Adversarial Variational Embedding for Robust Semi-supervised Learning

Abstract

Semi-supervised learning is sought for leveraging the unlabelled data when labelled data is difficult or expensive to acquire. Deep generative models (e.g., Variational Autoencoder (VAE)) and semisupervised Generative Adversarial Networks (GANs) have recently shown promising performance in semi-supervised classification for the excellent discriminative representing ability. However, the latent code learned by the traditional VAE is not exclusive (repeatable) for a specific input sample, which prevents it from excellent classification performance. In particular, the learned latent representation depends on a non-exclusive component which is stochastically sampled from the prior distribution. Moreover, the semi-supervised GAN models generate data from pre-defined distribution (e.g., Gaussian noises) which is independent of the input data distribution and may obstruct the convergence and is difficult to control the distribution of the generated data. To address the aforementioned issues, we propose a novel Adversarial Variational Embedding (AVAE) framework for robust and effective semi-supervised learning to leverage both the advantage of GAN as a high quality generative model and VAE as a posterior distribution learner. The proposed approach first produces an exclusive latent code by the model which we call VAE++, and meanwhile, provides a meaningful prior distribution for the generator of GAN. The proposed approach is evaluated over four different real-world applications and we show that our method outperforms the state-of-the-art models, which confirms that the combination of VAE++ and GAN can provide significant improvements in semisupervised classification.

RETRACTED ARTICLE: Human activity recognition with deep learning: overview, challenges and possibilities

“…This is important to remember that since this work uses profound networks, this relies on marks to ensure it is not unattended. Zhang et al [101] proposed the use for activity recognition of semi-supervised GAN. Unlike standard GANs, the discriminator allows a K + 1 classification in the semisupervised GAN classification, which involves operation and false recognition of results.…”

Analyzing the Effectiveness of the Brain–Computer Interface for Task Discerning Based on Machine Learning

“…As pointed out by Han et al [103], it is often reasonable to assume that the input-output mapping is similar across different models, so a better NN performance may be obtained by fitting all the parameters at the same time. Lastly, since poor generalization ability still limits the broader use of BCI, thus deep learning could be employed in the form of, e.g., autoencoders without manual feature selection [32,35]. In the case of training neural networks, it is required for the inputs and outputs of the network to be encoded as vectors of numbers.…”

“…Such a reduction enables to distinguish signals representing different types of mental activity that the BCI system is to recognize [10,30]. However, in deep learning classification, feature extraction is not always applied as signal characteristics may be automatically derived from autoencoders [31,32]. Moreover, Wu et al proposed an experimental scenario in which the feature selection and classification were performed simultaneously [33].…”

Data augmentation using generative adversarial neural networks on brain structural connectivity in multiple sclerosis

“…Interestingly, other applications of adversarial variational training frameworks have been reported. For example, Zhang et al [24] proposed a semi-supervised learning Adversarial Variational Embedding for leveraging both the power of GAN as a high quality generative model and Variational Au-toEncoder (VAE) as a posterior distribution learner. They demonstrated that the combination of VAE and GAN provided significant improvements of semisupervised classification.…”