Kernel size selection in Convolutional Neural Networks (CNNs) is a critical but often overlooked design decision that affects receptive field, feature extraction, computational cost, and model accuracy. This paper proposes the Best Kernel Size Estimation Function (BKSEF), a mathematically grounded and empirically validated framework for optimal, layer-wise kernel size determination. BKSEF balances information gain, computational efficiency, and accuracy improvements by integrating principles from information theory, signal processing, and learning theory. Extensive experiments on CIFAR-10, CIFAR-100, ImageNet-lite, ChestX-ray14, and GTSRB datasets demonstrate that BKSEF-guided architectures achieve up to 3.1 percent accuracy improvement and 42.8 percent reduction in FLOPs compared to traditional models using uniform 3x3 kernels. Two real-world case studies further validate the approach: one for medical image classification in a cloud-based setup, and another for traffic sign recognition on edge devices. The former achieved enhanced interpretability and accuracy, while the latter reduced latency and model size significantly, with minimal accuracy trade-off. These results show that kernel size can be an active, optimizable parameter rather than a fixed heuristic. BKSEF provides practical heuristics and theoretical support for researchers and developers seeking efficient and application-aware CNN designs. It is suitable for integration into neural architecture search pipelines and real-time systems, offering a new perspective on CNN optimization.
We study the task of bandit learning, also known as best-arm identification, under the assumption that the true reward function f belongs to a known, but arbitrary, function class F. We seek a general theory of bandit learnability, akin to the PAC framework for classification. Our investigation is guided by the following two questions: (1) which classes F are learnable, and (2) how they are learnable. For example, in the case of binary PAC classification, learnability is fully determined by a combinatorial dimension - the VC dimension- and can be attained via a simple algorithmic principle, namely, empirical risk minimization (ERM). In contrast to classical learning-theoretic results, our findings reveal limitations of learning in structured bandits, offering insights into the boundaries of bandit learnability. First, for the question of "which", we show that the paradigm of identifying the learnable classes via a dimension-like quantity fails for bandit learning. We give a simple proof demonstrating that no combinatorial dimension can characterize bandit learnability, even in finite classes, following a standard definition of dimension introduced by Ben-David et al. (2019). For the question of "how", we prove a computational hardness result: we construct a reward function class for which at most two queries are needed to find the optimal action, yet no algorithm can do so in polynomial time unless RP=NP. We also prove that this class admits efficient algorithms for standard algorithmic operations often considered in learning theory, such as an ERM. This implies that computational hardness is in this case inherent to the task of bandit learning. Beyond these results, we investigate additional themes such as learning under noise, trade-offs between noise models, and the relationship between query complexity and regret minimization.
Most of the widely used estimators of the average treatment effect (ATE) in causal inference rely on the assumptions of unconfoundedness and overlap. Unconfoundedness requires that the observed covariates account for all correlations between the outcome and treatment. Overlap requires the existence of randomness in treatment decisions for all individuals. Nevertheless, many types of studies frequently violate unconfoundedness or overlap, for instance, observational studies with deterministic treatment decisions -- popularly known as Regression Discontinuity designs -- violate overlap. In this paper, we initiate the study of general conditions that enable the identification of the average treatment effect, extending beyond unconfoundedness and overlap. In particular, following the paradigm of statistical learning theory, we provide an interpretable condition that is sufficient and nearly necessary for the identification of ATE. Moreover, this condition characterizes the identification of the average treatment effect on the treated (ATT) and can be used to characterize other treatment effects as well. To illustrate the utility of our condition, we present several well-studied scenarios where our condition is satisfied and, hence, we prove that ATE can be identified in regimes that prior works could not capture. For example, under mild assumptions on the data distributions, this holds for the models proposed by Tan (2006) and Rosenbaum (2002), and the Regression Discontinuity design model introduced by Thistlethwaite and Campbell (1960). For each of these scenarios, we also show that, under natural additional assumptions, ATE can be estimated from finite samples. We believe these findings open new avenues for bridging learning-theoretic insights and causal inference methodologies, particularly in observational studies with complex treatment mechanisms.
