Searcharxiv⌕ Search

arXiv subjects

Wenjing Liao

Publications and source records attributed to Wenjing Liao.

At least 19 recordsLinked to original sources

A Geometric Understanding of Denoising Autoencoder with Applications to Image Denoising

Denoising autoencoders have shown strong empirical performance in recovering clean data from their noisy observations, yet their underlying statistical behavior remains underexplored. In this paper, we study denoising autoencoders under the manifold hypothesis, assuming that clean data are supported on low-dimensional smooth manifolds embedded in a high-dimensional ambient space, while noisy observations are generated by isotropic Gaussian noise. We establish generalization guarantees in three geometric scenarios: a single manifold, a pairwise separated finite union of manifolds, and a finite union of intersecting manifolds. Using local chart representations, partitions of unity, and neural network approximation, we derive upper bounds for the expected squared reconstruction error for denoising autoencoder. In the single-manifold and separated-union scenarios, the error consists of a sample-dependent statistical term of order $n^{-2/(d+2)}\log^4 n$, a residual term of order $dσ^2/D$ caused by tangential noise, and an exponentially small tail term associated with failure of the manifold projection. Here $D$ is the data ambient dimension, $d$ is the intrinsic dimension of the clean-data manifold, and $σ^2$ is the expected squared noise magnitude. In the scenario of intersecting manifolds, we further identify an additional ambiguity term determined by the probability mass near the intersection neighborhood. Numerical experiments on synthetic manifolds and real image datasets support the predicted sample-size scaling law and illustrate the additional reconstruction ambiguity near manifold intersections.

stat.ML↗

Learning Theory of Transformers: Local-to-Global Approximation via Softmax Partition of Unity

This paper investigates the learning theory of Transformer networks for regression tasks on the compact Euclidean domain $[0,1]^d$ and $d$-dimensional compact Riemannian manifolds. We propose a novel constructive approximation framework for Transformers that builds local approximations of the target function and aggregates them into a global approximation via softmax partition of unity (POU). This approach leverages the attention mechanism to achieve spatial localization through affine transformations of the input. The softmax activation plays a crucial role in aggregating local approximations to a global output. From an approximation perspective, we prove that a dense Transformer equipped with only two encoder blocks and standard single-hidden-layer point-wise feed-forward networks can achieve a uniform $\varepsilon$-approximation error for $α$-Hölder continuous functions with $α\in (0,1]$ using $\mathcal{O}(\varepsilon^{-d/α})$ total parameters. Building upon this approximation guarantee, we establish a near minimax-optimal generalization error bound of order $\mathcal{O}\big(n^{-\frac{2α}{2α+d}} \log n\big)$ for the empirical risk minimizer, where $n$ is the training data size. The Transformer architecture studied in this paper uses native softmax attention and sinusoidal positional encodings, with a fixed depth of two encoder blocks. Finally, numerical experiments reveal spatially localized, Gaussian-like attention weights that emerge during training, resembling our Softmax POU weights.

stat.ML↗

Transformers as Adaptive Kernel Learners across Tasks for In-Context Learning

Transformers have emerged as a dominant architecture in modern machine learning, with in-context learning (ICL) enabling models to adapt to new tasks by conditioning on prompts, without any parameter updates. To understand the attention mechanism in transformer blocks, a growing line of theoretical work interprets ICL through the lens of classical statistical estimators. Recent works have established a connection between the attention mechanism and kernel methods, while the kernel function is usually assumed to be fixed. In practice, the query and key matrices in the attention mechanism are trained, which allows the attention mechanism to serve as an adaptive kernel method. In this work, we demonstrate that transformers can act as adaptive anisotropic kernel learners. Specifically, we study ICL for low-rank linear and multi-index nonlinear regressions, and conduct numerical simulations to show that the learned attention scores closely resemble the oracle kernels for these function classes. We then develop a general framework to analyze the generalization error of ICL by connecting transformers to the oracle kernel estimator for each function class. Our theory gives an explicit and constructive implementation of the oracle kernel estimators for these function classes using a wide and shallow transformer architecture, illustrating how the prediction is carried out by the interaction between the query token and the context tokens. Our work complements existing studies by providing numerical and theoretical evidence of how the attention mechanism adapts to anisotropic kernels in both linear and nonlinear settings.

