use simulations to study the performance of our proposed PPMs in various practical scenarios. Furthermore, we demonstrate that computational complexity is not an obstacle to deployment.

Mitral and tricuspid regurgitation are the most common regurgitant valvular lesions, yet only a minority of severe cases undergo corrective surgery. Rapid assessment of valve mechanics could enable earlier, more precise intervention, but finite element (FE) analysis is slow to repeat across the many loading and tissue-property values of interest, which for a given valve are not known in advance. We introduce the Physics-Conditioned Neural Operator (PCNO), a transformer-based neural operator predicting leaflet displacement, strain, and stress fields conditioned on systolic blood pressure and tissue properties. PCNO is trained on, and evaluated against, FEBio simulations of functional and regurgitant mitral and tricuspid valves and of three mitral pathologies. With pressure and all material parameters simultaneously outside the training support, displacement error against FE reaches 4.48% and the mean errors of unsupervised geometric measures of valve function stay within 3.5%, indicating a conditioned solution operator over parameter space rather than an interpolator of the training set. Under this shift, PCNO is also more accurate than graph neural network and graph neural operator baselines trained on the same simulations, with the largest margins in stress.

We present a comprehensive benchmark of Federated Learning (FL) for multilingual Automatic Speech Recognition (ASR), evaluating four Speech-LLM architectures on the Multilingual LibriSpeech dataset. We compare FedAvg and FedProx across frozen and unfrozen encoder configurations, demonstrating that optimized learning rates are critical for performance. Specifically, independently tuning the learning rates for the speech encoder, connector, and decoder yields the lowest error rates, with full three-component adaptation (LoRA for encoder and decoder, full training for the connector) producing the best FL results. We observe that FedProx efficacy is architecture-dependent, providing notable advantages in multilingual pre-trained architectures (e.g., EuroLLM over TinyLlama when keeping the encoder fixed); this indicates that LLM backbone capacity plays a key role in mediating resilience to heterogeneous data distributions. These findings offer concrete design guidance for deploying multilingual Speech-LLMs in privacy-sensitive, distributed environments.

Radiation drag from inverse-Compton scattering of an external radiation field changes the acoustic dispersion relation of a relativistic fluid. We derive the linear dispersion relation and show that, depending on the electron distribution and fluid sound speed, radiation drag either damps both modes or damps one while driving the other unstable. The unstable growth is robust at short wavelengths but slow and overdamped at long wavelengths, where the linear analysis is limited by the background acceleration. The resulting growth and damping rates are of order the bulk radiative acceleration rate, so whenever radiation drag is important for the background flow, its effects on fluid perturbations are also important. A purely damped region exists for a wide range of parameters. Radiation drag also makes acoustic waves dispersive: The phase speed deviates from the ordinary sound speed at longer wavelengths, and approaches the sound speed at short wavelengths. We validate the derived rates using special-relativistic hydrodynamic simulations. We apply the derived relation to astrophysical jets such as gamma-ray bursts, blazars in the broad-line region of the host active galactic nucleus, and quasar jets traveling through the CMB and galactic radiation field. We find that the derived instability, damping and dispersion may be relevant across a range of relativistic outflow conditions.

The flux of protons slow-extracted from the CERN Super Proton Synchrotron (SPS) to the North Area experiments is limited by the induced radioactivity of the beam lost on the electrostatic septum during third-integer resonant extraction. Crystal shadowing, in which a thin bent silicon crystal deflects the protons that would otherwise hit the septum wires, has been in operation at the SPS since 2021; a single crystal at a non-local position halved the extraction losses relative to the local crystal alone, short of the fourfold reduction that future high-intensity operation requires. In this paper, the design, deployment and operation of a 3-crystal Multi-Volume Reflection Array (MVRA) at the non-local position in the fourth long straight section (LSS4) are presented. The array geometry was designed by Multi-Fidelity Bayesian Optimization (MFBO), combining a multilayer-perceptron surrogate with full particle tracking; it predicts a fourfold loss reduction with three crystals and up to tenfold with an ideally aligned array of four to five. At its optimum the as-built array reaches, and at its best point exceeds, the predicted factor of four, and beam dynamics simulations matched to the measured position and angle scans reproduce their shape to within 0.04 of the crystals-retracted loss. In physics production an extremum-seeking controller has held the array on its operating plateau, with a median loss reduction by a factor of 3.6 over 1.14 x 10^5 extraction cycles relative to the pre-shadowing baseline. The remaining margin to the target, the limitations of the present installation and the upgrade to a 4- or 5-crystal array are discussed.

