Peer Revenue Past citations are no guarantee of future returns

Papers

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

Mathematics

math Watch Mathematics (sign in to use)

Thursday 7 May

Commutative Algebra

The derived depth formula for modules of finite quasi-projective dimension

Let $R$ be a commutative Noetherian local ring. We prove a variety of new formulae for modules of finite quasi-projective or finite quasi-injective dimension. These include the Derived Depth Formula, itself an extension of Auslander famous depth formula, a variation of the Derived Depth Formula for width, an extended version of Ischebeck's Formula, and a Dependency formula in the vein of Jorgensen. Several special cases of our main results are new even under stronger assumptions on the vanishing of various complete intersection dimensions.
Read the full abstract

Statistics Theory · Cross-listed from cs.AI

Adaptive auditing of AI systems with anytime-valid guarantees

A major bottleneck in characterizing the failure modes of generative AI systems is the cost and time of annotation and evaluation. Consequently, adaptive testing paradigms have gained popularity, where one opportunistically decides which cases and how many to annotate based on past results. While this framework is highly practical, its extreme flexibility makes it difficult to draw statistically rigorous conclusions, as it violates classical assumptions: the number of observations is typically limited (often 10 to 50 cases) and decisions regarding sampling and stopping are made in the midst of data collection rather than based a pre-specified rule. To characterize what statistical inferences can be drawn from highly adaptive audits, we introduce a hypothesis testing framework from two "dueling" perspectives: (i) the model's null that asserts there is no failure mode with performance below a target threshold versus (ii) the auditor's null that asserts they have a sampling strategy that will uncover a failure mode. Leveraging Safe Anytime-Valid Inference (SAVI), we formalize the auditor as conducting "testing by betting," which translates into simultaneous e-processes for testing the dueling null hypotheses. Furthermore, if the auditor is sufficiently powerful, we prove that these two hypotheses are asymptotically inverses of each other, in that passage of such an audit certifies the AI system as being globally robust. Empirically, we demonstrate that our proposed testing procedures maintain anytime-valid Type-I error control, outperform pre-specified testing methods, and can reach statistically rigorous conclusions sometimes with as few as 20 observations.
Read the full abstract

All authors Siyu Zhou, Patrick Vossler, Venkatesh Sivaraman, Yifan Mai, Jean Feng

Combinatorics

A note on the extensible no-three-in-line problem

We show the existence of a set $S\subset\mathbb{Z}^2$ avoiding collinear triples satisfying $|S\cap [n]^2|=\Omega(n/\sqrt{\log n})$ for sufficiently large $n$. This improves on the best-known lower bound on Erde's extensible no-three-in-line problem due to Nagy, Nagy and Woodroofe by $\sqrt{\log n}$, leaving the same gap to the trivial upper bound. Our construction is random.
Read the full abstract

Optimization and Control

A Semi-smooth Newton Method for the Constrained Optimal Control of Continuous-Time Linear Systems

This paper details a novel indirect method for solving constrained optimal control problems (OCPs) directly in continuous-time function space. The KKT conditions are embedded in a non-smooth complementarity function, which enables their reformulation as a rootfinding problem in Banach space. This problem is then solved using a non-smooth Newton method. Finally, the paper shows that the Newton update can be obtained by solving a modified differential Riccati equation, where the cost terms are reweighted at every iteration based on the constraint multipliers. Numerical simulations show the effectiveness of the method, which converges superlinearly up to the tolerance of the ODE solver.
Read the full abstract

Information Theory · Cross-listed from cs.MA

The Cost of Consensus: Malignant Epistemic Herding and Adaptive Gating in Distributed Multi-Agent Search

Distributed agents in real-world settings frequently must coordinate under uncertainty with only partial observations. Coordination is necessary to share beliefs to aid in task completion, but communication costs bandwidth, introduces latency, and if done poorly, can degrade collective reasoning. This tension is especially acute in bandwidth-constrained deployments such as distributed sensing networks, autonomous reconnaissance, and collaborative cyber defense, where excessive transmission carries direct operational costs. Existing work has focused on multi-agent exploration and communication strategies, but not on how communication frequency and content jointly shape the collective belief state. Central to this challenge is the degree to which agents maintain compatible internal beliefs about the environment, a property we term epistemic alignment. When agents share beliefs effectively, they converge on correct hypotheses; when communication is poorly designed, agents may converge confidently on wrong ones. We formalize this distinction and show it is not detectable from coordination metrics alone such as Jensen-Shannon Divergence or rate to consensus.
Read the full abstract

