Domain Adaptation Using Autoencoder

With the availability of speech data obtained from different devices and varied acquisition conditions we are often faced with scenarios where the intrinsic discrepancy between the training and the test data has an adverse impact on affective speech analysis.

Domain adaptation using autoencoder. For the feature extraction layer the marginal distributions of source and target domains are matched by using the nonparametric maximum mean discrepancy measurement. Which are referred to as marginalized denoising autoencoder with adaptation distribution mdaad. A feature extraction layer and a classification layer.

The unsupervised domain adaptation is typically solved using generative adversarial networks gan framework. Recently adversarial domain adaptation models are applied to learn representations with adversarial training manners in feature space. In the new space samples with the same labels are close while simultaneously those of different labels are away from each other and the topology of each input domain is preserved.

The dana structure consists of a couple of encoding layers. Proposed autoencoder based domain adaptation in this paper the idea of dae is extended to aeda by replacing denoising concept with domain adaptation. Autoencoder based unsupervised domain adaptation for speech emotion recognition abstract.

Invertible autoencoder for domain adaptation. The domain adaptation using manifold alignment dama was proposed by wang and mahadevan by projecting both domain data to a new feature space. Heterogeneous domain adaptation network based on autoencoder.

Domain adaption based on elm autoencoder research article report. Tuytelaars unsupervised visual domain adaptation using subspace alignment in proceedings of the 2013 14th ieee international conference on computer vision iccv2013 pp. Deep learning is a powerful tool for domain adaptation by learning robust high level domain invariant representations.

1 marginalized denoising autoencoder mdae is used to promote domain invariant features which are critical to domain adaptation. Then resource rich out of domain dataset could be more useful for. Our approach consists of two stages.