We provide a complex-geometric approach to the Blaschke conjecture, i.e., that a manifold whose injectivity radius equals its diameter is isometric to a compact rank-one symmetric space (CROSS). In particular, we introduce the great quadric fibration, a complex analogue of the great sphere fibration. Requiring its total space to be a complex submanifold simultaneously explains many properties that a Blaschke manifold is expected to possess, including its Clifford structure, Hopf fibration, and diffeomorphism class. Furthermore, the great quadric bundle coincides with the variety of minimal rational tangents (VMRT) of the ambient Fano variety in the standard CROSS cases, and we explain this through a bend-and-break argument in the setting of the Blaschke conjecture under a suitable complexification assumption. Combining this with Hwang-Mok VMRT recognition, we prove the Blaschke conjecture for manifolds admitting an adapted complex structure on their entire tangent bundles.

We introduce and study multiradial regular variation of random fields, allowing componentwise thresholds to diverge at unrelated rates. The limit measures are homogeneous in each component and finite on events where every component exceeds a positive level in absolute value somewhere on a compact window. We characterise convergence by anchor exceedance masses and conditional whole-path laws, together with a compact-window mass bound in the continuous-parameter setting. For fields indexed by a countable discrete abelian group or by R^m every shift-invariant tail measure in this class admits a strictly stationary realisation. Finite moving averages with shared volatility show that scalar row-tail measures and simultaneous-exceedance tail masses on every consecutive finite window can coincide while relative-lag tail masses differ.

Blackwell and Cover both proposed the same method for guessing the larger of a pair of two numbers when one can only observe one and the other of the two numbers remains unobserved. As originally stated, the procedure ended with a computation showing that the probability of correctly guessing the larger number was greater than one-half. In this paper, we investigate what we can say when about the number in the pair that was not observed we employ the Blackwell/Cover procedure. Let {\alpha}1 and {\alpha}2 be two different numbers in (0,1]. Let G be the uniform distribution on [0, {\alpha}1] and H be the uniform distribution on [0, {\alpha}2]. We think of G and H as being represented by the heads probabilities of coins. We use a variant of the Blackwell-Cover strategy ro take pairs of coins chosen from distributions created from G and H such that one of the coins is flipped, and as a result of that flip the unobserved (unflipped) coin is placed in one of two sets A and B. We show that there are situations in which this can be done so that the mean of the heads probabilities of the unobserved coins in A is less than the mean of the heads probabilities of the unobserved coins in B. We show that there are situations in which recursive application of this process results in a set consisting almost completely of coins from one of the two distributions G and H, and another set consisting almost completely of coins from the other distribution.

The Generalized Naive Bayes (GNB) model was introduced for discrete and categorical random variables as an extension of classic Naive Bayes. We now accommodate the GNB framework to continuous explanatory variables. A central result of the paper is that structure learning of the GNB depends only on the pair copulas of the bi-variate marginals. We proved that the GNB structure can be assigned to the basis of a matroid, therefore we give greedy algorithms for finding the optimal GNB structure on the training data, in sense of minimizing Kullback-Leibler divergence. Three cases are considered: joint Gaussian distribution, then a more flexible model where we suppose the dependence structure to be described by a Gaussian copula with arbitrary marginals, and an even more flexible case where the joint continuous probability distribution is arbitrary, i.e. copula and marginal distributions are arbitrary. A method for model reduction, based on the newly introduced concept of GNB forest is given. We close the paper by comparing the newly introduced GNB classification results to other classical "glass-box" algorithms on real datasets.

