研究园简易版 · 经典书籍 · 期刊文章 · 有原文 · 完整版(可多选筛选、搜索) · 首页
2025 年的期刊文章 · 第 30 页
共 1491 篇,按发表时间排,新的在前,每页 40 篇。这一页是 2025 到 2025。
按期刊、按主题看 →
Econometrica Other · 发表于 2025
Morten I. Lau, Hong Il Yoo
摘要
Structural measures of higher order risk attitudes have well‐developed foundations in Expected Utility Theory (EUT), but little is known about their empirical magnitudes. We introduce a novel experimental design and a companion econometric model that allows us to structurally estimate indices of risk aversion, prudence, and temperance under EUT without imposing restrictions on their interdependence. We find that indices of absolute risk aversion, prudence, and temperance exhibit distinct patterns of variation over income, and that predicted risk premia under EUT and Rank‐Dependent Utility Theory gradually converge as the order of risk increases. These findings are obscured by regular parametric utility functions, which inherently bias results toward prudence and temperance when subjects are risk averse. The results remain robust in subsamples of moderate size, which suggests that our approach can be adopted in broader studies that link higher order risk attitudes to other domains of latent individual preferences and economic behavior.
JMLR Machine Learning · 发表于 2025
Michael Albergo, Nicholas M. Boffi, Eric Vanden-Eijnden
摘要
A class of generative models that unifies flow-based and diffusion-based methods is introduced. These models extend the framework proposed in Albergo and Vanden-Eijnden (2023), enabling the use of a broad class of continuous-time stochastic processes called stochastic interpolants to bridge any two probability density functions exactly in finite time. These interpolants are built by combining data from the two prescribed densities with an additional latent variable that shapes the bridge in a flexible way. The time-dependent density function of the interpolant is shown to satisfy a transport equation as well as a family of forward and backward Fokker-Planck equations with tunable diffusion coefficient. Upon consideration of the time evolution of an individual sample, this viewpoint leads to both deterministic and stochastic generative models based on probability flow equations or stochastic differential equations with an adjustable level of noise. The drift coefficients entering these models are time-dependent velocity fields characterized as the unique minimizers of simple quadratic objective functions, one of which is a new objective for the score. We show that minimization of these quadratic objectives leads to control of the likelihood for generative models built upon stochastic dynamics, while likelihood control for deterministic dynamics is more stringent. We also construct estimators for the likelihood and the cross entropy of interpolant-based generative models, and we discuss connections with other methods such as score-based diffusion models, stochastic localization, probabilistic denoising, and rectifying flows. In addition, we demonstrate that stochastic interpolants recover the Schrödinger bridge between the two target densities when explicitly optimizing over the interpolant. Finally, algorithmic aspects are discussed and the approach is illustrated on numerical examples.
JMLR Optimization & Learning Theory · 发表于 2025
Chuan He, Zhanwang Deng
摘要
Conic optimization plays a crucial role in many machine learning (ML) problems. However, practical algorithms for conic constrained ML problems with large datasets are often limited to specific use cases, as stochastic algorithms for general conic optimization remain underdeveloped. To fill this gap, we introduce a stochastic interior-point method (SIPM) framework for general conic optimization, along with four novel SIPM variants leveraging distinct stochastic gradient estimators. Under mild assumptions, we establish the iteration complexity of our proposed SIPMs, which, up to a polylogarithmic factor, matches the best-known results in stochastic unconstrained optimization. Finally, our numerical experiments on robust linear regression, multi-task relationship learning, and clustering data streams demonstrate the effectiveness and efficiency of our approach.
JMLR Latent Variables & Mixtures · 发表于 2025
Luciano Vinas, Arash A. Amini
摘要
We investigate a class of recovery problems for which observations are a noisy combination of continuous and step functions. These problems can be seen as non-injective instances of non-linear ICA with direct applications to image decontamination for magnetic resonance imaging. Alternately, the problem can be viewed as clustering in the presence of structured (smooth) contaminant. We show that a global topological property (graph connectivity) interacts with a local property (the degree of smoothness of the continuous component) to determine conditions under which the components are identifiable. Additionally, a practical estimation algorithm is provided for the case when the contaminant lies in a reproducing kernel Hilbert space of continuous functions. Algorithm effectiveness is demonstrated through a series of simulations and real-world studies.
JMLR Optimization & Learning Theory Bayesian Methods · 发表于 2025
George Stamatescu
摘要
A statistical field theory is introduced for finite state and action Markov decision processes with unknown parameters, in a Bayesian setting. The Bellman equation, for policy evaluation and the optimal value function in finite and discounted infinite horizon problems, is studied as a disordered interacting dynamical system. The Markov decision process transition probabilities and mean-rewards are interpreted as quenched random variables and the value functions, or the iterates of the Bellman equation, are deterministic variables that evolve dynamically. The posterior over value functions is then equivalent to the quenched average of Fourier inverse of the Martin-Siggia-Rose-De Dominicis-Janssen generating function. The formalism enables the use of methods from field theory to compute posterior moments of value functions. The paper presents two such methods, corresponding to two distinct asymptotic limits. First, the classical approximation is applied, corresponding to the asymptotic data limit. This approximation recovers so-called plug-in estimators for the mean of the value functions. Second, a dynamic mean field theory is derived, showing that under certain assumptions the state-action values are statistically independent across state-action pairs in the asymptotic state space limit. The state-action value statistics can be computed from a set of self-consistent mean field equations, which we call dynamic mean field programming (DMFP). Collectively, the results provide analytic insight into the structure of model uncertainty in Markov decision processes, and pave the way toward more advanced field theoretic techniques and applications to planning and reinforcement learning problems.