All authors David Farr, Iain Cruickshank, Kate Starbird, Jevin West

Logic

Definable groups and fields in t-minimal theories

Let $T$ be a theory which is t-minimal, meaning that with respect to some definable topology, a unary definable set $D \subseteq M$ has non-empty interior iff it is infinite. If $K$ is a definable field in $T$, then $K$ is finite or "large" in the sense of Pop: any smooth algebraic curve $C$ over $K$ with at least one $K$-rational point has infinitely many $K$-rational points. We also assign a canonical topology to any abelian definable group $G$ in a t-minimal theory. In the case where the t-minimal theory is "visceral" in the sense of Dolich and Goodrick, meaning that the definable topology is induced by a definable uniformity, we can drop the assumption of abelianity of $G$, and the resulting topology on $G$ is a definable manifold in the style of Acosta López and Hasson.
Read the full abstract

Quantum Algebra

A Complexity Dichotomy for Quantum Invariants of 3-Manifolds

We determine the complexity of exact evaluation of the Reshetikhin–Turaev and Turaev–Viro invariants of closed connected oriented 3-manifolds, with the underlying tensor category fixed. If $\mathcal C$ is a modular category, then the Reshetikhin–Turaev invariant $Z_{\mathcal C}(M)$ can be computed in polynomial time from a framed-link surgery presentation of $M$ precisely when $\mathcal C$ is pointed; otherwise the problem is $\#\mathrm{P}$-hard. If $\mathcal A$ is a spherical fusion category, then the Turaev–Viro invariant $|M|_{\mathcal A}$ can be computed in polynomial time from a triangulation of $M$ precisely when the Drinfeld center $\mathcal Z(\mathcal A)$ is pointed, equivalently when $\mathcal A$ is trivializable pointed; otherwise the problem is $\#\mathrm{P}$-hard. This proves the dichotomy conjectured by Bridges and Samperton and identifies the categorical obstruction to polynomial-time evaluation.
Read the full abstract

Numerical Analysis

Accelerating the Simulation of Ordinary Differential Equations Through Physics-Preserving Neural Networks

Numerical simulation of ordinary differential equations (ODEs) can be challenging when the system exhibits high accelerations and rapidly changing dynamics. Under these conditions the ODE solver often needs to take very small time steps in order to resolve the solution accurately, resulting in increased computational cost. In order to accelerate the simulation of these ODEs we present a novel methodology that uses a pseudo-invertible neural network to map system states into a high-dimensional latent-space. The network is then trained so that the dynamics in this learned latent space are slow, and can be simulated with relatively few function calls. Unlike existing neural methods, the latent dynamic equations are not learned from trajectory data, but derived from the original system equations and the chain rule. This allows the method to generalize better than existing approaches because the derived equations are correct by construction. In this work, we derive latent state equations of motion for any general ODE, and describe the loss function used to enforce slow time evolution of the latent states. We then apply this technique to multiple example ODEs and show that these problems can be solved with $3$x to $20$x fewer function calls for the same accuracy when simulating in the learned latent space. This reduction in cost could decrease computational demands for scientific simulations across engineering and physics applications.
Read the full abstract

Information Theory · Cross-listed from cs.LG

$f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses

Reinforcement Learning from Human Feedback (RLHF) has become a cornerstone technique for post-training large language models. While most existing approaches rely on the reverse KL-regularization, recent empirical studies have begun exploring alternative divergences (e.g., forward KL, chi-squared) as regularizers in RLHF. However, a unified theoretical understanding of general $f$-divergence regularization remains under-explored. To fill this gap, this work develops a comprehensive theoretical framework for online RLHF with a general $f$-divergence regularized objective. Rather than treating each possible divergence function individually, we adopt a holistic perspective across the entire function class and propose two algorithms based on distinct sampling principles. The first extends the classical optimism principle with a carefully designed exploration bonus, while the second introduces a new method that exploits the sensitivity of the optimal policy to reward perturbations under $f$-divergence regularization. Theoretical analysis shows that $O(\log T)$ regret and $O(1/T)$ sub-optimality gap are achievable, establishing provable efficiency of both algorithms and, to the best of our knowledge, the first performance bounds for online RLHF under general $f$-divergence regularization.
Read the full abstract

