Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 6 November

Analysis of PDEs

The Whitney Extension Problem in Carnot groups

Let $\mathbb{G}$ be a Carnot group, let $E \subset \mathbb{G}$ be compact, and suppose we are given a function $f: E \to \mathbb{R}$. The $C^1_H$ Whitney Extension Problem asks the following: what must be true about $f$ to guarantee that there exists a smooth, horizontal function $F \in C^1_H(\mathbb{G})$ extending $f$, i.e., there is some $F:\mathbb{G} \to \mathbb{R}$ such that $F|_E = f|_E$ and $XF$ is continuous for every horizontal left invariant vector field $X$? We adapt the method of G. Glaeser and C. Fefferman to answer this question in this non-commutative setting. Moreover, we provide an alternative solution to the $C^1$ Whitney Extension Problem for horizontal curves in the Heisenberg group.
Read the full abstract

Data Structures and Algorithms

Randomized Greedy MIS Is Fast at Every Node of a Tree

The Randomized Greedy MIS (RGMIS) algorithm computes a maximal independent set (MIS) by processing the vertices in a uniformly random order and greedily adding each vertex to the independent set whenever possible. Despite its simplicity, this algorithm has proved effective in a range of parallel, distributed, and dynamic settings. For sufficiently small maximum degree, Nguyen and Onak (FOCS~'08) and Yoshida, Yamamoto, and Ito (STOC~'09) showed that a recursive simulation of RGMIS can be used to estimate the sizes of matching and vertex cover in sublinear time. The key subroutine in their approach is the following local query procedure: given a vertex $v$, to determine whether $v$ belongs to the MIS, recursively evaluate the neighbors of $v$ that precede $v$ in the random order. If any such neighbor is found to belong to the MIS, the procedure immediately concludes that $v$ is not in the MIS; otherwise, it concludes that $v$ is in the MIS. For graphs of maximum degree $\Delta$, Nguyen and Onak showed that, in expectation over the random ordering, this procedure makes $O(\exp(\Delta))$ recursive calls from a worst-case starting vertex. No better upper bound is known even for trees. We prove that on trees, from any starting vertex, this recursive RGMIS procedure makes $\mathrm{poly}(\Delta)$ recursive calls in expectation.
Read the full abstract

All authors Mina Dalirrooyfard, Konstantin Makarychev, Slobodan Mitrović, Srikkanth Ramachandran

Systems and Control

Control of Linear Dynamical Systems with Operational Constraints

This paper presents a method for designing servo-controllers that enforce operational constraints on multi-input multi-output linear time-invariant dynamical systems. The proposed method introduces a control augmentation that complements a baseline controller to ensure forward invariance of state/output trajectories within prescribed box constraints. The derivations of the control design are based on the Nagumo Theorem, the Comparison Lemma, and principles from min-norm optimal control akin to the theory of Control Barrier Functions. Importantly, an explicit piecewise-linear feedback control law is derived, which provides continuous enforcement of constraints without online optimization, enabling rigorous stability and robustness analysis via classical linear systems theory. Extensions to handle estimation errors in output feedback and integral action in proportional-integral control are provided. Theoretical results are validated through flight control simulation studies that demonstrate input/output constraint satisfaction, robustness to disturbances, and closed-loop stability. The proposed method offers a mathematically rigorous, practical, and analyzable alternative to common heuristic or optimization-intensive constrained control strategies.
Read the full abstract

Chemical Physics

Density Matrix Renormalization Group Method for Core-excited Electronic States

We present the formulation and implementation of the Core-Valence Separation (CVS) approximation within the Density Matrix Renormalization Group (DMRG) framework for the simulation of core-level X-ray spectra. Two symmetry implementations are developed and assessed, identifying a scheme that preserves spin symmetry and remains applicable to multiple core orbitals. Benchmark calculations for BeH$_2$ and [Fe(H$_2$O)$_6$]$^{2+}$ demonstrate excellent agreement with conventional multi-configurational methods for excitation energies and spectral intensities, while enabling substantially larger active spaces. Using quantum information theory measures, we show that the CVS constraint suppresses core-occupation fluctuations without significantly altering core–valence entanglement, providing a diagnostic for its applicability. Furthermore, we introduce a reference-free second-order perturbative correction that systematically reduces errors introduced by the CVS approximation. Finally, a generalized multi-core formulation enables the simultaneous treatment of excitations from different core shells. These developments establish CVS-DMRG as an accurate and computationally efficient approach to core-level spectroscopy of strongly correlated systems.
Read the full abstract

All authors Giacomo Bassi, Mihael Eraković, Markus Reiher, Sergey I. Bokarev

Machine Learning

DisMUSE: An Efficient Distributed Singular-Spectrum-Based Optimization Approach with Theoretical Convergence Guarantee