JMLR Networks & Graphs · 发表于 2025
Dingbo Wu, Fangzheng Xie
摘要
Spectral estimators have been broadly applied to statistical network analysis, but they do not incorporate the likelihood information of the network sampling model. This paper proposes a novel surrogate likelihood function for statistical inference of a class of popular network models referred to as random dot product graphs. In contrast to the structurally complicated exact likelihood function, the surrogate likelihood function has a separable structure and is log-concave yet approximates the exact likelihood function well. From the frequentist perspective, we study the maximum surrogate likelihood estimator and establish the accompanying theory. We show its existence, uniqueness, large sample properties, and that it improves upon the baseline spectral estimator with a smaller sum of squared errors. Furthermore, we derive the second-order bias of the proposed estimator and gain insight into why it outperforms some of the existing estimators. A computationally convenient stochastic gradient descent algorithm is designed to find the maximum surrogate likelihood estimator in practice. From the Bayesian perspective, we establish the Bernstein--von Mises theorem of the posterior distribution with the surrogate likelihood function and show that the resulting credible sets have the correct frequentist coverage. The empirical performance of the proposed surrogate-likelihood-based methods is validated through the analyses of simulation examples and two real-world data sets.
JMLR Optimization & Learning Theory · 发表于 2025
Sen Na, Michael Mahoney
摘要
We consider online statistical inference of constrained stochastic nonlinear optimization problems. We apply the Stochastic Sequential Quadratic Programming (StoSQP) method to solve these problems, which can be regarded as applying second-order Newton's method to the Karush-Kuhn-Tucker (KKT) conditions. In each iteration, the StoSQP method computes the Newton direction by solving a quadratic program, and then selects a proper adaptive stepsize $\bar{\alpha}_t$ to update the primal-dual iterate. To reduce dominant computational cost of the method, we inexactly solve the quadratic program in each iteration by employing an iterative sketching solver. Notably, the approximation error of the sketching solver need not vanish as iterations proceed, meaning that the per-iteration computational cost does not blow up. For the above StoSQP method, we show that under mild assumptions, the rescaled primal-dual sequence $1/\sqrt{\bar{\alpha}_t}\cdot (x_t -x^\star, \lambda_t - \lambda^\star)$ converges to a mean-zero Gaussian distribution with a nontrivial covariance matrix depending on the underlying sketching distribution. To perform inference in practice, we also analyze a plug-in covariance matrix estimator. We illustrate the asymptotic normality result of the method both on benchmark nonlinear problems in CUTEst test set and on linearly/nonlinearly constrained regression problems.
JMLR Machine Learning Networks & Graphs · 发表于 2025
A. Martina Neuman, Dominik Dold, Philipp Christian Petersen
摘要
We study the learning problem associated with spiking neural networks. Specifically, we focus on spiking neural networks composed of simple spiking neurons having only positive synaptic weights, equipped with an affine encoder and decoder; we refer to these as affine spiking neural networks. These neural networks are shown to depend continuously on their parameters, which facilitates classical covering number-based generalization statements and supports stable gradient-based training. We demonstrate that the positivity of the weights enables a wide range of expressivity results, including rate-optimal approximation of smooth functions and dimension-independent approximation of Barron regular functions. In particular, we show in theory and simulations that affine spiking neural networks are capable of approximating shallow ReLU neural networks. Furthermore, we apply these affine spiking neural networks to standard machine learning benchmarks and reach competitive results (with respect to deep feedforward networks). Finally, we observe that from a generalization perspective, contrary to feedforward neural networks or previous results for general spiking neural networks, the depth has little to no adverse effect on theoretical guarantees for the generalization capabilities.
JMLR Other · 发表于 2025
Chengli Tan, Jiangshe Zhang, Junmin Liu, Yicheng Wang, Yunda Hao
摘要
Recently, sharpness-aware minimization (SAM) has attracted much attention because of its surprising effectiveness in improving generalization performance. However, compared to stochastic gradient descent (SGD), it is more prone to getting stuck at the saddle points, which as a result may lead to performance degradation. To address this issue, we propose a simple renormalization strategy, dubbed Stable SAM (SSAM), so that the gradient norm of the descent step maintains the same as that of the ascent step. Our strategy is easy to implement and flexible enough to integrate with SAM and its variants, almost at no computational cost. With elementary tools from convex optimization and learning theory, we also conduct a theoretical analysis of sharpness-aware training, revealing that compared to SGD, the effectiveness of SAM is only assured in a limited regime of learning rate. In contrast, we show how SSAM extends this regime of learning rate and then it can consistently perform better than SAM with the minor modification. Finally, we demonstrate the improved performance of SSAM on several representative data sets and tasks.
JMLR High-Dimensional Nonparametric & Semiparametric Genetics & Omics · 发表于 2025
Hee Cheol Chung, Yang Ni, Irina Gaynanova
摘要
Sequencing-based technologies provide an abundance of high-dimensional biological data sets with highly skewed and zero-inflated measurements. Despite the computational efficiency and high interpretability offered by linear classification methods, the violation of underlying distribution assumptions, driven by high skewness and zero inflation, results in invalid classification rules and interpretations. Furthermore, existing data transformation methods addressing these violations introduce ambiguity, rendering the final model and classification performance contingent on the specific transformation employed. To tackle these challenges, we propose a novel semiparametric framework for discriminant analysis based on the truncated latent Gaussian copula model. This model accommodates skewness and zero inflation, and its estimation procedure ensures robustness against data transformations. To facilitate model interpretability, we incorporate $\ell_1$ sparsity regularization and establish the consistency of the classification directions in high-dimensional settings. We validate our approach using human gut microbiome, breast cancer microRNA, and single-cell RNA sequencing data, highlighting its superior classification accuracy and robustness to data transformations.