A fundamental question in modern machine learning is why large, over-parameterized models, such as deep neural networks and transformers, tend to generalize well, even when their number of parameters far exceeds the number of training samples. We investigate this phenomenon through the lens of information theory, grounded in universal learning theory. Specifically, we study a Bayesian mixture learner with log-loss and (almost) uniform prior over an expansive hypothesis class. Our key result shows that the learner's regret is not determined by the overall size of the hypothesis class, but rather by the cumulative probability of all models that are close, in Kullback-Leibler divergence distance, to the true data-generating process. We refer to this cumulative probability as the weight of the hypothesis. This leads to a natural notion of model simplicity: simple models are those with large weight and thus require fewer samples to generalize, while complex models have small weight and need more data. This perspective provides a rigorous and intuitive explanation for why over-parameterized models often avoid overfitting: the presence of simple hypotheses allows the posterior to concentrate on them when supported by the data. We further bridge theory and practice by recalling that stochastic gradient descent with Langevin dynamics samples from the correct posterior distribution, enabling our theoretical learner to be approximated using standard machine learning methods combined with ensemble learning. Our analysis yields non-uniform regret bounds and aligns with key practical concepts such as flat minima and model distillation. The results apply broadly across online, batch, and supervised learning settings, offering a unified and principled understanding of the generalization behavior of modern AI systems.
We propose a new decentralized robust kernel-based learning algorithm within the framework of reproducing kernel Hilbert space (RKHS) by utilizing a networked system that can be represented as a connected graph. The robust loss function $\mathcal{L}_\sigma$ induced by a windowing function $W$ and a robustness scaling parameter $\sigma>0$, can encompass a broad spectrum of robust losses. Consequently, the proposed algorithm effectively provides a unified decentralized learning framework for robust regression, which fundamentally differs from the existing distributed robust kernel learning schemes, all of which are divide-and-conquer based. We rigorously establish the learning theory and offer a comprehensive convergence analysis for the algorithm. We show each local robust estimator generated from the decentralized algorithm can be utilized to approximate the regression function. Based on kernel-based integral operator techniques, we derive general high confidence convergence bounds for each local approximating sequence in terms of the mean square distance, RKHS norm, and generalization error, respectively. Moreover, we provide rigorous selection rules for local sample size and show that, under properly selected step size and scaling parameter $\sigma$, the decentralized robust algorithm can achieve optimal learning rates (up to logarithmic factors) in both norms. The parameter $\sigma$ is shown to be essential for enhancing robustness while also ensuring favorable convergence behavior. The intrinsic connection among decentralization, sample selection, robustness of the algorithm, and its convergence is clearly reflected.
This paper proposes a novel stable learning theory for recurrent neural networks (RNNs), so-called variational adaptive noise and dropout (VAND). As stabilizing factors for RNNs, noise and dropout on the internal state of RNNs have been separately confirmed in previous studies. We reinterpret the optimization problem of RNNs as variational inference, showing that noise and dropout can be derived simultaneously by transforming the explicit regularization term arising in the optimization problem into implicit regularization. Their scale and ratio can also be adjusted appropriately to optimize the main objective of RNNs, respectively. In an imitation learning scenario with a mobile manipulator, only VAND is able to imitate sequential and periodic behaviors as instructed. https://youtu.be/UOho3Xr6A2w
Strategic litigation involves bringing a legal case to court with the goal of having a broader impact beyond resolving the case itself: for example, creating precedent which will influence future rulings. In this paper, we explore strategic litigation from the perspective of machine learning theory. We consider an abstract model of a common-law legal system where a lower court decides new cases by applying a decision rule learned from a higher court's past rulings. In this model, we explore the power of a strategic litigator, who strategically brings cases to the higher court to influence the learned decision rule, thereby affecting future cases. We explore questions including: What impact can a strategic litigator have? Which cases should a strategic litigator bring to court? Does it ever make sense for a strategic litigator to bring a case when they are sure the court will rule against them?