In recent years, singular-spectrum-based optimizers have attracted increasing attention due to their ability to preserve the matrix structure of neural network gradients, thereby enabling geometry-aware optimization. However, most singular-spectrum-based (SSB) approaches remain centralized, limiting their use in distributed learning infrastructures that exploit massive parallelism. Distributed SSB optimization is still at an early stage, with open questions about convergence guarantees and communication efficiency. To fill this gap, we propose the Distributed Momentum-based Spectral Estimator (DisMUSE), an efficient distributed SSB optimization framework for both data-center and federated settings. Using Muon-style exact orthogonalized momentum as the spectral update, DisMUSE integrates a recursive variance-reduced momentum estimator and achieves an $O((KT)^{-1/3})$ convergence rate, where $K$ and $T$ denote the number of nodes and local iterations per node, respectively. This rate matches the best-known rates for variance-reduced nonconvex distributed optimization. Building on this variance-reduced framework, we study two extensions. First, our DisMUSE method with finite Newton-Schulz (NS) iterations preserves the same convergence order under a uniform normalized support-spectrum condition, up to a multiplicative factor. Second, we develop a communication-efficient variant based on low-rank (LR) momentum transmission and quantify the error it introduces to the stationarity bound, characterizing the trade-off between accuracy and communication cost.
Read the full abstract

All authors Zhen Qin, Zhiyao Zhang, Yingbin Liang, Jia Liu

Analysis of PDEs

Non-Uniqueness of Weak Solutions to the Unforced Two-Dimensional Euler Equations with Vorticity in $L^1\cap L^p$, Part II: The Nonlinear Construction

This paper and its companion Part I prove that the incompressible Euler equations in the plane, without external force, admit two distinct renormalized weak solutions with the same initial vorticity $f$, where $f$ is radial and nonnegative and lies in $(L^1\cap L^q)(\mathbb{R}^2)$ for every $q<\infty$ but is not bounded. One solution is radially symmetric and stationary. The other is $3$-fold rotationally symmetric, is equimeasurable with $f$ at every time, and departs from $f$ in finite time; the two solutions are distinct in $L^\infty_t(L^1\cap L^q)_x$ on $[0,T]\times\mathbb{R}^2$ for every $q\in(4/3,\infty)$ and every sufficiently large $T$. The nonstationary solution belongs to $C([0,\infty);L^q(\mathbb{R}^2))$ for every $q\in[1,\infty)$. Part I establishes a compactness criterion for non-uniqueness and conditional linear instability results for a vortex ring. Part II establishes the hypotheses of those results and proves the nonlinear non-uniqueness theorem.
Read the full abstract

Numerical Analysis

Laplace approximation of the parameter posterior with embedded error for efficient uncertainty quantification of expensive model calibrations

In practical engineering, there is a clear need for uncertainty quantification (UQ) of models used in design because forward UQ puts probabilities on performance predictions. However, without surrogate building, most forward engineering models are too expensive for traditional UQ methods like Markov chain Monte Carlo (MCMC) sampling. Our goal is to provide inexpensive, approxi- mate UQ that can be applied after an expensive traditional least squares fit, where the UQ handles both uncorrelated noise and parameter variance as in the embedded-error treatment developed by Sargsyan, Najm, and Ghanem (2015). After developing the basic theory to connect a Laplace-like approximation to Bayesian calibration, we compare the proposed method to black-box variational inference in terms of cost and fidelity. Then we demonstrate the method on real-world calibration of solid material models.
Read the full abstract

All authors Reese E.Jones, Kyle N. Karlson, Matt W. Kury, Moe Khalil, Ray S. Tuminaro

Computation and Language

Compact Greek Language Models from Scratch: Pretraining 500M- and 1B-Parameter Decoder-Only Transformers

We present two compact decoder-only language models trained from scratch exclusively on Greek text: LexisLM-1B (1.09B parameters) and LexisLM-500M (502M parameters). Both use a dedicated 32k byte-level BPE tokenizer and a 22B-token pretraining corpus drawn primarily from Greek Wikipedia and FineWeb2. Both models complete the 22B-token schedule, with best held-out perplexities of 9.77 for LexisLM-1B and 11.31 for LexisLM-500M. Evaluation of both models on Greek-translated ARC, MMLU, and TruthfulQA shows above-chance performance on ARC-Easy, while ARC-Challenge and MMLU remain near chance. We also perform supervised fine-tuning of both models on Greek instruction data, primarily EuroAlpaca, reaching held-out instruction perplexities of 7.12 for LexisLM-500M and 6.37 for LexisLM-1B. The report describes the tokenizer, data preparation, training, and evaluation methods, providing a foundation for further work on compact Greek language models.
Read the full abstract