JMLR High-Dimensional · 发表于 2025
Penghe Zhang, Naihua Xiu, Hou-Duo Qi
摘要
The hard-margin loss function has been at the core of the support vector machine research from the very beginning due to its generalization capability. On the other hand, the cardinality constraint has been widely used for feature selection, leading to sparse solutions. This paper studies the sparse SVM with the hard-margin loss that integrates the virtues of both worlds, resulting in one of the most challenging models to solve. We cast the problem as a composite optimization with the cardinality constraint. We characterize its local minimizers in terms of pseudo KKT point that well captures the combinatorial structure of the problem, and investigate a sharper P-stationary point with a concise representation for algorithm design. We further develop an inexact proximal augmented Lagrangian method (iPAL). The different parts of the inexactness measurements from the {\rm P}-stationarity are controlled at different scales in a way that the generated sequence converges both globally and at a linear rate. To make iPAL practically efficient, we propose a gradient-Newton method in a subspace for the iPAL subproblem. This is accomplished by detecting active samples and features with the help of the proximal operator of the hard margin loss and the projection of the cardinality constraint. Extensive numerical results on both simulated and real data sets demonstrate that the proposed method is fast, produces sparse solution of high accuracy, and can lead to effective reduction on active samples and features when compared with several leading solvers.
Econometrica Econometrics & Economics · 发表于 2025
R. Andrew Butters, Jackson Dorsey, Gautam Gowrisankaran
摘要
Renewable energy and battery storage are seen as complementary technologies that can together facilitate reductions in carbon emissions. We develop and estimate a framework to calculate the equilibrium effects of large‐scale battery storage. Using data from California, we find that the first storage unit breaks even by 2024 without subsidies when the renewable energy share reaches 50%. Equilibrium effects are important: the first 5000 MWh of storage capacity would reduce wholesale electricity prices by 5.6%, but an increase from 25,000 to 50,000 MWh would only reduce these prices by 2.6%. Large‐scale batteries will reduce revenues to both dispatchable generators and renewable energy sources. The equilibrium effects lead battery adoption to be virtually non‐existent until 2030, without a storage mandate or subsidy. A 30% capital cost subsidy—such as the one in the U.S. Inflation Reduction Act—achieves 5000 MWh of battery capacity by 2024, similar to the level required under California's storage mandate.
JMLR Optimization & Learning Theory · 发表于 2025
Clément Bonet, Lucas Drumetz, Nicolas Courty
摘要
While many Machine Learning methods have been developed or transposed on Riemannian manifolds to tackle data with known non-Euclidean geometry, Optimal Transport (OT) methods on such spaces have not received much attention. The main OT tool on these spaces is the Wasserstein distance, which suffers from a heavy computational burden. On Euclidean spaces, a popular alternative is the Sliced-Wasserstein distance, which leverages a closed-form solution of the Wasserstein distance in one dimension, but which is not readily available on manifolds. In this work, we derive general constructions of Sliced-Wasserstein distances on Cartan-Hadamard manifolds, Riemannian manifolds with non-positive curvature, which include among others Hyperbolic spaces or the space of Symmetric Positive Definite matrices. Then, we propose different applications such as classification of documents with a suitably learned ground cost on a manifold, and data set comparison on a product manifold. Additionally, we derive non-parametric schemes to minimize these new distances by approximating their Wasserstein gradient flows.
JMLR High-Dimensional Bayesian Methods Clinical Trials · 发表于 2025
Rajarshi Guhaniyogi, Aaron Wolfe Scheffler
摘要
Bayesian computation of high-dimensional linear regression models with popular Gaussian scale mixture prior distributions using Markov Chain Monte Carlo (MCMC) or its variants can be extremely slow or completely prohibitive due to the heavy computational cost that grows in the order of $p^3$, with $p$ as the number of features. Although a few recently developed algorithms allow computational efficiency in presence of a small to moderately large sample size, the computational issues are considerably less explored when sample size $n$ is also large, except for a few recent articles. In this article we propose a sketching approach to compress the $n$ original samples by a random linear transformation to $m<<n$ samples in $p$ dimensions, and compute Bayesian regression with Gaussian scale mixture prior distributions with the randomly compressed response vector and feature matrix. Our proposed approach yields computational complexity growing in the cubic order of $m$. Our detailed empirical investigation with the Horseshoe prior from the class of Gaussian scale mixture priors shows closely similar inference and a considerable reduction in per iteration computation time of the proposed approach compared to the regression with the full sample. One notable contribution of this article is to derive posterior contraction rate for high-dimensional feature coefficient with a general class of shrinkage priors on the coefficients under data compression/sketching. In particular, we characterize the dimension of the compressed response vector $m$ as a function of the sample size, number of features and sparsity in the regression to guarantee accurate estimation of feature coefficients asymptotically, even after data sketching.
