On Mixtures of Markov Chains
Abstract
We study the problem of reconstructing a mixture of Markov chains from the trajectories generated by random walks through the state space. Under mild non-degeneracy conditions, we show that we can uniquely reconstruct the underlying chains by only considering trajectories of length three, which represent triples of states. Our algorithm is spectral in nature, and is easy to implement.
Cite
Text
Gupta et al. "On Mixtures of Markov Chains." Neural Information Processing Systems, 2016.Markdown
[Gupta et al. "On Mixtures of Markov Chains." Neural Information Processing Systems, 2016.](https://mlanthology.org/neurips/2016/gupta2016neurips-mixtures/)BibTeX
@inproceedings{gupta2016neurips-mixtures,
title = {{On Mixtures of Markov Chains}},
author = {Gupta, Rishi and Kumar, Ravi and Vassilvitskii, Sergei},
booktitle = {Neural Information Processing Systems},
year = {2016},
pages = {3441-3449},
url = {https://mlanthology.org/neurips/2016/gupta2016neurips-mixtures/}
}