Journal of Machine Learning Research最新文献

Discount regularization, using a shorter planning horizon when calculating the optimal policy, is a popular choice to avoid overfitting when faced with sparse or noisy data. It is commonly interpreted as de-emphasizing or ignoring delayed effects. In this paper, we prove two alternative views of discount regularization that expose unintended consequences and motivate novel regularization methods. In model-based RL, planning under a lower discount factor acts like a prior with stronger regularization on state-action pairs with more transition data. This leads to poor performance when the transition matrix is estimated from data sets with uneven amounts of data across state-action pairs. In model-free RL, discount regularization equates to planning using a weighted average Bellman update, where the agent plans as if the values of all state-action pairs are closer than implied by the data. Our equivalence theorems motivate simple methods that generalize discount regularization by setting parameters locally for individual state-action pairs rather than globally. We demonstrate the failures of discount regularization and how we remedy them using our state-action-specific methods across empirical examples with both tabular and continuous state spaces.

We consider the problem of testing for a difference in means between clusters of observations identified via $k$-means clustering. In this setting, classical hypothesis tests lead to an inflated Type I error rate. In recent work, Gao et al. (2022) considered a related problem in the context of hierarchical clustering. Unfortunately, their solution is highly-tailored to the context of hierarchical clustering, and thus cannot be applied in the setting of $k$-means clustering. In this paper, we propose a p-value that conditions on all of the intermediate clustering assignments in the $k$-means algorithm. We show that the p-value controls the selective Type I error for a test of the difference in means between a pair of clusters obtained using $k$-means clustering in finite samples, and can be efficiently computed. We apply our proposal on hand-written digits data and on single-cell RNA-sequencing data.

Bayesian mixture models are widely used for clustering of high-dimensional data with appropriate uncertainty quantification. However, as the dimension of the observations increases, posterior inference often tends to favor too many or too few clusters. This article explains this behavior by studying the random partition posterior in a non-standard setting with a fixed sample size and increasing data dimensionality. We provide conditions under which the finite sample posterior tends to either assign every observation to a different cluster or all observations to the same cluster as the dimension grows. Interestingly, the conditions do not depend on the choice of clustering prior, as long as all possible partitions of observations into clusters have positive prior probabilities, and hold irrespective of the true data-generating model. We then propose a class of latent mixtures for Bayesian clustering (Lamb) on a set of low-dimensional latent variables inducing a partition on the observed data. The model is amenable to scalable posterior inference and we show that it can avoid the pitfalls of high-dimensionality under mild assumptions. The proposed approach is shown to have good performance in simulation studies and an application to inferring cell types based on scRNAseq.