JMLR High-Dimensional Optimization & Learning Theory · 发表于 2025
Peng Chen, Jin Zhu, Junxian Zhu, Xueqin Wang
摘要
We consider the probabilistic simplex-constrained sparse recovery problem. The commonly used Lasso-type penalty for promoting sparsity is ineffective in this context since it is a constant within the simplex. Despite this challenge, fortunately, simplex constraint itself brings a self-regularization property, i.e., the empirical risk minimizer without any sparsity-promoting procedure obtains the usual Lasso-type estimation error. Moreover, we analyze the iterates of a projected gradient descent method and show its convergence to the ground truth sparse solution in the geometric rate until a satisfied statistical precision is attained. Although the estimation error is statistically optimal, the resulting solution is usually more dense than the sparse ground truth. To further sparsify the iterates, we propose a method called PERMITS via embedding a tail screening procedure, i.e., identifying negligible components and discarding them during iterations, into the projected gradient descent method. Furthermore, we combine tail screening and the special information criterion to balance the trade-off between fitness and complexity. Theoretically, the proposed PERMITS method can exactly recover the ground truth support set under mild conditions and thus obtain the oracle property. We demonstrate the statistical and computational efficiency of PERMITS with both synthetic and real data. The implementation of the proposed method can be found in https://github.com/abess-team/PERMITS.
JMLR Meta-Analysis & Data Integration · 发表于 2025
Yipeng Li, Xinchen Lyu
摘要
There are two paradigms in Federated Learning (FL): parallel FL (PFL), where models are trained in a parallel manner across clients, and sequential FL (SFL), where models are trained in a sequential manner across clients. Specifically, in PFL, clients perform local updates independently and send the updated model parameters to a global server for aggregation; in SFL, one client starts its local updates only after receiving the model parameters from the previous client in the sequence. In contrast to that of PFL, the convergence theory of SFL on heterogeneous data is still lacking. To resolve the theoretical dilemma of SFL, we establish sharp convergence guarantees for SFL on heterogeneous data with both upper and lower bounds. Specifically, we derive the upper bounds for the strongly convex, general convex and non-convex objective functions, and construct the matching lower bounds for the strongly convex and general convex objective functions. Then, we compare the upper bounds of SFL with those of PFL, showing that SFL outperforms PFL on heterogeneous data (at least, when the level of heterogeneity is relatively high). Experimental results validate the counterintuitive theoretical finding.
JMLR Testing & Uncertainty Meta-Analysis & Data Integration High-Dimensional · 发表于 2025
Sifan Liu, Snigdha Panigrahi
摘要
When data are distributed across multiple sites or machines rather than centralized in one location, researchers face the challenge of extracting meaningful information without directly sharing individual data points. While there are many distributed methods for point estimation using sparse regression, few options are available for estimating uncertainties or conducting hypothesis tests based on the estimated sparsity. In this paper, we introduce a procedure for performing selective inference with distributed data. We consider a scenario where each local machine solves a lasso problem and communicates the selected predictors to a central machine. The central machine then aggregates these selected predictors to form a generalized linear model (GLM). Our goal is to provide valid inference for the selected GLM while reusing data that have been used in the model selection process. Our proposed procedure only requires low-dimensional summary statistics from local machines, thus keeping communication costs low and preserving the privacy of individual data sets. Furthermore, this procedure can be applied in scenarios where model selection is repeatedly conducted on randomly subsampled data sets, addressing the p-value lottery problem linked with model selection. We demonstrate the effectiveness of our approach through simulations and an analysis of a medical data set on ICU admissions.
Econometrica Econometrics & Economics · 发表于 2025
Abhijit Banerjee, Arun G. Chandrasekhar, Suresh Dalpath, Esther Duflo, John Floretta, Matthew O. Jackson, Harini Kannan, Francine Loza, Anirudh Sankar, Anna Schrimpf, Maheshwor Shrestha
摘要
Policymakers often choose a policy bundle that is a combination of different interventions in different dosages. We develop a new technique— treatment variant aggregation (TVA)—to select a policy from a large factorial design. TVA pools together policy variants that are not meaningfully different and prunes those deemed ineffective. This allows us to restrict attention to aggregated policy variants, consistently estimate their effects on the outcome, and estimate the best policy effect adjusting for the winner's curse. We apply TVA to a large randomized controlled trial that tests interventions to stimulate demand for immunization in Haryana, India. The policies under consideration include reminders, incentives, and local ambassadors for community mobilization. Cross‐randomizing these interventions, with different dosages or types of each intervention, yields 75 combinations. The policy with the largest impact (which combines incentives, ambassadors who are information hubs, and reminders) increases the number of immunizations by 44% relative to the status quo. The most cost‐effective policy (information hubs, ambassadors, and SMS reminders, but no incentives) increases the number of immunizations per dollar by 9.1% relative to the status quo.
Econometrica Networks & Graphs · 发表于 2025
Evan Sadler
摘要
I propose a method for selecting seeds to maximize contagion. First, fit a random graph model using a coarse categorization of individuals. Next, compute a seed multiplier for each category—this is the average number of new infections a seed generates. Finally, seed the category with the highest multiplier. Relative to the most common methods, my approach requires far less granular data, and it consumes less computing power—the problem scales with the number of categories, not the number of individuals. I validate the methodology through simulations using real network data.