All authors Konstantinos Perifanos, Anne Laure Poite, Giannis Tsilikas, Ilias Katis, Stavros Deligiannidis, Dionysis Goutsos

Fluid Dynamics

Reaching separated flow states by release and recapture with learned latent feedback

Targeting a prescribed flow configuration rather than an aerodynamic coefficient is challenging because the desired state may lie outside the natural attractor. We demonstrate cluster-based control in a learned vorticity representation for two-dimensional flow over a NACA0015 airfoil at $Re=500$ and $\alpha=15^\circ$, using a localized streamwise momentum source near the leading edge. An actuation-enriched ensemble is compressed with a variational autoencoder and partitioned into macrostates. We optimize a continuous feedback law interpolating macrostate-dependent amplitudes without an explicit latent dynamical model. For an attached target, the optimized law approaches constant forcing, arrests vortex shedding and maintains the attached state. For a strongly separated target, the law instead establishes attachment, withdraws actuation so that the natural separation transient carries the flow through the target, and restores forcing to recapture attachment. Over the evaluation horizon, the closed-loop trajectory revisits the target approximately every three shedding periods. Companion open-loop calculations isolate the arrest, release and recapture processes.
Read the full abstract

Genomics

Synaptic Dendritic Neuron Model (scDNM) for Cell Type Classification in Single-Cell RNA Sequencing Data

Cell type classification in single-cell RNA sequencing (scRNA-seq) data is challenging because of high dimensionality, sparsity, and transcriptionally similar cell types. We developed scDNM, an adaptation of the dendritic neuron model incorporating learnable synaptic thresholds, a supervised gene pool, staged training, and ensemble prediction. Six binary classification tasks were constructed from the Paul15, PBMC3k, and Nestorowa datasets. All models were evaluated using leakage-free nested cross-validation, with feature selection, scaling, and hyperparameter tuning performed independently of the outer test folds. scDNM ranked first in three of six tasks (including one tie) and achieved the best mean accuracy rank (2.00), followed by XGBoost (2.75), Random Forest (3.33), logistic regression (3.33), and SVM (3.58). In a six-class PBMC3k experiment, scDNM achieved 95.997% accuracy and 0.944 macro-F1, compared with 94.243% and 0.914 for SingleR. Garnett returned only Unknown labels and was treated as a zero-coverage technical outcome. Ablation analysis supported the contributions of dendrites and synaptic thresholds. However, XGBoost achieved higher mean binary-task accuracy (96.20% versus 95.32% for scDNM), and across-task rank differences were not statistically significant. These findings demonstrate that scDNM is competitive for scRNA-seq cell type classification, although broader independent validation is required.
Read the full abstract

Quantum Physics

Symplectic Reck: In-Situ Learning of Gaussian Quantum Operations

Programmable photonic circuits are a natural platform for physical learning, where hardware parameters are trained directly from measured feedback. We propose a physical learning algorithm and architecture for continuous-variable quantum systems where a photonic circuit is trained in situ to invert an unknown lossless Gaussian quantum operation. We prove that a generic symplectic matrix $S\in\mathrm{Sp}(2N,\mathbb{R})$ admits a decomposition into a triangular array of two-mode symplectic gates on neighboring modes, followed by single-mode symplectic gates. We then introduce the symplectic Reck mesh, which physically realizes this factorization, generalizing the unitary Reck mesh widely used in optical neural networks. When placed after an unknown Gaussian unitary, the mesh is trained in situ by nulling local response blocks one at a time using optical probes and measurements. We identify a local condition number that quantifies the squeezing required for a nulling step and its sensitivity to measurement error. We numerically demonstrate finite-shot training and robustness to systematic gate-calibration errors. Our work extends self-configuring optical networks from unitary to symplectic transformations, bringing physical learning to continuous-variable quantum photonic hardware.
Read the full abstract

All authors Janet F. Zhong, Renwen Yu, Paul-Alexis Mor, Charles Roques-Carmes, Aviv Karnieli, David A. B. Miller, Shanhui Fan

Optimization and Control

Linear Convergence of Nesterov's Acceleration for All Momentum Parameter

We study the convergence rate of the Nesetrov's accelerated proximal gradient method for a smooth, strongly convex composite minimization objective across a range of value of the momentum extrapolation. The literature proves a convergence rate of the method for a specific momentum extrapolation parameter choice based on the knowledge of strong convexity and Lipschitz smoothness constants. For the heavy ball method it is established that every choice of the momentum parameter either fails to accelerate or fails to converge on the same function class. We prove in this paper that a range of Nesterov style constant momentum extrapolation parameters produces a linear convergence rate. We corroborate these results numerically via performance estimation which computes a method's exact worst case rate as a semidefinite program.
Read the full abstract

High Energy Physics (Phenomenology)

Real Intermediate States in Dark Matter Freeze-In from First Principles: Single Mediator