cs.LG↗

Latent Evolution Operator (LEO): State-to-State Forecasting of Parametric PDEs

Solving time-dependent parametric partial differential equations (PDEs) with high-fidelity solvers is expensive in multi-query settings. This motivates reduced-order models in latent space. Existing methods are typically either (i) recurrent latent dynamics models, which require step-by-step integration and accumulate error over long horizons, or (ii) neural operators, which enable direct prediction but are usually trained from initial conditions and do not explicitly enforce generalization from intermediate states. We introduce the Latent Evolution Operator (LEO), a parameterized map that takes a state at an arbitrary time, together with physical parameters and a forecast horizon, and predicts the future state in a single latent-space evaluation. Under standard manifold and regularity assumptions, finite-time evolution operators admit an encoder-propagator-decoder decomposition. We approximate this structure using neural networks trained on state pairs sampled across anchor times and horizons. We evaluate the method on the 1D Burgers' equation, a 2D advection-diffusion system, and the shallow-water equations. The model produces accurate predictions across parameters and time horizons, achieves significant speedups compared to Autoencoder-LSTM baselines, and maintains accuracy when initialized from intermediate states. These results indicate that learning finite-time evolution operators in latent space provides an effective alternative to recurrent latent dynamics models and operator learning methods trained only from initial conditions.

cs.LG↗

Transformer-based In-Context Score Operator Learning for Diffusion Models

Diffusion models typically learn the generative dynamics of a single target distribution or a family of distributions indexed by conditioning variables fixed during training. This limits their ability to adapt to a previously unseen distribution from only a small set of examples. We introduce an in-context solution operator learning framework for diffusion models, in which each target distribution is treated as a task drawn from a meta-distribution and is specified at inference time by an empirical context set. Our method learns an in-context score operator that maps the context samples, a noisy query point, and the diffusion time to the task-dependent score, enabling reverse-diffusion sampling on unseen tasks without retraining or parameter updates. A key observation is that the task-dependent score admits a Gaussian-kernel conditional-mean representation. Its empirical Nadaraya--Watson approximation can be represented exactly by a model in our constructed transformer class, providing a principled foundation for the proposed query-dependent attention architecture. The operator is trained through amortized denoising score matching across task distributions. We establish generalization-error bounds that quantify the effects of the context size, the numbers of training tasks and within-task samples, model complexity, and intrinsic and ambient dimensions. Numerically, we evaluate the method on a range of scientific and visual datasets, including Navier-Stokes, ERA5 climate data, QuickDraw sketches, and Food-101. The results show that the proposed method performs well on relatively simple task families, while its performance degrades as task complexity increases, thereby illustrating both the promise and the limitations of in-context learning for diffusion models.

cs.LG↗

Neural Scaling Laws of Transformer Operator Network

Transformers have emerged as powerful architectures for learning solution operators of physical systems. Empirically the prediction error has been observed to decrease when the data size and model size increase, suggesting neural scaling behavior. Yet a theoretical understanding of such scaling laws for transformer-based operator learning remains limited. In this work, we develop a theoretical framework for characterizing the approximation and generalization errors of transformer-based operator learning. Our analysis builds on a local-to-global approximation principle that is naturally aligned with the softmax attention mechanism and yields discretization-invariant output functions. On approximation theory, we derive a universal approximation error of transformer-based operator learning for Hölder-regular operators. On generalization theory, we establish a power scaling law between the prediction error and the training data size. The rate of convergence represented by the scaling exponent explicitly reflects the dimensions of the input and output domains, the regularity of the underlying functions and operators, and crucially, the intrinsic dimension of the input function class. By exploiting this intrinsic low-dimensional structure, our analysis yields a power-law generalization rate for operator learning, in contrast to the logarithmic-type power-law rates appearing in existing analyses of operator learning with feedforward neural networks. Numerical experiments validate the predicted power-law scaling and confirm that the convergence rate varies systematically with the intrinsic dimension of the input function class.

stat.ML↗

Transformers as Cross-Task Learners: Shared Structure Drives Sample Efficiency in In-Context Learning

