Efficient non-linear feature adaptation using Maxout networks

In this paper we present a simple and effective method for doing non-linear feature adaptation using Maxout networks. The technique overcomes the need to sample the partition function during training, and overcomes the need to compute the Jacobian term and its gradient for each training case. Results on the Switchboard 1 task demonstrate that the approach can improve a state-of-the-art hybrid ASR system that utilizes i-vectors.