JMLR Other · 发表于 2025
Sachin Garg, Albert S. Berahas, Michał Dereziński
摘要
We show that, for finite-sum minimization problems, incorporating partial second-order information of the objective function can dramatically improve the robustness to mini-batch size of variance-reduced stochastic gradient methods, making them more scalable while retaining their benefits over traditional Newton-type approaches. We demonstrate this phenomenon on a prototypical stochastic second-order algorithm, called Mini-Batch Stochastic Variance-Reduced Newton ($\texttt{Mb-SVRN}$), which combines variance-reduced gradient estimates with access to an approximate Hessian oracle. In particular, we show that when the data size $n$ is sufficiently large, i.e., $n\gg \alpha^2\kappa$, where $\kappa$ is the condition number and $\alpha$ is the Hessian approximation factor, then $\texttt{Mb-SVRN}$ achieves a fast linear convergence rate that is independent of the gradient mini-batch size $b$, as long $b$ is in the range between $1$ and $b_{\max}=O(n/(\alpha \log n))$. Only after increasing the mini-batch size past this critical point $b_{\max}$, the method begins to transition into a standard Newton-type algorithm which is much more sensitive to the Hessian approximation quality. We verify this empirically on benchmark tasks showing that, after tuning the step size, the convergence rate of $\texttt{Mb-SVRN}$ remains fast for a wide range of mini-batch sizes, and the dependence of the phase transition point $b_{\max}$ on the Hessian approximation factor $\alpha$ agrees with our theory.
Econometrica Econometrics & Economics · 发表于 2025
Giulia Brancaccio, Karam Kang
摘要
This paper shows that product design shapes search frictions and that intermediaries leverage this channel to increase their rents in the context of the U.S. municipal bond market. The majority of bonds are designed via negotiation between a local government and its underwriter. They are then traded in a decentralized market, where the underwriter often also acts as an intermediary. Exploiting variations in state regulations that limit government officials' conflicts of interest, we provide evidence that the underwriter benefits from designing and trading complex bonds, which induces an increase in search frictions. Interestingly, a simpler bond may not necessarily benefit the government, as bond complexity affords flexibility in debt repayment. Motivated by these findings, we build and estimate a model of bond origination and trading to quantify the welfare implications of a policy mandating bond standardization.
JMLR Causal Inference Latent Variables & Mixtures · 发表于 2025
Burak Varici, Emre Acartürk, Karthikeyan Shanmugam, Abhishek Kumar, Ali Tajer
摘要
This paper addresses intervention-based causal representation learning (CRL) under a general nonparametric latent causal model and an unknown transformation that maps the latent variables to the observed variables. Linear and general transformations are investigated. The paper addresses both the identifiability and achievability aspects. Identifiability refers to determining algorithm-agnostic conditions that ensure the recovery of the true latent causal variables and the underlying latent causal graph. Achievability refers to the algorithmic aspects and addresses designing algorithms that achieve identifiability guarantees. By drawing novel connections between score functions (i.e., the gradients of the logarithm of density functions) and CRL, this paper designs a score-based class of algorithms that ensures both identifiability and achievability. First, the paper focuses on linear transformations and shows that one stochastic hard intervention per node suffices to guarantee identifiability. It also provides partial identifiability guarantees for soft interventions, including identifiability up to mixing with parents for general causal models and perfect recovery of the latent graph for sufficiently nonlinear causal models. Secondly, it focuses on general transformations and demonstrates that two stochastic hard interventions per node are sufficient for identifiability. This is achieved by defining a differentiable loss function whose global optima ensure identifiability for general CRL. Notably, one does not need to know which pair of interventional environments has the same node intervened. Finally, the theoretical results are empirically validated via experiments on structured synthetic data and image data.
JMLR Machine Learning · 发表于 2025
Jae Hyun Lim, Nikola B. Kovachki, Ricardo Baptista, Christopher Beckham, Kamyar Azizzadenesheli, Jean Kossaifi, Vikram Voleti, Jiaming Song, Karsten Kreis, Jan Kautz, Christopher Pal, Arash Vahdat, Anima Anandkumar
摘要
Diffusion models have recently emerged as a powerful framework for generative modeling. They consist of a forward process that perturbs input data with Gaussian white noise and a reverse process that learns a score function to generate samples by denoising. Despite their tremendous success, they are mostly formulated on finite-dimensional spaces, e.g., Euclidean, limiting their applications to many domains where the data has a functional form, such as in scientific computing and 3D geometric data analysis. This work introduces a mathematically rigorous framework called Denoising Diffusion Operators (DDOs) for training diffusion models in function space. In DDOs, the forward process perturbs input functions gradually using a Gaussian process. The generative process is formulated by a function-valued annealed Langevin dynamic. Our approach requires an appropriate notion of the score for the perturbed data distribution, which we obtain by generalizing denoising score matching to function spaces that can be infinite-dimensional. We show that the corresponding discretized algorithm generates accurate samples at a fixed cost independent of the data resolution. We theoretically and numerically verify the applicability of our approach on a set of function-valued problems, including generating solutions to the Navier-Stokes equation viewed as the push-forward distribution of forcings from a Gaussian Random Field (GRF), as well as volcano InSAR and MNIST-SDF.
JMLR Optimization & Learning Theory · 发表于 2025
Céline Comte, Matthieu Jonckheere, Jaron Sanders, Albert Senen-Cerda
摘要
In this paper, we introduce a policy-gradient method for model-based reinforcement learning (RL) that exploits a type of stationary distributions commonly obtained from Markov decision processes (MDPs) in stochastic networks, queueing systems, and statistical mechanics. Specifically, when the stationary distribution of the MDP belongs to an exponential family that is parametrized by policy parameters, we can improve existing policy gradient methods for average-reward RL. Our key identification is a family of gradient estimators, called score-aware gradient estimators (SAGEs), that enable policy gradient estimation without relying on value-function estimation in the aforementioned setting. We show that SAGE-based policy-gradient locally converges, and we obtain its regret. This includes cases when the state space of the MDP is countable and unstable policies can exist. Under appropriate assumptions such as starting sufficiently close to a maximizer and the existence of a local Lyapunov function, the policy under SAGE-based stochastic gradient ascent has an overwhelming probability of converging to the associated optimal policy. Furthermore, we conduct a numerical comparison between a SAGE-based policy-gradient method and an actor-critic method on several examples inspired from stochastic networks, queueing systems, and models derived from statistical physics. Our results demonstrate that a SAGE-based method finds close-to-optimal policies faster than an actor-critic method.