The Fourier representation for the uniform distribution over the Boolean cube has found numerous applications in algorithms and complexity analysis. Notably, in learning theory, learnability of Disjunctive Normal Form (DNF) under uniform as well as product distributions has been established through such representations. This paper makes five main contributions. First, it introduces a generalized Fourier expansion that can be used with any distribution $D$ through the representation of the distribution as a Bayesian network (BN). Second, it shows that the main algorithmic tools for learning with the Fourier representation, that use membership queries to approximate functions by recovering their heavy Fourier coefficients, can be used with slight modifications with the generalized expansion. These results hold for any distribution. Third, it analyzes the $L_1$ spectral norm of conjunctions under the new expansion, showing that it is bounded for a class of distributions which can be represented by difference bounded tree BN, where a parent node in the BN representation can change the conditional expectation of a child node by at most $\alpha<0.5$. Lower bounds are presented to show that such constraints are necessary. The fourth contribution uses these results to show the learnability of DNF with membership queries under difference bounded tree BN. The final contribution is to develop an algorithm for learning difference-bounded tree BN distributions, thus extending the DNF learnability result to cases where the distribution is not known in advance.
Attribute-efficient learning of sparse halfspaces has been a fundamental problem in machine learning theory. In recent years, machine learning algorithms are faced with prevalent data corruptions or even adversarial attacks. It is of central interest to design efficient algorithms that are robust to noise corruptions. In this paper, we consider that there exists a constant amount of malicious noise in the data and the goal is to learn an underlying $s$-sparse halfspace $w^* \in \mathbb{R}^d$ with $\text{poly}(s,\log d)$ samples. Specifically, we follow a recent line of works and assume that the underlying distribution satisfies a certain concentration condition and a margin condition at the same time. Under such conditions, we show that attribute-efficiency can be achieved by simple variants to existing hinge loss minimization programs. Our key contribution includes: 1) an attribute-efficient PAC learning algorithm that works under constant malicious noise rate; 2) a new gradient analysis that carefully handles the sparsity constraint in hinge loss minimization.
The basic question of delineating those statistical problems that are solvable without making any assumptions on the underlying data distribution has long animated statistics and learning theory. This paper characterizes when a convex M-estimation or stochastic optimization problem is solvable in such an assumption-free setting, providing a precise dividing line between solvable and unsolvable problems. The conditions we identify show, perhaps surprisingly, that Lipschitz continuity of the loss being minimized is not necessary for distribution free minimization, and they are also distinct from classical characterizations of learnability in machine learning.
Semantic change detection (CD) not only helps pinpoint the locations where changes occur, but also identifies the specific types of changes in land cover and land use. Currently, the mainstream approach for semantic CD (SCD) decomposes the task into semantic segmentation (SS) and CD tasks. Although these methods have achieved good results, they do not consider the incentive effect of task correlation on the entire model. Given this issue, this article further elucidates the SCD task through the lens of multitask learning theory and proposes a semantic change detection network based on boundary detection and task interaction (BT-SCD). In BT-SCD, the boundary detection (BD) task is introduced to enhance the correlation between the SS task and the CD task in SCD, thereby promoting positive reinforcement between SS and CD tasks. Furthermore, to enhance the communication of information between the SS and CD tasks, the pixel-level interaction strategy and the logit-level interaction strategy are proposed. Finally, to fully capture the temporal change information of the bitemporal features and eliminate their temporal dependency, a bidirectional change feature extraction module is proposed. Extensive experimental results on three commonly used datasets and a nonagriculturalization dataset (NAFZ) show that our BT-SCD achieves state-of-the-art performance.
We introduce an encoder-only approach to learn the evolution operators of large-scale non-linear dynamical systems, such as those describing complex natural phenomena. Evolution operators are particularly well-suited for analyzing systems that exhibit complex spatio-temporal patterns and have become a key analytical tool across various scientific communities. As terabyte-scale weather datasets and simulation tools capable of running millions of molecular dynamics steps per day are becoming commodities, our approach provides an effective tool to make sense of them from a data-driven perspective. The core of it lies in a remarkable connection between self-supervised representation learning methods and the recently established learning theory of evolution operators. To show the usefulness of the proposed method, we test it across multiple scientific domains: explaining the folding dynamics of small proteins, the binding process of drug-like molecules in host sites, and autonomously finding patterns in climate data. Code and data to reproduce the experiments are made available open source.