Transformers achieve remarkable performance by jointly learning broad families of tasks during pretraining and adapting to unseen tasks from only a short prompt. Yet a rigorous mathematical and statistical understanding of this phenomenon remains limited. This paper aims to study how Transformers exploit shared cross-task structure and how this structure affects the sample complexity of in-context learning (ICL). Specifically, we characterize task-space complexity through covering numbers under a prescribed metric, thereby quantifying the low-dimensional cross-task structure without requiring an explicit parametric representation. The resulting cover provides a set of anchor functions, which we use to introduce a task-identification-and-evaluation procedure: context observations localize an unseen task among the anchor functions, and the response at a query is predicted by aggregating the corresponding anchor function query evaluations. For approximation, we explicitly construct a Transformer with Softmax attention to approximate this procedure. For generalization, we derive an error bound that separates the effects of the number of pretraining tasks and the prompt length. The scaling with respect to the number of pretraining tasks is governed by the intrinsic dimensions of the task space and input domain; once sufficiently many tasks are available, the dependence on the prompt context length becomes dimension-free. To the best of our knowledge, this is the first work to quantify cross-task complexity for general nonlinear task families and explicitly construct a Transformer that exploits their low-dimensional structure to perform ICL. Our theory provides a quantitative explanation of how joint pretraining across related tasks improves in-context generalization.

stat.ML↗

Optimality of Gradient-MUSIC for spectral estimation

We introduce the Gradient-MUSIC algorithm for estimating the unknown frequencies and amplitudes of a nonharmonic signal from noisy time samples. While the classical MUSIC algorithm performs a computationally expensive search over a fine grid, Gradient-MUSIC is significantly more efficient and eliminates the need for discretization over a fine grid by using optimization techniques. It coarsely scans the 1D landscape to find initialization simultaneously for all frequencies followed by parallelizable local refinement via gradient descent. We also analyze its performance when the noise level is sufficiently small and the signal frequencies are separated by at least $8π/m$, where $π/m$ is the standard resolution of this problem. Even though the 1D landscape is nonconvex, we prove a global convergence result for Gradient-MUSIC: coarse scanning provably finds suitable initialization and gradient descent converges at a linear rate. In addition to convergence results, we also upper bound the error between the true signal frequencies and amplitudes with those found by Gradient-MUSIC. For example, if the noise has $\ell^\infty$ norm at most $\varepsilon$, then the frequencies and amplitudes are recovered up to error at most $C\varepsilon/m$ and $C\varepsilon$ respectively for a universal $C>0$, which are minimax optimal in $m$, $\varepsilon$, and number of frequencies. Our theory can also handle stochastic noise with performance guarantees under nonstationary independent Gaussian noise. Our main approach is a comprehensive geometric analysis of the landscape, a perspective that has not been explored before.

cs.IT↗

Generalization Guarantees for Multi-Input Neural Operator Learning in Sobolev Spaces

We develop approximation and generalization error estimates for multi-input neural operators, with the output error measured in Sobolev norms. In contrast to standard operator-learning settings with a single input function, our framework allows multiple input functions defined on possibly different domains, with different dimensions and Sobolev regularities. The derived rates explicitly quantify the contribution of each input space to the final error bound. In particular, in the balanced regime, the approximation and generalization rates are governed by the interaction between the input dimensions, regularities, and Sobolev orders, while the dependence on the model complexity retains a \(\log\log/\log\)-type structure. Our analysis provides a general theoretical framework for multi-input operator learning, including Sobolev training, and is applicable to operator learning problems arising from partial differential equations and scientific computing.

cs.LG↗

Neural Scaling Laws of Deep ReLU and Deep Operator Network: A Theoretical Study

Neural scaling laws play a pivotal role in the performance of deep neural networks and have been observed in a wide range of tasks. However, a complete theoretical framework for understanding these scaling laws remains underdeveloped. In this paper, we explore the neural scaling laws for deep operator networks, which involve learning mappings between function spaces, with a focus on the Chen and Chen style architecture. These approaches, which include the popular Deep Operator Network (DeepONet), approximate the output functions using a linear combination of learnable basis functions and coefficients that depend on the input functions. We establish a theoretical framework to quantify the neural scaling laws by analyzing its approximation and generalization errors. We articulate the relationship between the approximation and generalization errors of deep operator networks and key factors such as network model size and training data size. Moreover, we address cases where input functions exhibit low-dimensional structures, allowing us to derive tighter error bounds. These results also hold for deep ReLU networks and other similar structures. Our results offer a partial explanation of the neural scaling laws in operator learning and provide a theoretical foundation for their applications.