We investigate the treatment of real intermediate states (RIS) in freeze-in production of $s$-channel Dark Matter (DM) with unstable mediators, as such models feature a double-counting issue when decay and annihilation collision terms are simply added within the semi-classical Boltzmann approach and a finite width is used for the regularization of the resonance. While several ad-hoc RIS-subtraction schemes are available to mitigate this problem, a first-principles treatment in this context is so far missing. We therefore start from the Closed-Time-Path (CTP) formulation of non-equilibrium Quantum Field Theory and present a CTP-based description of $s$-channel freeze-in pair production of a single scalar, fermion or vector DM species. In addition to being intrinsically free of non-unitary overcounting, the CTP ansatz also has the advantage of systematically capturing thermal corrections known to be important for DM freeze-in. Separate evolution equations assuming a single scalar or vector mediator in thermal equilibrium with the surrounding bath are presented. The exact propagators of the unstable mediators are approximated by resumming one particle irreducible (1PI) self-energy diagrams generated by bath particles and are evaluated with their full mass and momentum dependence in contrast to the commonly adopted Hard Thermal Loop limit. The CTP relic densities are compared numerically with those obtained using the most common RIS-subtraction schemes for two example $s$-channel freeze-in models, demonstrating the phenomenological relevance of a consistent RIS treatment.
Read the full abstract

All authors Jannis Brirup, Adrian Finke, Michael Klasen, Luca Paolo Wiggering

High Energy Physics (Theory)

Large-D approach to nearly critical holographic matter and Yang-Mills deconfinement

We study a systematic expansion of dynamical black hole solutions in a class of five-dimensional Einstein-dilaton theories that are holographically dual to nearly critical, strongly coupled four-dimensional field theories. The expansion is motivated, through dimensional reduction, by the large-D limit of pure gravity, where D is the number of spacetime dimensions. At leading nontrivial order, the Einstein equations reduce to a closed set of non-relativistic hydrodynamic equations in four dimensions, as first shown in our previous letter; these equations generalize the large-D effective theory of black branes. We derive the thermodynamics and hydrodynamics of the system from black hole thermodynamics and the holographic dictionary, and show that the energy-momentum tensor extracted from the near-boundary geometry is conserved precisely when the effective equations hold. We analyze planar, cylindrical, and spherical domain walls between coexisting phases, as well as non-equilibrium steady-state walls with nonzero momentum flux. We also show how the near-horizon geometry can be completed by a holographic flow to an asymptotically AdS5 region. Finally, as an application to the confinement-deconfinement transition of Yang-Mills theory, we construct the domain wall between the confined and deconfined phases in improved holographic QCD, compute its surface tension, and compare it with recent lattice results.
Read the full abstract

Machine Learning

Semiprime Factorization with Discrete Diffusion and Beam Search: Scaling Laws

We demonstrate that transformer architectures trained with discrete denoising diffusion and decoded via guided test-time beam search can learn to recover the prime factors of semiprime integers directly from data. Our approach establishes a new frontier for learning-based semiprime factorization, successfully factoring 68-bit integers-a substantial advance over the prior 56-bit state of the art. While computationally less efficient than classical factoring algorithms, our model requires no domain-specific number-theoretic priors and learns purely from training examples. In practice, the diffusion model acts as a generative prior, identifying promising factor candidates where exact verification or lattice-based post-processing can complete the factorization. Furthermore, we derive empirical scaling laws relating approximation quality to training data volume, model size, and test-time computation. In contrast to language, problem size is the key driver of scaling behavior. Achieving a fixed accuracy on a held-out test set requires compute that scales exponentially with bit length. Consequently, our findings pose no threat to security, as the projected compute cost required to factorize cryptographically relevant integers remains prohibitive.
Read the full abstract

All authors Brendan O Donoghue, Nenad Tomasev, Volodymyr Mnih, Timothy Nguyen, Dalin Guo, Arun Nair, Kenneth Yang, Alexandre Frechette, Jasper Xian, Jean Tarbouriech, Pavel Tyletski, Ingrid von Glehn, Johannes Bausch, Demis Hassabis, Raia Hadsell

Atmospheric and Oceanic Physics

New Astronomical, Meteorological and Geological Study of the "Torre Spaccata District" - VII district of the Municipality of Rome