JMLR Other · 发表于 2025
Pierre Marion, Adeline Fermanian, Gérard Biau, Jean-Philippe Vert
摘要
Deep ResNets are recognized for achieving state-of-the-art results in complex machine learning tasks. However, the remarkable performance of these architectures relies on a training procedure that needs to be carefully crafted to avoid vanishing or exploding gradients, particularly as the depth $L$ increases. No consensus has been reached on how to mitigate this issue, although a widely discussed strategy consists in scaling the output of each layer by a factor $\alpha_L$. We show in a probabilistic setting that with standard i.i.d. initializations, the only non-trivial dynamics is for $\alpha_L = \frac{1}{\sqrt{L}}$---other choices lead either to explosion or to identity mapping. This scaling factor corresponds in the continuous-time limit to a neural stochastic differential equation, contrarily to a widespread interpretation that deep ResNets are discretizations of neural ordinary differential equations. By contrast, in the latter regime, stability is obtained with specific correlated initializations and $\alpha_L = \frac{1}{L}$. Our analysis suggests a strong interplay between scaling and regularity of the weights as a function of the layer index. Finally, in a series of experiments, we exhibit a continuous range of regimes driven by these two parameters, which jointly impact performance before and after training.
JMLR Other · 发表于 2025
Niklas Muennighoff, Alexander M. Rush, Boaz Barak, Teven Le Scao, Aleksandra Piktus, Nouamane Tazi, Sampo Pyysalo, Thomas Wolf, Colin Raffel
摘要
The current trend of scaling language models involves increasing both parameter count and training data set size. Extrapolating this trend suggests that training data set size may soon be limited by the amount of text data available on the internet. Motivated by this limit, we investigate scaling language models in data-constrained regimes. Specifically, we run a large set of experiments varying the extent of data repetition and compute budget, ranging up to 900 billion training tokens and 9 billion parameter models. We find that with constrained data for a fixed compute budget, training with up to 4 epochs of repeated data yields negligible changes to loss compared to having unique data. However, with more repetition, the value of adding compute eventually decays to zero. We propose and empirically validate a scaling law for compute optimality that accounts for the decreasing value of repeated tokens and excess parameters. Finally, we experiment with approaches mitigating data scarcity, including augmenting the training data set with code data or removing commonly used filters. Models and data sets from our 400 training runs are freely available at https://github.com/huggingface/datablations.
JMLR Other · 发表于 2025
Tenghui Li, Guoxu Zhou, Xuyang Zhao, Qibin Zhao
摘要
Large language models have demonstrated predictable scaling behaviors with respect to model parameters and training data. This study investigates whether a similar scaling relationship exist for vision-language models with respect to the number of vision tokens. A mathematical framework is developed to characterize a relationship between vision token number and the expected divergence of distance between vision-referencing sequences. The theoretical analysis reveals two distinct scaling regimes: sublinear scaling for less vision tokens and linear scaling for more vision tokens. This aligns with model performance relationships of the form \(S(n) \approx c / n^{\alpha(n)}\), where the scaling exponent relates to the correlation structure between vision token representations. Empirical validations across multiple vision-language benchmarks show that model performance matches the prediction from scaling relationship. The findings contribute to understanding vision token scaling in transformers through a theoretical framework that complements empirical observations.
JMLR Bayesian Methods Spatial & Time Series Econometrics & Economics · 发表于 2025
Deborah Sulem, Vincent Rivoirard, Judith Rousseau
摘要
Hawkes processes are often applied to model dependence and interaction phenomena in multivariate event data sets, such as neuronal spike trains, social interactions, and financial transactions. In the nonparametric setting, learning the temporal dependence structure of Hawkes processes is generally a computationally expensive task, all the more with Bayesian estimation methods. In particular, for multivariate nonlinear Hawkes processes, Monte-Carlo Markov Chain (MCMC) methods used to sample from the posterior distribution do not scale well to the dimension of the process. Recently, efficient algorithms targeting a mean-field variational approximation of the posterior distribution have been proposed, however, these methods do not allow to perform model selection on the graph of interactions of the Hawkes model. In this work, we propose a novel adaptive Bayesian variational method that performs model selection and can estimate a sparse graphical parameter. For the popular sigmoid Hawkes processes, we design a parallel algorithm which is scalable to high-dimensional point processes and large sequences of events. Furthermore, we unify existing variational Bayes approaches under a general nonparametric inference framework, and analyse the asymptotic properties of these methods under easily verifiable conditions on the prior, the variational class, and the nonlinear model. Finally, through an extensive set of numerical simulations, we demonstrate that our method is able to adapt to the dimensionality of the parameter of the Hawkes process, and is partially robust to certain types of model misspecification.