All authors Di Wu, Chengshuai Shi, Jing Yang, Cong Shen

Numerical Analysis

Efficient symplectic integrators for cubic and quartic potentials

We present a set of new, efficient high-order symplectic methods designed for Hamiltonian systems with cubic or quartic potentials. By demonstrating that polynomial potentials require fewer order conditions, we develop schemes that outperform both standard symmetric compositions of second-order methods and existing RKN splitting methods. Numerical results confirm their improved efficiency over state-of-the-art alternatives found in the literature.
Read the full abstract

Number Theory

Poissonian correlations of $\alpha n^d$ mod $1$

Let $x(n):=\alpha n^d \mod 1$ for integer $d >1$ and non-zero real $\alpha$. We show that $\{x(n)\}_{n>0}$ has Poissonian $\ell$-point correlations for almost all choices of $\alpha$ when $d$ is large (depending on $\ell$). This falls in line with the expected behavior from the Berry–Tabor conjecture. Further, in the spirit of a conjecture of Rudnick–Sarnak, we show Poissonian $\ell$-point correlations for a set of badly approximable $\alpha$ of full Hausdorff dimension by a Fourier analytic transference principle. The proof makes use of an application of the determinant method to count points on a diagonal hypersurface of degree $d$ in such a way as to capture the contribution of points belonging to lower dimensional varieties. As $d$ grows, these `special solutions' dominate the count and non-special solutions become increasingly rare. This stratified counting statement allows us to control the number of points on average very effectively.
Read the full abstract

Mathematical Physics

Quantitative propagation of chaos for Lindblad dynamics

We consider an open quantum system governed $N$-body Lindblad equation and study mean-field limits in this setting. We prove that the $N$-particle dynamics converges, in the sense of quantum relative entropy, to the tensorized solution of the limiting nonlinear equation. More precisely, we establish explicit bounds of order $1/N$ on the relative entropy between the $N$-particle density operator and the corresponding product state, thereby providing a quantitative propagation of chaos.
Read the full abstract

Geometric Topology

Translation Surfaces arising from Right Regular Prisms

We study flat metrics arising from right regular $n$-prisms by viewing them as $n$-differentials and analyzing their associated unfoldings. We show that the unfolding of a right regular $n$-prism is never a lattice surface unless $n=4$, in contrast with the case of Platonic solids. Despite this, we prove that these surfaces admit translation coverings to hyperelliptic surfaces, allowing us to determine their $\mathrm{GL}(2,\mathbb{R})$-orbit closures using the classification of hyperelliptic components of strata. As a consequence, we obtain exact quadratic asymptotics for a certain average of the number of saddle connections on the base surfaces, their unfoldings, and the original prisms, including their Siegel–Veech constants. This provides a natural infinite family of non-lattice surfaces for which orbit closures and counting problems can be computed explicitly.
Read the full abstract

Dynamical Systems

Expanding Maps on Flowers, Interval Exchange Transformations, and Ergodic Optimization

In this paper, we discuss expanding maps on a class of invariant sets called flowers. We show that any set contained in a flower has at most linear complexity, and we present a relationship between flowers and a special class of interval exchange transformations. This extends work of Bullett and Sentenac, who showed that any Sturmian system may be embedded into the circle as a doubling-invariant subset that is contained in a half circle. Flowers were first introduced in the context of ergodic optimization, as candidate sets for supporting maximizing measures. We discuss the relationship to ergodic optimization, and present numerical results that support the conjecture that trigonometric polynomials are maximized on flowers.
Read the full abstract

Optimization and Control

Unlocking Deep Demand Flexibility via Dynamic Signals

The rapid proliferation of distributed energy resources (DERs) and the electrification of residential loads offer significant potential for grid flexibility but pose stability challenges under static pricing regimes. Specifically, high levels of automation under static Time-of-Use (TOU) tariffs often induce “device synchronization,” where simultaneous responses from home energy management systems (HEMS) create artificial demand peaks that threaten grid stability. This paper proposes a privacy-preserving, one-way dynamic signaling framework to unlock deep demand flexibility from HEMS. We utilize a feedback-based learning algorithm that updates day-ahead price profiles based on aggregate substation demand and environmental contexts, effectively closing the loop between utility objectives and aggregated edge behaviors. The framework is rigorously validated using high-fidelity simulations on an 84-bus distribution network populated with hundreds of HEMS controlling diverse devices, including HVAC, PV, batteries, and flexible loads. Results demonstrate that the proposed mechanism achieves substantial reductions in both peak demand and total load variation. Extensive analyses across diverse climates and scalable deployments confirm the framework's robustness, indicating that dynamic pricing acts as a force multiplier for DERs, with peak shaving potential increasing significantly under high renewable penetration scenarios.
Read the full abstract