This study, part of the NGICS (New Italian City Geological Study) project, analyzes 25 years of multi-disciplinary data to assess long-term environmental, climatic, and geological trends in Torrespaccata, an urban district within Rome's Seventh Municipality (VII Administrative District). Specifically, it investigates the temporal evolution of Heating Degree Days (HDD) by comparing the historical stationary climate period (1961-1989) with the recent multidecadal interval (1990-2014). Statistical analysis reveals a clear climatic transition: while the pre-1989 period showed a stationary trend (m = -1.14 HDD/year), the modern era exhibits a sharp decline in heating demand, characterized by a negative trend of -11.52 HDD/year (R2 = 0.511). This significant reduction of 80.14 HDD between the two periods is robustly validated via Welch's t-test (t = 3.045, p = 0.0039). Concurrently, interannual variability increased by 29.8%, driven by the synergistic feedback between global warming and the Urban Heat Island (UHI) effect. Furthermore, the correlation between climatic anomalies and local geological events is evaluated using robust statistical imputation to mitigate uncertainties caused by data gaps. These findings provide critical insights into long-term anthropic and natural changes within the urban ecosystem.
Read the full abstract

Combinatorics

Brown-Gap Healing and Exact Completeness Thresholds for Leading-Unit Positive Linear Recurrences

We determine the exact completeness threshold for standard positive linear recurrence sequences whose coefficient lists consist of leading units, followed by zeros and a final positive coefficient. Using the generating function of Brown's gaps, we prove a necessary and sufficient condition for completeness, including the boundary case in which the critical gap vanishes. We further establish strict monotonicity of the unrounded threshold in the number of leading units, thereby resolving the corresponding leading-unit monotonicity conjecture.
Read the full abstract

Computer Vision

RigidGS: Geometry-Aware Object Decoupling for Rigid-Body Interaction in 3D Gaussian Splatting

Interactive VR/MR applications demand photorealistic 3D representations that support real-time rendering and object-level manipulation. While 3D Gaussian Splatting (3DGS) achieves high rendering quality, it represents scenes as unstructured, monolithic primitives. Existing object-aware 3DGS methods jointly optimize geometry and semantic states but suffer from two key failure modes: noisy foundation-model masks on complex geometry, and label-geometry coupling, in which co-optimized identity gradients mislabel boundary Gaussians. We introduce RigidGS, an end-to-end pipeline that decouples geometric learning from object identity. We propose a depth- and temporally-aware mask refinement module that addresses two complementary SAM2 failures: over-inclusion on thin structures and under-inclusion on multimodal-depth objects. We then determine object identities entirely upstream via COLMAP projection voting and freeze them during training, directing mask supervision solely toward geometric optimization. Refined masks alone substantially improve competing methods, yet a residual gap in label-geometry coupling persists that our fixed-identity design eliminates. RigidGS outperforms prior object-aware 3DGS methods on LLFF/NVOS, Mip-NeRF~360, and iNeRF360, improving segmentation and boundary IoU by 7.7 and 14.5 points on LLFF/NVOS. Finally, RigidGS enables real-time rigid-body manipulation on Meta Quest~2 without mesh extraction. Our code and project page are available at https://gyreyirisansah.github.io/rigidgs.
Read the full abstract

Machine Learning

BO-Arena: An Evolving Benchmark for High-Dimensional Bayesian Optimisation

High-dimensional Bayesian optimisation (HDBO) is an increasingly crowded field with new methods proposed at every major conference, making it ever harder to answer a deceptively simple question: "Which method is currently state-of-the-art?" In this paper, we question the efficacy of benchmarking practices in the field, and find them to be systematically ineffective: multiple published works fail to benchmark against the strongest available baselines, or even include incorrect implementations of key methods. Without a shared reference implementation, such errors are difficult to detect during peer review. We therefore introduce BO-Arena, an actively maintained and extensible package providing canonical implementations of state-of-the-art algorithms, which allows researchers to develop new methods and benchmark them rigorously. Using BO-Arena, we find that what currently drives performance in HDBO is even simpler than previously thought. We use these insights to propose a new algorithm which is competitive with the state of the art on problems with many observations, without resorting to the complex GP approximations used in recent methods. Finally, we question whether there is currently sufficient evidence that any more complex method outperforms a simple Gaussian process baseline for HDBO.
Read the full abstract

All authors Colin Doumont, Thomas Christie, Henry Moss, Philipp Hennig

Medical Physics

Physics-Based Differentiable Beamforming for Deep Learning-Based Contrast Enhanced Ultrasound Super-Resolution Imaging

Ultrasound is widely used for vascular imaging because it is non-invasive, real-time, and cost-effective. However, its spatial resolution remains limited by diffraction to an order of the acoustic wavelength. This limitation is particularly restrictive for imaging of the microvasculature and high-resolution vector-flow-imaging. In this work, we present a multi-channel physics-based deep learning approach to super-resolve monodisperse microbubbles directly from multi-channel radio-frequency (RF) data. We also present a novel image-based metric for model evaluation. The proposed pipeline combines a learnable deconvolution network with a differentiable beamforming layer, enabling a direct deconvolution of the microbubble (MB) signature prior to image formation. By operating on multi-channel RF data, the method aims to improve localization robustness in the presence of dense microbubble populations and deconvolve the lateral direction in a relatively low-frequency regime (1.7 MHz) where MBs exhibit nonlinear behavior. The model is trained and evaluated on simulated data using monodisperse lipid-coated MBs over a broad range of acoustic pressures and microbubble densities. The proposed method improves the signal-to-noise ratio (SNR) compared to the single-channel approach. It achieves sub-wavelength axial and lateral resolution while slightly improving robustness to dense microbubble populations outside the training domain. It also demonstrates generalizability across transmit configurations without requiring explicit training on each transmit configuration.
Read the full abstract