JMLR Other · 发表于 2025
Karthik Bharath, Alexander Lewis, Akash Sharma, Michael V. Tretyakov
摘要
Error bounds are derived for sampling and estimation using a discretization of an intrinsically defined Langevin diffusion with invariant measure $\text{d}\mu_\phi \propto e^{-\phi} \mathrm{dvol}_g $ on a compact Riemannian manifold. Two estimators of linear functionals of $\mu_\phi $ based on the discretized Markov process are considered: a time-averaging estimator based on a single trajectory and an ensemble-averaging estimator based on multiple independent trajectories. Imposing no restrictions beyond a nominal level of smoothness on $\phi$, first-order error bounds, in discretization step size, on the bias and variance/mean-square error of both estimators are derived. The order of error matches the optimal rate in Euclidean and flat spaces, and leads to a first-order bound on distance between the invariant measure $\mu_\phi$ and a stationary measure of the discretized Markov process. This order is preserved even upon using retractions when exponential maps are unavailable in closed form, thus enhancing practicality of the proposed algorithms. Generality of the proof techniques, which exploit links between two partial differential equations and the semigroup of operators corresponding to the Langevin diffusion, renders them amenable for the study of a more general class of sampling algorithms related to the Langevin diffusion. Conditions for extending analysis to the case of non-compact manifolds are discussed. Numerical illustrations with distributions, log-concave and otherwise, on the manifolds of positive and negative curvature elucidate on the derived bounds and demonstrate practical utility of the sampling algorithm.
JMLR Machine Learning Optimization & Learning Theory · 发表于 2025
Amirreza Neshaei Moghaddam, Alex Olshevsky, Bahman Gharesifard
摘要
We provide the first known algorithm that provably achieves $\varepsilon$-optimality within $\widetilde{O}(1/\varepsilon)$ function evaluations for the discounted discrete-time linear quadratic regulator problem with unknown parameters, without relying on two-point gradient estimates. These estimates are known to be unrealistic in many settings, as they depend on using the exact same initialization, which is to be selected randomly, for two different policies. Our results substantially improve upon the existing literature outside the realm of two-point gradient estimates, which either leads to $\widetilde{O}(1/\varepsilon^2)$ rates or heavily relies on stability assumptions.
Econometrica Econometrics & Economics · 发表于 2025
Qingen Gai, Naijia Guo, Bingjing Li, Qinghua Shi, Xiaodong Zhu
摘要
We exploit the implementation of a rural pension policy in China to estimate the average rural‐to‐urban migration cost for workers affected by the policy and the average underlying sectoral productivity difference. Our estimates, based on a large panel data set, reveal significant migration costs and substantial sectoral productivity differences, with sorting playing a minor role in accounting for sectoral labor income gaps. We construct and structurally estimate a general equilibrium household model with endogenous labor supply and migration. The results of this model align with the reduced‐form findings and illustrate how the rural pension policy influences migration, GDP, and welfare through improving within‐household labor allocation. Counterfactual analyses based on the model show that the positive effects of the policy remain even if migration costs were significantly lower, and that scaling up the rural pension policy would lead to even larger improvements in labor allocation, GDP, and welfare.
Econometrica Econometrics & Economics · 发表于 2025
Andrew B. Abel, Stavros Panageas
摘要
Government debt can be rolled over forever without primary surpluses in some stochastic economies, including some economies that are dynamically efficient. In an overlapping‐generations model with constant growth rate, g , of labor‐augmenting productivity, and with shocks to the durability of capital, we show that along a balanced growth path, the maximum sustainable ratio of bonds to capital is attained when the risk‐free interest rate, r f , equals g . Furthermore, this maximal ratio maximizes utility per capita along a balanced growth path and ensures that the economy is dynamically efficient.
JMLR Other · 发表于 2025
YoonHaeng Hur, Yuehaw Khoo
摘要
Computational difficulty of quadratic matching and the Gromov-Wasserstein distance has led to various approximation and relaxation schemes. One of such methods, relying on the notion of distance profiles, has been widely used in practice, but its theoretical understanding is limited. By delving into the statistical complexity of the previously proposed method based on distance profiles, we show that it suffers from the curse of dimensionality unless we make certain assumptions on the underlying metric measure spaces. Building on this insight, we propose and analyze a modified matching procedure that can be used to robustly match points under a certain probabilistic setting. We demonstrate the performance of the proposed methods using simulations and real data applications to complement the theoretical findings. As a result, we contribute to the literature by providing theoretical underpinnings of the matching procedures based on distance invariants like distance profiles, which have been widely used in practice but rarely analyzed theoretically.
Econometrica Optimization & Learning Theory Bayesian Methods · 发表于 2025
Karun Adusumilli
摘要
We provide a decision‐theoretic analysis of bandit experiments under local asymptotics. Working within the framework of diffusion processes, we define suitable notions of asymptotic Bayes and minimax risk for these experiments. For normally distributed rewards, the minimal Bayes risk can be characterized as the solution to a second‐order partial differential equation (PDE). Using a limit of experiments approach, we show that this PDE characterization also holds asymptotically under both parametric and non‐parametric distributions of the rewards. The approach further describes the state variables it is asymptotically sufficient to restrict attention to, and thereby suggests a practical strategy for dimension reduction. The PDEs characterizing minimal Bayes risk can be solved efficiently using sparse matrix routines or Monte Carlo methods. We derive the optimal Bayes and minimax policies from their numerical solutions. These optimal policies substantially dominate existing methods such as Thompson sampling; the risk of the latter is often twice as high.
JMLR Optimization & Learning Theory · 发表于 2025
Jiaxiang Li, Shiqian Ma
摘要
In this work, we consider the bilevel optimization problem on Riemannian manifolds. We inspect the calculation of the hypergradient of such problems on general manifolds and thus enable the utilization of gradient-based algorithms to solve such problems. The calculation of the hypergradient requires utilizing the notion of Riemannian cross-derivative and we inspect the properties and the numerical calculations of Riemannian cross-derivatives. Algorithms in both deterministic and stochastic settings, named respectively RieBO and RieSBO, are proposed that include the existing Euclidean bilevel optimization algorithms as special cases. Numerical experiments on robust optimization on Riemannian manifolds are presented to show the applicability and efficiency of the proposed methods.
