Learning the Dependence Graph of Time Series with Latent Factors

摘要: This paper considers the problem of learning, from samples, dependency structure a system linear stochastic differential equations, when some variables are latent. We observe time evolution variables, and never other variables; this, we would like to find observed - separating out spurious interactions caused by latent variables' series. develop new convex optimization based method do so in case number is smaller than ones. For between sparse, theoretically establish high-dimensional scaling result for recovery. verify our theoretical with both synthetic real data (from stock market).