All authors Anass Hameddine, Bram de Wilde, Michel Versluis, Jelmer M. Wolterink, Guillaume Lajoinie

Combinatorics

Questions of minimality for DDMOGs and cycle ornaments

Difference distance magic oriented graphs (DDMOGs) are oriented graphs whose vertices admit a bijective labeling by consecutive positive integers such that every vertex has weight zero, where the weight is the sum of the labels of its in-neighbors minus the sum of the labels of its out-neighbors. We investigate minimality and sparsity questions for DDMOGs and their ornaments. First, we develop new constructions of cycle-based ornaments that reduce the number of cycles required for a given number of attachment nodes and identify minimal cycle lengths in several cases. Second, we determine the minimum number of arcs in a nontrivial DDMOG of order $n$, combining computational enumeration for small orders with theoretical lower bounds and constructions. Langford sequences provide a key tool in these constructions, enabling us to partition consecutive labels into difference triples and thereby construct sparse DDMOGs. Finally, we use related constructions to obtain generalized ornaments with no internal arcs. These results advance the study of minimal constructions in difference distance magic labeling and provide new methods for building sparse DDMOGs.
Read the full abstract

All authors Roza Aceska, McKailyn Lort, Allison Ripperger, Zach Wilcher

Signal Processing

Two-Timescale Receive Combining for Fronthaul-Constrained Uplink XL-MIMO with Visibility Regions

Extremely large-scale multiple-input multiple-output (XL-MIMO) arrays are hard to deploy in virtualized radio access networks, because the fronthaul data rate grows with the number of antennas when the radio unit forwards every antenna sample. We propose a two-timescale receiver that allocates the per-user combining ranks under the fronthaul stream budget from the channel statistics. The radio unit applies the long-timescale transform, after which the distributed unit applies a minimum mean-squared-error receiver in the reduced dimension. For disjoint visibility regions, we prove that the reduced-dimension receiver decouples across users. We characterize the sum-rate loss to second order and show that the greedy allocation, which we call saturating capture, minimizes its leading term. In simulations, the allocation gains over a uniform-rank baseline, matches a budget-trimmed threshold rule at a small number of users and exceeds it as the array and the user count grow. The gain grows with the heterogeneity of the per-user eigenvalue profiles and vanishes for identical profiles. Under out-of-cell interference offset by at least half a window width, the gain survives with clean covariances, which window-restricted estimation keeps for disjoint windows.
Read the full abstract

Instrumentation and Detectors

Test beam results for machine-learning-based reconstruction of resistive silicon sensors

The advent of Low-Gain Avalanche Diodes (LGADs) and AC-coupled Low-Gain Avalanche Diodes (AC-LGADs) as candidate technologies for four-dimensional tracking has moved part of the reconstruction burden onto the analog waveforms of the sensor, in which the charge shared between electrodes encodes the impact position well below the electrode pitch. Recovering that position is a continuous, non-linear inverse problem. Transient Current Technique measurements characterize this problem at fixed injected charge and high signal-to-noise, whereas a minimum-ionizing beam adds Landau fluctuations in the deposited charge under detector operating conditions. We report test beam results for full-waveform reconstruction of an AC-LGAD of 500 $\mu$m pitch read out on a $2\times2$ pad matrix, exposed to the 120 GeV pion beam of the CERN Super Proton Synchrotron (SPS) and read out alongside a six-plane MIMOSA-26 telescope. The telescope is aligned, and its per-plane resolutions are calibrated in situ, giving a track pointing resolution of about 2.5 $\mu$m per axis at the device, so that the measured residual is the sensor response rather than a convolution with the reference. We then compare two analytic estimators, the parameter-free charge-imbalance formula and a fitted linear matrix inversion, with the topology-agnostic transformer developed for resistive silicon sensors. Over the full active pad, the transformer reaches a median total position with up to factor 3 improvement. Within the charge-sharing core, the three methods converge to the floor of this sensor at the achieved signal-to-noise.
Read the full abstract

All authors Don C. Wong, Leena Diehl, Spencer Ellis, Marcos Fernández García, Gabriele Giacomini, Vagelis Gkougkousis, Ben Kilminster, Jingyu Luo, Anna Macchiolo, Yajna Mahabeer, Yevhenii Padniuk, Daria Senina, Samantha Sunnarborg, Jessica Tang, Alessandro Tricoli, Lixing Wang, Gaetano Barone