JMLR Machine Learning Optimization & Learning Theory · 发表于 2025
Xuefeng Gao, Jiale Zha, Xun Yu Zhou
摘要
We propose a new reinforcement learning (RL) formulation for training continuous-time score-based diffusion models for generative AI to generate samples that maximize reward functions while keeping the generated distributions close to the unknown target data distributions. Different from most existing studies, ours does not involve any pretrained model for the unknown score functions of the noise-perturbed data distributions, nor does it attempt to learn the score functions. Instead, we formulate the problem as entropy-regularized continuous-time RL and show that the optimal stochastic policy has a Gaussian distribution with a known covariance matrix. Based on this result, we parameterize the mean of Gaussian policies and develop an actor--critic type (little) $q$-learning algorithm to solve the RL problem. A key ingredient in our algorithm design is to obtain noisy observations from the unknown score function via a ratio estimator. Our formulation can also be adapted to solve pure score-matching and fine-tuning pretrained models. Numerically, we show the effectiveness of our approach by comparing its performance with two state-of-the-art RL methods that fine-tune pretrained models on several generative tasks including high-dimensional image generations. Finally, we discuss extensions of our RL formulation to probability flow ODE implementation of diffusion models and to conditional diffusion models.
JMLR Other · 发表于 2025
Tao Sun, Xinwang Liu, Kun Yuan
摘要
Gradient clipping has long been considered essential for ensuring the convergence of Stochastic Gradient Descent (SGD) in the presence of heavy-tailed gradient noise. In this paper, we revisit this belief and explore whether gradient normalization can serve as an effective alternative or complement. We prove that, under individual smoothness assumptions, gradient normalization alone is sufficient to guarantee convergence of the nonconvex SGD. Moreover, when combined with clipping, it yields far better rates of convergence under more challenging noise distributions. We provide a unifying theory describing normalization-only, clipping-only, and combined approaches. Moving forward, we investigate existing variance-reduced algorithms, establishing that, in such a setting, normalization alone is sufficient for convergence. Finally, we present an accelerated variant that under second-order smoothness improves convergence. Our results provide theoretical insights and practical guidance for using normalization and clipping in nonconvex optimization with heavy-tailed noise.
JMLR High-Dimensional Nonparametric & Semiparametric · 发表于 2025
Chengde Qian, Guanghui Wang, Changliang Zou
摘要
Changepoint detection typically relies on a grid-search strategy for optimal data segmentation. When model fitting itself is expensive, repeatedly fitting a model on every candidate segment dominates the computation. Existing approaches mitigate this by pruning the grid, thus reducing the number of segments (and model fits). We propose Reliever, which instead cuts the number of model fits directly and nests seamlessly within standard grid-search routines. Reliever fits a small, deterministic collection of proxy models and reuses them wherever they apply, making it compatible with a wide range of existing algorithms. For high-dimensional regression with changepoints, coupling Reliever with an optimal grid-search method yields changepoint and coefficient estimators that are rate-optimal up to a logarithmic factor. Extensive numerical experiments demonstrate that Reliever rapidly and accurately detects changepoints across a wide range of high-dimensional and nonparametric models.
JMLR Optimization & Learning Theory Bayesian Methods · 发表于 2025
Sébastien J. Petit, Julien Bect, Emmanuel Vazquez
摘要
This work presents a new procedure for obtaining predictive distributions in the context of Gaussian process (GP) modeling, with a relaxation of the interpolation constraints outside ranges of interest: the mean of the predictive distribution no longer necessarily interpolates the observed values when they are outside ranges of interest, but is simply constrained to remain outside. This method called relaxed Gaussian process (reGP) interpolation provides better predictive distributions in ranges of interest, especially in cases where a stationarity assumption for the GP model is not appropriate. It can be viewed as a goal-oriented method and becomes particularly interesting in Bayesian optimization, for example, for the minimization of an objective function, where good predictive distributions for low function values are important. When the expected improvement criterion and reGP are used for sequentially choosing evaluation points, the convergence of the resulting optimization algorithm is theoretically guaranteed (provided that the function to be optimized lies in the reproducing kernel Hilbert space attached to the known covariance of the underlying Gaussian process). Experiments indicate that using reGP instead of stationary GP models in Bayesian optimization is beneficial.
JMLR Machine Learning · 发表于 2025
Wei Zhang, Jr-Shin Li
摘要
Interest in reinforcement learning (RL) for large-scale systems, comprising extensive populations of intelligent agents interacting with heterogeneous environments, has surged significantly across diverse scientific domains in recent years. However, the large-scale nature of these systems often leads to high computational costs or reduced performance for most state-of-the-art RL techniques. To address these challenges, we propose a novel RL architecture and derive effective algorithms to learn optimal policies for arbitrarily large systems of agents. In our formulation, we model such systems as parameterized control systems defined on an infinite-dimensional function space. We then develop a moment kernel transform that maps the parameterized system and the value function into a reproducing kernel Hilbert space. This transformation generates a sequence of finite-dimensional moment representations for the RL problem, organized into a filtrated structure. Leveraging this RL filtration, we develop a hierarchical algorithm for learning optimal policies for the infinite-dimensional parameterized system. To enhance the algorithm's efficiency, we incorporate early stopping at each hierarchy, demonstrating the fast convergence property of the algorithm through the construction of a convergent spectral sequence. The performance and efficiency of the proposed algorithm are validated using practical examples in engineering and quantum systems.