329 papers
This paper proposes Randomized Kriging Believer, a parallel Bayesian optimization method that combines low computational complexity and asynchronous applicability with rigorous Bayesian expected regret guarantees to effectively optimize expensive black-box functions.
This paper introduces a partition-based functional ridge regression framework that decomposes coefficient functions into dominant and weaker components to apply differential penalization, thereby improving numerical stability, interpretability, and predictive performance in high-dimensional functional linear models without relying on explicit variable selection.
Co-Diffusion is a novel two-stage latent diffusion framework that enhances generalizable drug-target affinity prediction by aligning embeddings in an affinity-steered manifold and employing modality-specific diffusion as a regularizer, thereby achieving superior zero-shot performance on unseen molecular scaffolds and protein families compared to state-of-the-art methods.
This paper introduces a three-dimensional height-augmented ReLU network architecture that achieves significantly more efficient exponential approximation rates for analytic functions and provides the first quantitative, non-asymptotic high-order approximation for general functions, thereby advancing the theoretical foundation for designing parameter-efficient neural networks.
This paper proposes and analyzes non-penalized and sparse-penalized deep neural network estimators based on the minimum error entropy principle for nonparametric regression with strongly mixing dependent observations, establishing that they achieve minimax optimal convergence rates up to a logarithmic factor.
This paper formally defines algorithmic capture as the ability of infinite-width transformers to generalize to arbitrary problem sizes and demonstrates that, despite their universal expressivity, their inductive bias restricts them to learning only low-complexity algorithms within the Efficient Polynomial Time Heuristic Scheme (EPTHS) class.
This paper introduces diagnostic transport maps as a method to recalibrate potentially misspecified predictive distributions for high-dimensional inputs, enabling real-time local diagnostics and improved accuracy for rare events, as demonstrated in tropical cyclone intensity forecasting.
This paper proposes RIE-Greedy, a contextual bandit algorithm that leverages the inherent stochasticity of cross-validation-based regularization in black-box models to induce effective exploration, theoretically matching Thompson Sampling in simple cases and empirically outperforming existing methods in large-scale applications.
This paper introduces the "outrigger" local polynomial estimator, a distributionally adaptive method that achieves minimax optimality across various conditional error distributions without requiring structural assumptions like independence or symmetry, while guaranteeing performance at least as good as standard estimators under Gaussian errors.
This paper introduces a unified framework called wcPCA for developing worst-case optimal low-rank approximations across heterogeneous domains, extending the approach to new estimators and matrix completion while proving theoretical guarantees and demonstrating improved robustness in real-world applications with minimal loss to average performance.
This paper demonstrates that Langevin dynamics is fundamentally non-robust to small (or ) errors in the estimated score function, producing distributions far from the target even in polynomial time, thereby providing a theoretical justification for preferring diffusion models over Langevin dynamics when scores are learned from data.
This paper proposes a doubly robust estimator combined with a novel jackknife spatial HAC variance correction to achieve valid asymptotic inference when using machine learning predictions for missing-at-random labels in the presence of spatial dependence, addressing the instability caused by cross-fitting-induced correlations.
This paper introduces a formal framework decomposing observational multiplicity in decision trees into leaf and structural regret, demonstrating that structural instability is the primary driver of prediction variability and that leveraging these measures for selective prediction significantly enhances model safety and recall.
This paper proposes and validates a free energy-based social bandit learning algorithm that enables agents to effectively identify and leverage the expertise of both expert and non-expert peers without reward information, thereby achieving optimal policy convergence and logarithmic regret while significantly outperforming existing methods in diverse social learning scenarios.
This paper establishes that the Follow-the-Perturbed-Leader (FTPL) algorithm, when combined with Fréchet or Pareto distributions and an improved conditional geometric resampling technique, achieves optimal best-of-both-worlds regret guarantees for -set semi-bandit problems while reducing computational complexity from to .
This paper provides a learning-theoretic analysis of model collapse by introducing a replay adversary framework, demonstrating that while replay does not hinder uniform generation, it fundamentally limits non-uniform and limit-based generation, thereby offering theoretical justification for practical mitigation strategies like data cleaning and watermarking while also revealing their potential failure points.
This paper establishes the fundamental framework for quaternion machine learning by introducing augmented statistics, widely linear models, quaternion calculus, and mean square estimation to facilitate the adoption of this multidimensional domain.
This paper proposes a novel test-then-pool framework that leverages kernel-based distributional equivalence testing and resampling methods to safely fuse historical and concurrent control data in randomized controlled trials, thereby improving statistical power while rigorously controlling Type-I error rates.
EnTransformer is a deep generative framework that combines Transformers with the stochastic "engression" paradigm and energy-based scoring to produce well-calibrated, non-parametric multivariate probabilistic forecasts that outperform existing benchmarks on complex, correlated time series data.
This paper introduces Local Adjacency Spectral Embedding (LASE), a novel method that overcomes the limitations of global spectral embedding by uncovering locally low-dimensional network structures through weighted spectral decomposition, thereby improving local reconstruction, visualization, and theoretical guarantees via finite-sample bounds and spectral gap analysis.