All authors Xinyang Zhou, Jing Shang, Andrey Bernstein, Stefan Wager, Moody Saleh, Lara Pierpoint

Statistics Theory · Cross-listed from stat.ML

Locally Near Optimal Piecewise Linear Regression in High Dimensions via Difference of Max-Affine Functions

This paper presents a parametric solution to piecewise linear regression through the Adaptive Block Gradient Descent (ABGD) algorithm. The heart of the method is the parametrization of piecewise linear functions as the difference of max-affine (DoMA) functions. A non-asymptotic local convergence analysis for ABGD is provided under sub-Gaussian covariate and noise distributions. To initialize ABGD, we adapt a prior algorithm originally developed for the simpler setting of max-affine functions. When suitably initialized, ABGD converges linearly to an $\epsilon$-accurate estimate given $\tilde{\mathcal{O}}(d\max(\sigma_z/\epsilon,1)^2)$ observations where $\sigma_z^2$ denotes the noise variance. This implies exact recovery given $\tilde{\mathcal{O}}(d)$ samples in the noiseless case. Also, such a rate is shown to be minimax optimal up to logarithmic factors. Synthetic numerical results corroborate the theoretical guarantees for ABGD. We also observe competitive performance compared to the state-of-the-art methods on real-world datasets.
Read the full abstract

Information Theory

Hybrid Multiport Receivers for Slow Fluid Antenna Multiple Access

We propose a novel receiver architecture that preserves the performance benefits of multiport selection in fluid-antenna systems while requiring only a very small number of radio-frequency (RF) chains. The resulting fluid-antenna hybrid multiport (FAHM) receiver effectively decouples port selection from signal combining by integrating a low-complexity analog combining network similar to those used in conventional hybrid multiantenna designs. We develop an effective-port criterion to determine the selected-port dimension before hybrid combining, thereby limiting the performance loss associated with port selection, and then design the hybrid combiner for a given RF-chain budget. The FAHM architecture is evaluated in a multiuser set-up operating under slow fluid-antenna multiple access (FAMA). In this scenario, a FAHM implementation with only 2 RF chains showcases a performance comparable to a fully-digital conventional multiport scheme with a much larger number of RF chains. Additionally, the proposed Rank-1 implementation attains a reduction in average GEPort execution time by about two orders of magnitude.
Read the full abstract

Algebraic Geometry

On the Bourbaki Degree of Plane Projective Curves

Let $F$ be a reduced plane projective curve defined over an algebraically closed field $k$. The Bourbaki degree of $F$, denoted by $\mathrm{Bour}(F)$, measures its freeness and was introduced in \cite{JNS2024}. It is defined as the degree of $R/I_\epsilon$, where $R = k[x,y,z]$ and $I_\epsilon \subseteq R$ is the Bourbaki ideal associated with a minimal generator $\epsilon$ of the module of first syzygies of the Jacobian ideal $J_F$. In this article, we investigate the local Bourbaki degree and apply it to refine the well-known upper bound for the global Bourbaki degree. We then study curves with prescribed Bourbaki degree through the minimal graded free resolution of $R/J_F$, describing the possible resolution forms and expressing $\mathrm{Bour}(F)$ in terms of the corresponding graded shifts. Finally, we study the realization problem in fixed degree. We prove that for $\deg F \geq 5$, the du Plessis-Wall bounds impose no numerical obstruction to the realization of the Bourbaki degree. Moreover, we show that every possible Bourbaki degree is realized by reduced plane curves of degrees $5$ and $6$.
Read the full abstract

2605.06956 · 7 May · The previous version has been substantially modified. We now address several problems concerning the Bourbaki degree, extending the scope beyond its local version. The title has been changed accordingly. 25 pages. Comments are welcome

Optimization and Control

Prescriptive Optimization for Adaptive Auto-insurance Pricing with Telematics Data