This paper presents a comprehensive synthesis of major breakthroughs in artificial intelligence (AI) over the past fifteen years, integrating historical, theoretical, and technological perspectives. It identifies key inflection points in AI' s evolution by tracing the convergence of computational resources, data access, and algorithmic innovation. The analysis highlights how researchers enabled GPU based model training, triggered a data centric shift with ImageNet, simplified architectures through the Transformer, and expanded modeling capabilities with the GPT series. Rather than treating these advances as isolated milestones, the paper frames them as indicators of deeper paradigm shifts. By applying concepts from statistical learning theory such as sample complexity and data efficiency, the paper explains how researchers translated breakthroughs into scalable solutions and why the field must now embrace data centric approaches. In response to rising privacy concerns and tightening regulations, the paper evaluates emerging solutions like federated learning, privacy enhancing technologies (PETs), and the data site paradigm, which reframe data access and security. In cases where real world data remains inaccessible, the paper also assesses the utility and constraints of mock and synthetic data generation. By aligning technical insights with evolving data infrastructure, this study offers strategic guidance for future AI research and policy development.
In statistical learning, models are classified as regular or singular depending on whether the mapping from parameters to probability distributions is injective. Most models with hierarchical structures or latent variables are singular, for which conventional criteria such as the Akaike Information Criterion and the Bayesian Information Criterion are inapplicable due to the breakdown of normal approximations for the likelihood and posterior. To address this, the Widely Applicable Information Criterion (WAIC) and the Widely Applicable Bayesian Information Criterion (WBIC) have been proposed. Since WAIC and WBIC are computed using posterior distributions at different temperature settings, separate posterior sampling is generally required. In this paper, we theoretically derive an asymptotic equation that links WAIC and WBIC, despite their dependence on different posteriors. This equation yields an asymptotically unbiased expression of WAIC in terms of the posterior distribution used for WBIC. The result clarifies the structural relationship between these criteria within the framework of singular learning theory, and deepens understanding of their asymptotic behavior. This theoretical contribution provides a foundation for future developments in the computational efficiency of model selection in singular models.
[Objective] This study focuses on addressing the current lack of a unified formal theoretical framework in machine learning, as well as the deficiencies in interpretability and ethical safety assurance. [Methods] A formal information model is first constructed, utilizing sets of well-formed formulas to explicitly define the ontological states and carrier mappings of typical components in machine learning. Learnable and processable predicates, along with learning and processing functions, are introduced to analyze the logical deduction and constraint rules of the causal chains within models. [Results] A meta-framework for machine learning theory (MLT-MF) is established. Based on this framework, universal definitions for model interpretability and ethical safety are proposed. Furthermore, three key theorems are proved: the equivalence of model interpretability and information recoverability, the assurance of ethical safety, and the estimation of generalization error. [Limitations] The current framework assumes ideal conditions with noiseless information-enabling mappings and primarily targets model learning and processing logic in static scenarios. It does not yet address information fusion and conflict resolution across ontological spaces in multimodal or multi-agent systems. [Conclusions] This work overcomes the limitations of fragmented research and provides a unified theoretical foundation for systematically addressing the critical challenges currently faced in machine learning.
In this work, we study the learning theory of reward modeling with pairwise comparison data using deep neural networks. We establish a novel non-asymptotic regret bound for deep reward estimators in a non-parametric setting, which depends explicitly on the network architecture. Furthermore, to underscore the critical importance of clear human beliefs, we introduce a margin-type condition that assumes the conditional winning probability of the optimal action in pairwise comparisons is significantly distanced from 1/2. This condition enables a sharper regret bound, which substantiates the empirical efficiency of Reinforcement Learning from Human Feedback and highlights clear human beliefs in its success. Notably, this improvement stems from high-quality pairwise comparison data implied by the margin-type condition, is independent of the specific estimators used, and thus applies to various learning algorithms and models.
