Efficient non-linear feature adaptation using Maxout networks

Steven J. Rennie; Xiaodong Cui; Vaibhava Goel

ICASSP 2016accepted0 citations

Efficient non-linear feature adaptation using Maxout networks

Steven J. Rennie, Xiaodong Cui, Vaibhava Goel

Abstract

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.

BibTeX

@inproceedings{icassp2016_efficientnonline,
  title = {Efficient non-linear feature adaptation using Maxout networks},
  author = {Steven J. Rennie and Xiaodong Cui and Vaibhava Goel},
  booktitle = {ICASSP 2016},
  year = {2016}
}