Usage-based insurance (UBI) uses telematics to align premiums with risk and encourage safe driving. However, deploying these programs is challenging due to heavy-tailed claim costs, nonstationary driver behavior, and limited incentive budgets. While existing research focuses on profiling drivers, prescriptive pricing remains underexplored. We propose an optimal control framework that integrates telematics directly into dynamic pricing. Our approach (i) learns claim frequency and severity, (ii) models multi-period behavioral evolution in response to discounts, and (iii) optimizes portfolio-wide discount allocation using a Lagrangian relaxation. This decomposes the non-convex centralized problem into independent dynamical systems. We theoretically prove this relaxation's duality gap vanishes as the portfolio scales, guaranteeing asymptotic optimality. We validate our approach computationally on a simulated industry-scale portfolio. Our results demonstrate not only the computational tractability of our approach but also that it outperforms static baselines, reducing both expected losses and claim probabilities to benefit insurers and policyholders alike.
Read the full abstract

Probability

Asymptotic Results for Uniform Group Drawing in the Coupon Collector's Problem

The article explores the asymptotic behavior of the expected number of drawings in the Coupon Collector's Problem with group-drawing under the uniform distribution. In this variant, each draw consists of a package of $s$ distinct coupons selected uniformly at random from a set of $n$ coupons. We focus on three regimes of the package size $s$: (i) constant $s$, (ii) $s$ proportional to $n$, and (iii) $s$ "very close" to $n$. For each case, we provide precise asymptotic expressions for the expected collection time. Keywords: Coupon Collector's Problem, Group Drawings, Uniform Distribution, Asymptotic Analysis, Expected Collection Time
Read the full abstract

2605.06953 · 7 May · This article and arXiv:2509.17201 together supersede the original preprint, following their acceptance as two separate papers in Stochastic Models. This part focuses on asymptotic results for uniform group drawing in the coupon collector's problem

Dynamical Systems

Analytical solutions for some quadratic ODEs found via linear rational eigenfunctions and the rational eigenfunction variety

Many important systems across biology, engineering, physics, and economics are characterized by polynomial ordinary differential equations (ODEs), yet analytical solutions are rare. We develop a framework for identifying and solving a broad class of two-dimensional quadratic ODEs using linear rational Koopman eigenfunctions. By imposing a linear rational form on the eigenfunctions, we convert the Koopman eigenfunction PDE into a large algebraic system of polynomials. We then study the solutions of this polynomial system that satisfy the ODE restrictions; we call the solution set the rational eigenfunction variety of an ODE system. The nonlinear algebra method uses formal algebraic geometry theory to analyze and solve systems otherwise intractable and to discover relationships between ODE and eigenfunction parameters that must hold to extract eigenfunctions. We identify families of quadratic ODEs that can be solved analytically, characterize their eigenfunction parameters, and use the resulting eigenfunctions to produce closed-form analytical solutions.
Read the full abstract

Optimization and Control

Low-Order Explicit Hessian Imitation Method for Large-Scale Supervised Machine Learning

An algorithm is proposed for solving optimization problems arising in neural network training for supervised learning. The unique feature of the algorithm is the use of an auxiliary loss, in addition to the original loss employed for model training. The purpose of the auxiliary loss is to provide a mechanism for creating a low-order Hessian-type approximation for the original loss. The proposed algorithm employs the resulting low-order second-derivative approximation terms in place of the second-order momentum terms (i.e., squared elements of the gradient of the loss function) in an overall scheme that has computational cost on par with an Adam-type approach. Whereas the squared elements of a gradient vector do not necessarily approximate second-order derivatives well, by careful construction of the auxiliary loss, second-order derivative-type approximations for the original loss can be computed and employed by the algorithm in an efficient manner. A convergence guarantee is provided for the proposed algorithm that is on par with guarantees available for similar stochastic diagonal-scaling methods. The results of numerical experiments show situations when the proposed algorithm outperforms Adam and other popular modern optimizers.
Read the full abstract

All authors Yunlang Zhu, Lingjun Guo, Zahra Khatti, Xiaoyi Qu, Chia-Yuan Wu, Lara Zebiane, Frank E. Curtis