We study the problem of learning junta distributions on $\{0, 1\}^n$, where a distribution is a $k$-junta if its probability mass function depends on a subset of at most $k$ variables. We make two main contributions: - We show that learning $k$-junta distributions is \emph{computationally} equivalent to learning $k$-parity functions with noise (LPN), a landmark problem in computational learning theory. - We design an algorithm for learning junta distributions whose statistical complexity is optimal, up to polylogarithmic factors. Computationally, our algorithm matches the complexity of previous (non-sample-optimal) algorithms. Combined, our two contributions imply that our algorithm cannot be significantly improved, statistically or computationally, barring a breakthrough for LPN.
Computer-generated holography (CGH) enables applications in holographic augmented reality (AR), 3D displays, systems neuroscience, and optical trapping. The fundamental challenge in CGH is solving the inverse problem of phase retrieval from intensity measurements. Physics-inspired neural networks (PINNs), especially Gerchberg-Saxton-based PINNs (GS-PINNs), have advanced phase retrieval capabilities. However, their performance strongly depends on forward models (FMs) and their hyperparameters (FMHs), limiting generalization, complicating benchmarking, and hindering hardware optimization. We present a systematic sensitivity analysis framework based on Saltelli's extension of Sobol's method to quantify FMH impacts on GS-PINN performance. Our analysis demonstrates that SLM pixel-resolution is the primary factor affecting neural network sensitivity, followed by pixel-pitch, propagation distance, and wavelength. Free space propagation forward models demonstrate superior neural network performance compared to Fourier holography, providing enhanced parameterization and generalization. We introduce a composite evaluation metric combining performance consistency, generalization capability, and hyperparameter perturbation resilience, establishing a unified benchmarking standard across CGH configurations. Our research connects physics-inspired deep learning theory with practical CGH implementations through concrete guidelines for forward model selection, neural network architecture, and performance evaluation. Our contributions advance the development of robust, interpretable, and generalizable neural networks for diverse holographic applications, supporting evidence-based decisions in CGH research and implementation.
A fundamental open problem in learning theory is to characterize the best-case teaching dimension $\operatorname{TS}_{\min}$ of a concept class $\mathcal{C}$ with finite VC dimension $d$. Resolving this problem will, in particular, settle the conjectured upper bound on Recursive Teaching Dimension posed by [Simon and Zilles; COLT 2015]. Prior work used a natural greedy algorithm to construct teaching sets recursively, thereby proving upper bounds on $\operatorname{TS}_{\min}$, with the best known bound being $O(d^2)$ [Hu, Wu, Li, and Wang; COLT 2017]. In each iteration, this greedy algorithm chooses to add to the teaching set the $k$ labeled points that restrict the concept class the most. In this work, we prove lower bounds on the performance of this greedy approach for small $k$. Specifically, we show that for $k = 1$, the algorithm does not improve upon the halving-based bound of $O(\log(|\mathcal{C}|))$. Furthermore, for $k = 2$, we complement the upper bound of $O\left(\log(\log(|\mathcal{C}|))\right)$ from [Moran, Shpilka, Wigderson, and Yuhudayoff; FOCS 2015] with a matching lower bound. Most consequentially, our lower bound extends up to $k \le \lceil c d \rceil$ for small constant $c>0$: suggesting that studying higher-order interactions may be necessary to resolve the conjecture that $\operatorname{TS}_{\min} = O(d)$.
Learning theory has traditionally followed a model-centric approach, focusing on designing optimal algorithms for a fixed natural learning task (e.g., linear classification or regression). In this paper, we adopt a complementary data-centric perspective, whereby we fix a natural learning rule and focus on optimizing the training data. Specifically, we study the following question: given a learning rule $\mathcal{A}$ and a data selection budget $n$, how well can $\mathcal{A}$ perform when trained on at most $n$ data points selected from a population of $N$ points? We investigate when it is possible to select $n \ll N$ points and achieve performance comparable to training on the entire population. We address this question across a variety of empirical risk minimizers. Our results include optimal data-selection bounds for mean estimation, linear classification, and linear regression. Additionally, we establish two general results: a taxonomy of error rates in binary classification and in stochastic convex optimization. Finally, we propose several open questions and directions for future research.