Learning Vector Autoregressive Models With Latent Processes