IZS 2018 Plenaries

Bayesian Suffix Trees and Context Tree Weighting

Ioannis Kontoyiannis (Cambridge University)

Abstract: The context tree weighting (CTW) and related algorithms, initially developed by Willems, Shtarkov, Tjalkens and their collaborators since the early 1990s, can be rephrased as methodologies for performing very effective Bayesian inference on a class of hierarchical models for discrete time-series data. We describe how these methods can be extended in several directions, both algorithmically and theoretically, to provide effective tools for statistical inference in much more general settings. In particular, we give a precise description of a new class of prior distributions on model space, and we describe a novel MCMC Metropolis-within-Gibbs algorithm for exploring the full posterior distribution. Our results are illustrated by extensive computational experiments on both synthetic and real data.

Slides (PDF) (additional material)

My Little Toolbox for Code Ensemble Performance Analysis

Neri Merhav (Technion – Israel Institute of Technology)

Slides (PDF)

A Differential View of Network Capacity

Michelle Effros (California Institute of Technology)

Slides (PDF)

Back to the IZS 2018 home page.