Number Theory

On almost primes solutions to forms of odd degrees in many variables

Let $\mathcal{F}=\{f_1,\ldots,f_R\}$ be a family of forms of odd degrees at most $d$ in $s$ variables. We study the solutions to the system $f_1(\mathbf{x})=\ldots=f_R(\mathbf{x})=0$ of the form $x_i=y_ip_i$ with $|y_i|\leq Y_\mathcal{F}$ and $p_i$ being a prime for all $i\in [s]$ inside the box $[-N,N]^s$, for large $N$. We show that if the number of variables $s$ is sufficiently large with respect to the parameters $R$ and $d$, then there are at least $C_\mathcal{F} N^{s-D}/(\log\,N)^s$ such solutions for some constants $C_\mathcal{F}>0$ and $D\in\mathbb{N}$, with $D$ depending only on the initial parameters $R$ and $d$.
Read the full abstract

Optimization and Control · Cross-listed from cs.LG

Constrained Triple-Head Price Optimization (C3PO) Network for Discrete-Choice Models

We propose a Constrained Triple-Head Price Optimization (C3PO) network for real-time bi-level decision making in discrete-choice environments. In these settings, a seller selects product or service offerings to achieve business objectives, while heterogeneous users make personalized acceptance or rejection decisions based on their preferences, ultimately determining seller outcomes. Existing approaches that fine-tune pre-trained tabular foundation models on synthetic data are limited by their imitation-learning paradigm and inability to enforce common pricing rules, including cross-product and global constraints. C3PO addresses these limitations through a novel transformer-based architecture that jointly integrates price imitation learning, multi-task revenue prediction, and in-context elasticity learning to generate robust pricing recommendations across domains without retraining. A lightweight differentiable constraint layer enables generated prices to satisfy business constraints while optimizing the revenue objective. We demonstrate strong out-of-distribution performance on simulated, synthetic, and real-world datasets. Across simulated benchmarks where the calibrated MNL model is misspecified relative to the ground-truth choice model, C3PO improves expected revenue by 1.4 to 7.9 percentage points over a calibrated MNL baseline. The constraint layer reduces average absolute constraint violations by more than 50%. We further deploy the model in real-world applications, including healthcare tender pricing and airline ancillary pricing, where C3PO improves directional pricing effectiveness, measured by price-increase and price-decrease recall, by 18–34 percentage points depending on the baselines considered, including calibrated MNL, classical optimization, and off-the-shelf tabular foundation-models.
Read the full abstract

All authors Shivaram Subramanian, Zhengliang Xue, Markus Ettl, Yingdong Lu, Jayant Kalagnanam

Representation Theory

On the refined local Langlands conjecture for discrete $L$-parameters of inner forms of quasi-split disconnected real reductive groups

Given a quasi-split connected reductive $\mathbb{R}$-group $G$ and a finite group $A$ acting on $G$ by $\mathbb{R}$-automorphisms that preserve an $\mathbb{R}$-pinning, we construct for each discrete $L$-parameter for $G$ a corresponding $L$-packet of irreducible discrete series representations on each inner forms $\tilde G_z(\mathbb{R})$ of the disconnected group $\tilde G = G \rtimes A$. We prove that these $L$-packets satisfy the endoscopic character identities with respect to normalized transfer factors. This proves the conjectural refined local Langlands correspondence for inner forms of quasi-split disconnected real reductive groups, as recently formulated by the first author.
Read the full abstract

Probability

Law of large numbers for activated random walk on villages

We consider activated random walk (ARW), an interacting particle system and prototypical model of self-organized criticality in a setting which combines mean-field behavior with the geometry of an arbitrary graph, which we call the village model of ARW, or VARW for short. VARW is obtained from a fixed graph by replacing each vertex with a 'village' that consists of n replicas of that vertex. We focus on VARW where particles walk according to a strictly sub-stochastic transition kernel on a finite underlying graph, so mass is sometimes lost (which guarantees that the system eventually stabilizes almost surely). Under a subcriticality assumption on the initial state we prove a law of large numbers as n goes to infinity for the resulting stable configuration of particles and the odometer of the process, to a limit which is uniquely characterized by a system of non-linear equations.
Read the full abstract

Optimization and Control

Muon with Nesterov Momentum: Heavy-Tailed Noise and (Randomized) Inexact Polar Decomposition

