Reviews: Learning HMMs with Nonparametric Emissions via Spectral Decompositions of Continuous Matrices

Strengths and Weaknesses Positive aspects: The paper is well written and has a clear and coherent structure. The discussion of related work is comprehensive. Non-parametric emission distributions add flexibility to the general HMM framework and reduce bias due to wrong modeling assumptions. Progress in this area should have theoretical and practical impact. The paper builds upon existing spectral methods for parametric HMMs but introduces novel techniques to extend those approaches to the non-parametric case.