Artificial Intelligence

Ensemble Routing for Prompt-Adaptive Cost-Aware LLM Inference

Large language models (LLMs) increasingly support Web-based services that must balance response quality and inference cost across heterogeneous requests. Conventional routing selects one model per prompt, while fixed ensembles cannot adapt their composition and cost to individual requests. We introduce PISA, a prompt-adaptive, cost-aware single-model router, and PISA-E, its extension to ensemble routing. At the core of both methods, PISA combines soft cluster memberships with model-specific predictors to estimate individual model capabilities. Building on these estimates, PISA-E jointly selects ensemble composition and cardinality under an explicit cost–quality preference that can be adjusted at inference time without retraining. A tractable ensemble-success surrogate and efficient candidate selection enable this extension without ensemble-specific training. Across four LLM routing benchmarks under in-distribution (ID) and out-of-distribution (OOD) conditions, PISA achieves the highest macro-average cost–performance hypervolume among the evaluated single-model routers in both settings. Meanwhile, PISA-E achieves the highest macro-average task performance among the evaluated ensemble approaches when prioritizing response quality, surpassing the strongest competitors by 2.09 and 2.87 percentage points under ID and OOD conditions, respectively. At selected lower-cost operating points, it also exceeds the strongest ensemble competitors' macro-average task performance while reducing average inference cost by approximately 36% and 25%, respectively. These results demonstrate the effectiveness of prompt-adaptive capability estimation for both single-model routing and selective ensembling.
Read the full abstract

All authors Saul Urso, Emanuele Carlini, Patrizio Dazzi, Matteo Mordacchini

Audio and Speech Processing

Lightweight Neural Prediction for MDCT-Based Perceptual Audio Coding

Perceptual audio codecs often use predictive coding to further reduce signal redundancy. Rather than learning the entire codec pipeline end-to-end, we insert a small neural network as a drop-in, deterministic predictor in an otherwise conventional codec: the network estimates the next Modified Discrete Cosine Transform (MDCT) coefficient block from a short context of reconstructed blocks, and the encoder transmits the quantized residual. We compare Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN) predictors against a non-predictive baseline and a per-line Normalized Least Mean Squares (NLMS) linear predictor (similar to the MPEG-2 AAC Main Profile predictor). The CNN consistently outperforms the MLP, the linear predictor, and the baseline in signal-to-noise ratio (SNR), noise-to-mask ratio (NMR), and MUSHRA evaluation scores. Using a 512-sample analysis window (w = 512), our neural predictors recover, and the CNN exceeds, the coded quality obtained by the non-predictive baseline after quadrupling the analysis-window length to w = 2048, at substantially lower algorithmic delay. The CNN uses only 0.13M parameters (roughly 20x fewer than the MLP) and has a decoder real-time factor (RTF) of ~0.10 on a consumer CPU.
Read the full abstract

2611.10192 · 6 Nov · 10 pages, 2 figures, 9 tables. Presented at the 161st AES Convention, Nashville, TN, USA, October 30-November 1, 2026 (published version: https://aes.org/publications/elibrary/elibrary-page/?id=23480). Audio samples and supplemental materials: https://github.com/NiccoloAbate/neural-predictive-audio-coding

Data Structures and Algorithms

Strong Connectivity Oracle for Multiple Failures via Rooted Flow-Augmentation

We present a significantly improved $k$-fault-tolerant strong connectivity oracle, one of the most basic data structures on a directed graph $G$ that, given any set $F$ of failed edges of size $|F|\le k$, answers whether $G \setminus F$ is strongly connected. For every constant $k$, our oracle only uses $n^{1+o(1)}$ space and $n^{o(1)}$ query time. No prior result can achieve $o(n^{2})$ space and $o(n)$ query time, \emph{even when $k=2$}. Previous oracles either require $\Omega(n)$ query time [BCR18], or $\Omega(n^2)$ space [vdBS19]. Our main technical contribution is a rooted version of the highly successful flow augmentation technique, which may be of independent interest. Given a directed graph $G$, an integer $k$, and a source vertex $s$, we can sample a set of edges $A$ such that each minimal $s$-rooted cut of size at most $k$ becomes an $s$-rooted min-cut in $G + A$ with probability $2^{-poly(k)}n^{-o(1)}$.
Read the full abstract

