Alternatives to Direct Supervision Content Autoencoder What is

  • Slides: 48
Download presentation
可视计算 Alternatives to Direct Supervision 陈奕邻

可视计算 Alternatives to Direct Supervision 陈奕邻

Content • Autoencoder • What is Autoencoder • Sparse Autoencoder • De-nose Autoencoder •

Content • Autoencoder • What is Autoencoder • Sparse Autoencoder • De-nose Autoencoder • Variational Autoencoders (VAEs) • • Introduction to VAE Monte-Carlo and KL divergence Loss function Some applications • Generative Adversarial Networks (GANs) • • Introduction to GAN DCGAN CVAE-GAN 2

Autoencoder

Autoencoder

What is autoencoder • 损失函数 7

What is autoencoder • 损失函数 7

What is autoencoder • 梯度下降法 8

What is autoencoder • 梯度下降法 8

Autoencoder vs PCA Autoencoder PCA 14

Autoencoder vs PCA Autoencoder PCA 14

Varaitonal Autoencoder

Varaitonal Autoencoder

Introduction to VAE • Generative model • Goal: creat a new sample from Pdata(x)

Introduction to VAE • Generative model • Goal: creat a new sample from Pdata(x) that is not in the dataset • Difference between AE … dataset generated 16

Introduction to VAE • How to meagure similarity between pθ(x) and pdata(x) • Likelihood

Introduction to VAE • How to meagure similarity between pθ(x) and pdata(x) • Likelihood of data in pθ(x) • Variational Autoencoders (VAE) • Adversarial Game • Discriminator vs Generator • Generative Adversarial Networks (GAN) 17

Introduction to VAE Autoencoder Variational Autoencoder 18

Introduction to VAE Autoencoder Variational Autoencoder 18

Reparameterization trick 28

Reparameterization trick 28

Some applications • Generating Sentences from a Continuous Space 29

Some applications • Generating Sentences from a Continuous Space 29

Some applications 30

Some applications 30

Generative Adversarial Networks

Generative Adversarial Networks

Introduction to GAN • Generative Adversarial Networks 34

Introduction to GAN • Generative Adversarial Networks 34

DCGAN 36

DCGAN 36

DCGAN 37

DCGAN 37

CGAN 39

CGAN 39

CGAN 40

CGAN 40

CVAE-GAN 43

CVAE-GAN 43

CVAE-GAN 44

CVAE-GAN 44

CVAE-GAN 45

CVAE-GAN 45

Q&A 46

Q&A 46

References [1] Diederik P. K. , Max W. “Auto-Encoding Variational Bayes. ” 2014, ar.

References [1] Diederik P. K. , Max W. “Auto-Encoding Variational Bayes. ” 2014, ar. Xiv [2] Samuel R. B. , Luke V. , et al. “Generating Sentences from a Continuous Space. ” 2016, ar. Xiv [3] Ian J. G. , Jean P. A. , Mehdi M. et al. “Generative Adversarial Networks” 2014, NIPS [4] Alec R. , Luke M. “Unsupervised representation learning with deep convolutional generative adversarial networks. ” 2016, ar. Xiv [5] Mehdi M. , Simon O. “Conditional Generative Adversarial Nets. ” 2014 ar. Xiv [6] Bao J. M, Chen D. , et al. “CVAE-GAN: Fine-Grained Image Generation through Asymmetric Training” 2017 ar. Xiv [7] Phillip I. , Zhu J. Y. , Zhou T. H. “Image-to-Image Translation with Conditional Adversarial Networks” 2018 ar. Xiv 47

DOWNLOADS at http: //vcc. szu. edu. cn Thank You!

DOWNLOADS at http: //vcc. szu. edu. cn Thank You!