Deep Learners and Deep Learner Descriptors for Medical Applications by Unknown

Author:Unknown
Language: eng
Format: epub
ISBN: 9783030427504
Publisher: Springer International Publishing

5.2.3 Recurrent Sequence to Sequence Autoencoders

Acoustic sequences are typically varying length signals; this highlights a drawback for CNNs-based deep representation learning architectures which generally require inputs of fixed dimensionality. Moreover, many DNN systems applied for representation learning, including Restricted Boltzmann Machines (RBMs) or stacked autoencoders, do not explicitly account for the inherent sequential nature of audio signals [53]. To learn fixed-length representations from variable-length data with sequential nature, sequence to sequence learning with recurrent neural networks (RNNs) has been proposed in machine translation [54–56]. Sequence to sequence autoencoders (S2SAEs) have been used for unsupervised pre-training of RNNs with state-of-the-art results on image recognition or text classification tasks [57]. Variational S2SAEs have been employed to learn representations of sentences, and to create new sentences from the latent space [58, 59]. Furthermore, Weninger et al. used denoising recurrent autoencoders to learn variable-length representations of audio for reverberated speech recognition [60].

Fig. 5.10A high-level overview of the recurrent autoencoder used in auDeep

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.