Probability

A Geometric Construction of the Radial Dunkl and Heckman-Opdam Processes

We provide a geometric construction of the radial Dunkl and Heckman-Opdam processes for arbitrary root systems and arbitrary nonnegative real root multiplicity parameters. This is a broad class of processes including the classical Bessel process, Dyson Brownian motion, and the Dyson Bessel process. We find that, when $x_t$ is a certain anisotropic Markov diffusion process in a symmetric space $G/K$ or its tangent space at the origin, invariant under an appropriate $K$-action, the evolution in the space of orbits can be described as a radial Dunkl or Heckman-Opdam process (respective to whether the diffusion is on the tangent space or $G/K$ itself). These processes can have arbitrary nonnegative real root multiplicities, analogous to arbitrary values of $\beta$ in Dyson Brownian motion. In a freezing limit, analogous to $\beta \to \infty$ for Dyson Brownian motion, these processes become motion by mean curvature of the orbits. A connection to finite free probability is also discussed, in which we provide a geometric interpretation of certain properties related to finite free position.
Read the full abstract

Data Structures and Algorithms

Flexible Edge Cover: A Disjunctive Programming driven Approximation

We introduce the flexible edge cover problem, denoted FLEX-EDGE-COVER: the input here is an undirected graph with non-negative edge costs along with a partition of the edges into safe and unsafe edges, non-negative integers $p$ and $q$, and the goal is to find a minimum cost subset of edges such that each vertex has either at least $p$ safe edges incident to it or at least $p+q$ total edges incident to it. We show that FLEX-EDGE-COVER is APX-hard in simple bipartite graphs. A greedy approach leads to a $2$-approximation in all graphs. We show that the $2$-factor can be improved in simple graphs via a novel disjunctive programming-based LP formulation. AI-tools were used in the analysis of our rounding algorithm. Our motivation for introducing and investigating FLEX-EDGE-COVER is its close connection to the Flexible Graph Connectivity problem (denoted $(p,q)$-FGC) which has garnered much recent attention. For $(p,q)$-FGC, we develop a tighter LP-formulation via the disjunctive programming approach. We hope that our disjunctive-programming based LP and rounding algorithms for FLEX-EDGE-COVER will serve as a starting point towards a constant factor approximation for $(p,q)$-FGC.
Read the full abstract

Machine Learning

BiMuon: A Geometry-Aware Stochastic Bilevel Optimization Approach

Stochastic bilevel optimization provides a general framework for many machine learning applications, such as hyperparameter learning, data reweighting, language model alignment, and actor-critic reinforcement learning. However, most existing bilevel optimization methods treat matrix-valued gradients as flattened vectors and perform optimization in the resulting vectorized space, ignoring the matrix structures common in modern neural networks. Moreover, the lower-level (LL) problem curvature can induce significant spectral imbalance in the upper-level (UL) geometry. Such limitations of existing bilevel optimization methods motivate us to develop BiMuon, a geometry-aware stochastic bilevel optimization method that integrates implicit hypergradient estimation in bilevel optimization with Muon-style orthogonalized updates at the UL. To our knowledge, BiMuon is the first stochastic BLO method with Muon-style orthogonalized matrix updates and an $O(\epsilon^{-4})$ iteration-complexity guarantee, matching the best-known rates achieved by Muon-type methods based on matrix parameterizations. To improve computational and parallel efficiency for models with multiple matrix parameters, we further study BiMuon for blockwise diagonal matrix approximation and derive a relative stepsize allocation rule based on UL progress and blockwise LL sensitivity. We evaluate BiMuon on representation learning and blockwise bilevel optimization, providing empirical support for our methods.
Read the full abstract

All authors Zhen Qin, Zhiyao Zhang, Yingbin Liang, Jia Liu

Machine Learning

Fast Accurate Quantum Monte Carlo without Metropolis Adjustment

In quantum systems, physical quantities of interest can be expressed as expectation values of a probability distribution, which allows their estimation by Markov Chain Monte Carlo (MCMC) methods. This approach, an instance of Quantum Monte Carlo (QMC), plays a central role in obtaining predictions of system properties relevant to condensed matter and high energy physics. The use of Hamiltonian dynamics to design a Markov kernel, known as Hybrid or Hamiltonian Monte Carlo (HMC), is widespread in quantum applications. In this context, it is standard to use an HMC kernel that is “adjusted” with the Metropolis-Hastings (MH) criterion, so that estimates of expectations converge to their true value in the limit of infinite sample size. However, recent work in computational statistics suggests that an unadjusted HMC kernel can be significantly more efficient, while safely maintaining an asymptotic bias which is small relative to the error arising from the variance of the finite sample size. We adapt this approach to the QMC setting, focusing on a challenging and high-dimensional spin-fermion model of a quantum phase transition. We find that unadjusted methods outperform adjusted HMC in terms of effective sample size per gradient call, while retaining accurate results.
Read the full abstract

All authors Reuben Cohn-Gordon, Gabriel Pescia, Sumner N. Hearth, Jakob Robnik, Uroš Seljak, Peter Lunts