cs.LG↗

GemDepth: Geometry-Embedded Features for 3D-Consistent Video Depth

Video depth estimation extends monocular prediction into the temporal domain to ensure coherence. However, existing methods often suffer from spatial blurring in fine-detail regions and temporal inconsistencies. We argue that current approaches, which primarily rely on temporal smoothing via Transformers, struggle to maintain strict 3D geometric consistency-particularly under rotations or drastic view changes. To address this, we propose GemDepth, a framework built on the insight that an explicit awareness of camera motion and global 3D structure is a prerequisite for 3D consistency. Distinctively, GemDepth introduces a Geometry-Embedding Module (GEM) that predicts inter-frame camera poses to generate implicit geometric embeddings. This injection of motion priors equips the network with intrinsic 3D perception and alignment capabilities. Guided by these geometric cues, our Alternating Spatio-Temporal Transformer (ASTT) captures latent point-level correspondences to simultaneously enhance spatial precision for sharp details and enforce rigorous temporal consistency. Furthermore, GemDepth employs a data-efficient training strategy, effectively bridging the gap between high efficiency and robust geometric consistency. As shown in Fig.2, comprehensive evaluations demonstrate that GemDepth achieves state-of-the-art performance across multiple datasets, particularly in complex dynamic scenarios. The code is publicly available at: https://github.com/Yuecheng919/GemDepth.

cs.CV↗

Transformers for Learning on Noisy and Task-Level Manifolds: Approximation and Generalization Insights

Transformers serve as the foundational architecture for large language and video generation models, such as GPT, BERT, SORA and their successors. Empirical studies have demonstrated that real-world data and learning tasks exhibit low-dimensional structures, along with some noise or measurement error. The performance of transformers tends to depend on the intrinsic dimension of the data/tasks, though theoretical understandings remain largely unexplored for transformers. This work establishes a theoretical foundation by analyzing the performance of transformers for regression tasks involving noisy input data near a manifold. Specifically, the input data are in a tubular neighborhood of a manifold, while the ground truth function depends on the projection of the noisy data onto this manifold, referred to as the task-level manifold. We prove approximation and generalization errors which crucially depend on the intrinsic dimension of the task-level manifold. Our results demonstrate that transformers can leverage low-complexity structures in learning task even when the input data are perturbed by high-dimensional noise. Our novel proof technique constructs representations of basic arithmetic operations by transformers, which may hold independent interest.

cs.LG↗

Understanding In-Context Learning on Structured Manifolds: Bridging Attention to Kernel Methods

While in-context learning (ICL) has achieved remarkable success in natural language and vision domains, its theoretical understanding-particularly in the context of structured geometric data-remains unexplored. This paper initiates a theoretical study of ICL for regression of Hölder functions on manifolds. We establish a novel connection between the attention mechanism and classical kernel methods, demonstrating that transformers effectively perform kernel-based prediction at a new query through its interaction with the prompt. This connection is validated by numerical experiments, revealing that the learned query-prompt scores for Hölder functions are highly correlated with the Gaussian kernel. Building on this insight, we derive generalization error bounds in terms of the prompt length and the number of training tasks. When a sufficient number of training tasks are observed, transformers give rise to the minimax regression rate of Hölder functions on manifolds, which scales exponentially with respect to the prompt length with the exponent depending on the intrinsic dimension of the manifold, rather than the ambient space dimension. Our result also characterizes how the generalization error scales with the number of training tasks, shedding light on the complexity of transformers as in-context kernel algorithm learners. Our findings provide foundational insights into the role of geometry in ICL and novels tools to study ICL of nonlinear models.

cs.LG↗

Understanding In-Context Learning for Nonlinear Regression with Transformers: Attention as Featurizer

Pre-trained transformers are able to learn from examples provided as part of the prompt without any weight updates, a remarkable ability known as in-context learning (ICL). Despite its demonstrated efficacy across various domains, the theoretical understanding of ICL is still developing. Whereas most existing theory has focused on linear models, we study ICL in the nonlinear regression setting. Through the interaction mechanism in attention, we explicitly construct transformer networks to realize nonlinear features, such as polynomial or spline bases, which span a wide class of functions. Based on this construction, we establish a framework to analyze end-to-end in-context nonlinear regression with the constructed features. Our theory provides finite-sample generalization error bounds in terms of context length and training set size. We numerically validate the theory on synthetic regression tasks.

cs.LG↗

Data-Driven Model Reduction using WeldNet: Windowed Encoders for Learning Dynamics