Twin-beam (TWB) states generated by parametric down-conversion exhibit strong photon-number correlations that can be exploited for quantum communication. In the mesoscopic regime, these correlations can be directly investigated using photon-number-resolving detectors, such as silicon photomultipliers (SiPMs). Their nonclassical nature is quantified by the noise reduction factor (NRF), which provides a witness of quantum correlations and a means of monitoring channel security. In this work, a quantum communication protocol based on TWB states, originally proposed for binary encoding, is extended to a four-state scheme capable of transmitting two bits per use. Information is encoded by combining two mean photon-number values and two mode numbers of thermal noise superimposed on one arm of the twin beam, defining four distinguishable states in the (m,R) plane. While the mean detected photon number enables the retrieval of one bit, the NRF provides discrimination of the second bit and a security check against intercept-and-resend attacks, which introduce additional noise and degrade the quantum correlations. The protocol is investigated through two experimental measurement campaigns. First, the twin-beam source is characterized in terms of photon-number statistics, NRF, and detection efficiency to establish the operating conditions of the communication channel. Subsequently, the four-state protocol is implemented and evaluated through the analysis of state distributions and their 95% confidence regions, error probability as a function of sample size, and a comparative assessment of machine learning classifiers for state discrimination. Finally, the response of the NRF to an intercept-and-resend attack is investigated to assess the protocol's ability to detect eavesdropping.

We present VISTA, a two-stage framework for generating stylized 3D human motion by fusing structural content from text prompts with expressive style from reference videos, without requiring jointly paired (text, video, stylized motion) triplets. A Dual-channel Autoencoder first maps motion sequences and video clips into a shared latent manifold. A masked autoregressive diffusion backbone then operates within this manifold, injecting video-derived style through a dedicated late-fusion Dual-AdaLN pathway while preserving text-conditioned content structure. A cross-batch unpaired training protocol with latent cycle consistency enables joint learning across separate semantically rich and stylistically diverse datasets. As a proof-of-concept for controllable animation synthesis, we validate VISTA on rendered motion-capture references: it achieves the highest style recognition accuracy among video-conditioned methods while preserving competitive content alignment, and its decomposed 3-way classifier-free guidance provides independent, user-controllable calibration of the content-style balance at inference time.

The key-value (KV) cache has become the dominant consumer of memory in large language model (LLM) serving systems as context lengths, concurrency, and request lifetimes grow. High-bandwidth memory (HBM) provides the bandwidth attention decode needs but limited capacity, while off-package memory and storage add capacity but lack the bandwidth to sustain attention decode. High-Bandwidth Flash (HBF) is a promising substrate that combines terabyte-scale capacity with near-HBM read bandwidth. Limited write endurance makes read-only model weights its natural use, but we argue that HBF paired with HBM as a hierarchy can also hold the KV cache. Unlike prior hierarchies, whose secondary tiers are bandwidth bottlenecked, the comparable bandwidths let the two act as one logical memory for the long-context KV cache. HBF capacity enables long-context serving, and sparse attention makes it efficient by limiting KV-cache reads during memory-bound decode. Since HBF reads full flash pages and aggregates bandwidth by accessing thousands of parallel flash planes, sparse attention must be co-designed with these physical properties. We present SPLASH, an algorithm and architecture co-design that virtualizes the KV cache across HBM and HBF and adapts sparse attention to HBF's page granularity and plane-level parallelism. Across models and context lengths, SPLASH improves decode throughput per GPU by 3.5x-11.4x over the evaluated baselines under a 100 ms per-token latency objective, while keeping accuracy within 4% of dense attention across long-context suites.