nonparametric mixture estimation

To address the challenge of nonparametrically modeling mixed component distributions in heterogeneous data, this paper proposes a finite mixture model with nonparametric components, where each component density is itself modeled via a Dirichlet process mixture (DPM) prior. First, we establish identifiability conditions for the mixing components under this framework. Second, we theoretically prove that the posterior contraction rate for component densities is polynomial—significantly faster than the logarithmic rate typical of conventional deconvolution for mixing measures. Third, to enable efficient Bayesian inference, we design a tailored MCMC algorithm. Extensive simulations and real-data analyses demonstrate the method’s high accuracy and robustness in identifying latent subgroups, estimating population-level and component-specific densities. The approach thus offers both rigorous theoretical guarantees and practical utility for complex heterogeneous data analysis.

Existing Bayesian mixture models assume independence among atoms of random probability measures, neglecting inter-atomic interactions—such as repulsion or attraction—that are critical for meaningful clustering. Method: We propose the first unified framework enabling both repulsive and attractive interactions among atoms, abandoning the classical independence assumption. We derive closed-form expressions for the posterior, marginal, and predictive distributions of interacting-atom random measures; avoid prespecifying finite point process structures; and pioneer the integration of Palm calculus into Bayesian nonparametric inference. Our hierarchical model combines Poisson, Gibbs, and determinantal point processes with shot-noise Cox processes to flexibly encode interaction patterns. Contribution/Results: The framework supports efficient MCMC sampling, ensures prior interpretability and algorithmic convergence, and significantly improves cluster separation and interpretability. Extensive experiments on synthetic and real-world datasets validate its effectiveness and robustness.

Gibbs sampling for Bayesian mixture models suffers from slow mixing in the marginal posterior over component assignments and struggles to jointly perform model selection and parameter inference. Method: We propose two novel joint-sampling MCMC algorithms: (1) a collapsed Gibbs sampler incorporating unconventional move sets, and (2) a prior-driven, rejection-free component allocation sampler. Both methods jointly update observation assignments and the number of components, unifying model fitting and dimensionality inference. Contribution/Results: Our approaches eliminate the need for post-hoc model selection and substantially improve Markov chain mixing efficiency. In latent class analysis tasks, they reduce mixing time by several-fold compared to state-of-the-art methods while achieving comparable or superior posterior inference accuracy. The framework provides an efficient, fully automated computational solution for high-dimensional Bayesian nonparametric modeling.

This paper addresses modeling population heterogeneity in mixed linear regression (i.e., random-coefficient) models. We propose a fully nonparametric maximum likelihood estimator (NPMLE) for the unknown mixing distribution (G^*), without prespecifying its parametric form or the number of components. Our method directly computes the NPMLE via convex optimization—yielding the first rigorous proof of its existence—and establishes, for finite samples, an optimal parametric-rate bound (up to logarithmic factors) on the Hellinger estimation error, circumventing discretization-induced bias inherent in conventional approaches. Theoretically and empirically, the estimator achieves both statistical efficiency and computational tractability: it significantly outperforms EM-based parametric methods on both discrete and continuous mixture simulations, as well as two real-world datasets, demonstrating strong robustness and practical utility.

This paper addresses the longstanding limitation in posterior summarization for nonparametric Bayesian mixture models—where inference has predominantly focused on random partition point estimates, neglecting direct inference on the mixing measure itself. We propose a decision-theoretic framework that prioritizes the mixing measure as the primary inferential target. Methodologically, we introduce, for the first time, a model-agnostic variant of the sliced Wasserstein distance, integrated with generalized geodesic projection and optimization on the symmetric positive-definite matrix manifold; leveraging the linear structure of Gaussian mixing measures, our approach delivers coherent point estimates of the mixing measure, density function, and random partition simultaneously. Compared to conventional paradigms, our method preserves statistical validity under complex dependency structures, substantially improves geometric coherence and computational efficiency in posterior summarization, and unifies support for both density estimation and clustering inference.

This work addresses the high computational cost and lack of convergence rate guarantees associated with nonparametric maximum likelihood estimation (NPMLE) in exponential family mixture models. The authors propose a data-compression-based acceleration strategy that, for the first time, reduces the likelihood evaluation complexity of NPMLE to logarithmic order. They establish rigorous statistical theory for the resulting approximate estimator, demonstrating that the proposed method achieves near-parametric convergence rates for marginal density estimation while substantially lowering computational overhead.

This study addresses nonparametric statistical inference for rational infinitely divisible mixture distributions comprising both discrete and continuous components, with the goal of jointly estimating the discrete part, the continuous part, and the quasi-Lévy measure. To this end, the work introduces, for the first time, a systematic estimation framework based on band-limited kernel functions with compact support in the frequency domain, effectively integrating nonparametric estimation theory with the analytic structure inherent to this class of distributions. Under mild regularity conditions, the proposed estimators achieve polynomial convergence rates and, in certain cases, approach parametric rates. Numerical simulations further corroborate the efficacy of the methodology.

This study addresses the problem of recovering latent components and estimating the mixing matrix from unlabeled finite mixture data, where only distributions sharing the same latent components but with unknown mixing weights are observed. The authors propose a novel identifiability theory based on marginal independence, decoupling irreducibility from marginal independence for the first time. They develop the Product-Marginal Maximum Mean Discrepancy (PM-MMD) estimation framework, which integrates affine combination analysis, the maximum mean discrepancy metric, and uniform convergence theory to achieve stable estimation. Experiments demonstrate that the proposed method significantly outperforms baseline approaches such as clustering and matrix factorization on both synthetic and flow cytometry data, with condition-aware representative selection substantially enhancing the accuracy and stability of component recovery.

This work addresses the challenge of efficiently clustering discrete distributions—such as mixtures of Bernoulli models—while simultaneously handling continuous distributions within a unified framework. To this end, the authors propose a simple projection-based clustering algorithm: it first computes the best rank-$k$ approximation of the data matrix and then applies $k$-means to this low-rank representation to obtain approximate cluster centers. Samples are subsequently projected onto these centers for final clustering assignments. The method is rotationally invariant, thereby avoiding the reliance on coordinate-specific projections inherent in traditional approaches. This approach validates McSherry’s conjecture that a geometric clustering algorithm exists for discrete distributions. Under natural separation conditions on the cluster centers, the algorithm provably achieves accurate clustering for both high-dimensional Gaussian and other continuous distributions as well as discrete ones.

This study addresses the challenge of statistical inference for multimodal geometric distributions in complex biological systems, particularly when density functions are defined only up to a normalizing constant. Building upon variational inference, this work integrates black-box variational inference, multiple importance sampling ELBO, and flow matching techniques to construct a highly expressive inference framework tailored for spatial transcriptomics. Key contributions include proposing CoLN, an unnormalized target density; revealing and refuting the conventional belief regarding performance gains of mixture models in variational inference; and developing a hybrid approach that combines multi-marginal flow matching with variational interpolation. Ultimately, this framework substantially enhances both the efficiency of approximate inference and the analytical capability for multimodal biological data, such as three-dimensional spatial transcriptomics.