Many problems in science and engineering involve time-dependent, high dimensional datasets arising from complex physical processes, which are costly to simulate. In this work, we propose WeldNet: Windowed Encoders for Learning Dynamics, a data-driven nonlinear model reduction framework to build a low-dimensional surrogate model for complex evolution systems. Given time-dependent training data, we split the time domain into multiple overlapping windows, within which nonlinear dimension reduction is performed by auto-encoders to capture latent codes. Once a low-dimensional representation of the data is learned, a propagator network is trained to capture the evolution of the latent codes in each window, and a transcoder is trained to connect the latent codes between adjacent windows. The proposed windowed decomposition significantly simplifies propagator training by breaking long-horizon dynamics into multiple short, manageable segments, while the transcoders ensure consistency across windows. In addition to the algorithmic framework, we develop a mathematical theory establishing the representation power of WeldNet under the manifold hypothesis, justifying the success of nonlinear model reduction via deep autoencoder-based architectures. Our numerical experiments on various differential equations indicate that WeldNet can capture nonlinear latent structures and their underlying dynamics, outperforming both traditional projection-based approaches and recently developed nonlinear model reduction methods.

stat.ML↗

MonSter++: Unified Stereo Matching, Multi-view Stereo, and Real-time Stereo with Monodepth Priors

We introduce MonSter++, a geometric foundation model for multi-view depth estimation, unifying rectified stereo matching and unrectified multi-view stereo. Both tasks fundamentally recover metric depth from correspondence search and consequently face the same dilemma: struggling to handle ill-posed regions with limited matching cues. To address this, we propose MonSter++, a novel method that integrates monocular depth priors into multi-view depth estimation, effectively combining the complementary strengths of single-view and multi-view cues. MonSter++ fuses monocular depth and multi-view depth into a dual-branched architecture. Confidence-based guidance adaptively selects reliable multi-view cues to correct scale ambiguity in monocular depth. The refined monocular predictions, in turn, effectively guide multi-view estimation in ill-posed regions. This iterative mutual enhancement enables MonSter++ to evolve coarse object-level monocular priors into fine-grained, pixel-level geometry, fully unlocking the potential of multi-view depth estimation. MonSter++ achieves new state-of-the-art on both stereo matching and multi-view stereo. By effectively incorporating monocular priors through our cascaded search and multi-scale depth fusion strategy, our real-time variant RT-MonSter++ also outperforms previous real-time methods by a large margin. As shown in Fig.1, MonSter++ achieves significant improvements over previous methods across eight benchmarks from three tasks -- stereo matching, real-time stereo matching, and multi-view stereo, demonstrating the strong generality of our framework. Besides high accuracy, MonSter++ also demonstrates superior zero-shot generalization capability. We will release both the large and the real-time models to facilitate their use by the open-source community.

cs.CV↗

IDENT Review: Recent Advances in Identification of Differential Equations from Noisy Data

Differential equations and numerical methods are extensively used to model various real-world phenomena in science and engineering. With modern developments, we aim to find the underlying differential equation from a single observation of time-dependent data. If we assume that the differential equation is a linear combination of various linear and nonlinear differential terms, then the identification problem can be formulated as solving a linear system. The goal then reduces to finding the optimal coefficient vector that best represents the time derivative of the given data. We review some recent works on the identification of differential equations. We find some common themes for the improved accuracy: (i) The formulation of linear system with proper denoising is important, (ii) how to utilize sparsity and model selection to find the correct coefficient support needs careful attention, and (iii) there are ways to improve the coefficient recovery. We present an overview and analysis of these approaches about some recent developments on the topic.

math.NA↗

Identification of Differential Equations by Dynamics-Guided Weighted Weak Form with Voting

In the identification of differential equations from data, significant progresses have been made with the weak/integral formulation. In this paper, we explore the direction of finding more efficient and robust test functions adaptively given the observed data. While this is a difficult task, we propose weighting a collection of localized test functions for better identification of differential equations from a single trajectory of noisy observations on the differential equation. We find that using high dynamic regions is effective in finding the equation as well as the coefficients, and propose a dynamics indicator per differential term and weight the weak form accordingly. For stable identification against noise, we further introduce a voting strategy to identify the active features from an ensemble of recovered results by selecting the features that frequently occur in different weighting of test functions. Systematic numerical experiments are provided to demonstrate the robustness of our method.

math.NA↗