Ladder networks are a notable new concept in the field of semi-supervisedlearning by showing state-of-the-art results in image recognition tasks whilebeing compatible with many existing neural architectures. We present therecurrent ladder network, a novel modification of the ladder network, forsemi-supervised learning of recurrent neural networks which we evaluate with aphoneme recognition task on the TIMIT corpus. Our results show that the modelis able to consistently outperform the baseline and achieve fully-supervisedbaseline performance with only 75% of all labels which demonstrates that themodel is capable of using unsupervised data as an effective regulariser.
展开▼
