Research indexJournal article
Generative Model with Kernel Density Estimation
Tan, Yin, Zhao
1 Aug 2018
NobleIDark:/48914/w1/20261004/13D1B7B0DOI10.1109/icnidc.2018.8525628
Generative Model with Kernel Density Estimation
Tan, Yin, Zhao · 2018 · Subscription
NobleIDark:/48914/w1/20261004/13D1B7B0DOI10.1109/icnidc.2018.8525628
The full text is at the publisher
You can read the abstract here, or open it at the publisher.
Abstract
In this paper, a novel generative model is proposed based on kernel density estimation. Different from other generative models like GANs [1] and VAEs [2], the model directly learns the distribution of training dataset in a non-parametric way. The mean absolute error (MAE) between the estimated probability density of the real and generated samples is used as loss function for model to train. To address the sparsity of high-dimensional distribution, we associate the proposed generative model with an encoder, by which the observed high-dimensional data can be transformed to low-dimensional latent features. Results on the MNIST dataset show that our proposed model is able to generate samples visually indistinguishable from the real ones.