arXiv subjects
Publications and source records attributed to James Y. Zhao.
We present a new approach for random sampling of contingency tables of any size and constraints based on a recently introduced $\textit{probabilistic divide-and-conquer}$ technique. A simple exact sampling algorithm is presented for $2\times n$ tables, as well as a generalization where each entry of the table has a specified marginal distribution.
Sampling from combinatorial families can be difficult. However, complicated families can often be embedded within larger, simpler ones, for which easy sampling algorithms are known. We take advantage of such a relationship to describe a sampling algorithm for the smaller family, via a Markov chain started at a random sample of the larger family. The utility of the method is demonstrated via several examples, with particular emphasis on sampling labelled graphs with given degree sequence, a well-studied problem for which existing algorithms leave much room for improvement. For graphs with given degrees, with maximum degree $O(m^{1/4})$ where $m$ is the number of edges, we obtain an asymptotically uniform sample in $O(m)$ steps, which substantially improves upon existing algorithms.
We study the Hopfield model where the ratio $α$ of patterns to sites grows large. We prove that the free energy with inverse temperature $β$ and external field $B$ behaves like $β\sqrtα+γ$, where $γ$ is the limiting free energy of the Sherrington-Kirkpatrick model with inverse temperature $\sqrt2β$ and external field $B$.
We introduce a universality theorem for functionals of measures on partitions which "behave like" the Ewens measure. Various limit theorems for the Ewens measure, most notably the Poisson-Dirichlet limit for the longest parts, the functional central limit theorem for the number of parts, and the Erdos-Turan limit for the product of parts, extend to these asymptotically Ewens measures as easy corollaries. Our major contributions are: (1) extending the classes of measures for which these limit theorems hold; (2) characterising universality by a single, easily-checked criterion; and (3) greatly shortening the proofs of the limit theorems using the Feller coupling.
We consider a hidden Markov model with multiple observation processes, one of which is chosen at each point in time by a policy---a deterministic function of the information state---and attempt to determine which policy minimises the limiting expected entropy of the information state. Focusing on a special case, we prove analytically that the information state always converges in distribution, and derive a formula for the limiting entropy which can be used for calculations with high precision. Using this fomula, we find computationally that the optimal policy is always a threshold policy, allowing it to be easily found. We also find that the greedy policy is almost optimal.