Most first-order optimizers treat matrix-valued parameters as vectors, ignoring the intrinsic geometry of hidden-layer weights in neural networks. Muon addresses this mismatch by updating along the polar factor of a momentum matrix, but its theoretical understanding has lagged behind practice. In particular, practical implementations incorporate Nesterov momentum, compute the polar factor only approximately, and operate with stochastic gradients that may be heavy-tailed. We close this gap by developing a convergence theory for Muon with Nesterov momentum and inexact polar decomposition in non-convex matrix optimization under heavy-tailed noise. Our analysis builds on a unified framework for inexact polar decomposition that captures practical iterative approximations such as Newton-Schulz and quantifies how their errors propagate through the optimization dynamics. Under uniform bounds on polar errors, we establish an optimal iteration and sample complexity of $O \left(\varepsilon^{\frac{-(3\alpha-2)}{(\alpha-1)}} \right)$ for finding an $\varepsilon$-stationary point, where $\alpha\in(1,2]$~denotes the heavy-tail index. For the inexact-polar setting with $\sigma_1=0$, we also provide guarantees that do not require prior knowledge of $\alpha$. We analyze a randomized low-rank polar decomposition that is substantially more efficient than full-space methods while remaining compatible with our theory. Numerical experiments further demonstrate the effectiveness of the proposed inexact and randomized~variants.
Read the full abstract

All authors Sayantan Choudhury, Xiaoran Cheng, Martin Takáč, Sen Na, Mladen Kolar

Numerical Analysis · Cross-listed from cs.AR

EULER-ADAS: Energy-Efficient & SIMD-Unified Logarithmic-Posit Engine for Precision-Reconfigurable Approximate ADAS Acceleration

Advanced driver-assistance systems (ADAS) require neural compute engines that deliver low-latency inference under strict power and area constraints. Posit arithmetic is attractive for such accelerators because it provides high numerical fidelity at low precision, but its variable-length regime encoding increases encode/decode cost and exposes the datapath to large regime-field fault effects. This paper presents EULER-ADAS, a SIMD-enabled logarithmic bounded-Posit neural compute engine for energyefficient and reliability-aware ADAS acceleration. The proposed datapath combines bounded-regime Posit representation, stageadaptive logarithmic mantissa multiplication with bit truncation, and a SIMD-shared quire accumulation path supporting Posit- (8,0), Posit-(16,1), and Posit-(32,2) execution. The unified architecture enables 4xPosit-8, 2xPosit-16, or 1xPosit-32 operation without duplicating precision-specific hardware. FPGA implementation shows that the proposed configurations reduce LUT count by up to 41.4%, delay by up to 76.1%, and power by up to 71.9% relative to exact Posit neural compute engines, while achieving up to 10x lower energy-delay product than radix-4 Booth-based Posit multipliers. In 28-nm CMOS, the bounded variants occupy 0.013-0.016 mm2 , consume 19.8-22.1 mW, and operate at up to 1.84 GHz. Application-level evaluation across image-classification, ADAS, and edge-inference workloads shows that the evaluated Posit-16 and Posit-32 configurations remain within about 1.5 percentage points of FP32 accuracy. A TinyYOLOv3 prototype on Pynq-Z2 achieves 78 ms latency at 0.29 W and 22.6 mJ/frame, demonstrating the suitability of EULERADAS for low-power real-time ADAS inference.
Read the full abstract

All authors Mukul Lokhande, Ratko Pilipovic, Omkar Kokane, Adam Teman, Santosh Kumar Vishvakarma

Numerical Analysis · Cross-listed from stat.ML

One Operator for Many Densities: Amortized Approximation of Conditioning by Neural Operators

Probabilistic conditioning is concerned with the identification of a distribution of a random variable $X$ given a random variable $Y$. It is a cornerstone of scientific and engineering applications where modeling uncertainty is key. This problem has traditionally been addressed in machine learning by directly learning the conditional distribution of a fixed joint distribution. This paper introduces a novel perspective: we propose to solve the conditioning problem by identifying a single operator that maps any joint density to its conditional, thus amortizing over joint-conditional pairs. We establish that the conditioning operator can be approximated to arbitrary accuracy by neural operators. Our proof relies on new results establishing continuity of the conditioning operator over suitable classes of densities. Finally, we learn the conditioning map for a class of Gaussian mixtures using neural operators, illustrating the promise of our framework. This work provides the theoretical underpinnings for general-purpose, amortized methods for probabilistic conditioning, such as foundation models for Bayesian inference.
Read the full abstract

All authors Panos Tsimpos, Edoardo Calvello, Ayoub Belhadji, Nicholas H. Nelsen