Under the auspices of the Computational Complexity Foundation (CCF)
We present an algorithm for learning a mixture of distributions.The algorithm is based on spectral projection and is efficient when the components of the mixture are logconcave distributions.