Topic 0: Deep Generative Models
What does "Generative" mean in Deep Generative Models?
"Generative" describes a class of statistical models that contrasts with discriminative models.
Informally:
-
Generative models can generate new data instances.
-
Discriminative models discriminate between different kinds of data instances.
Example: A generative model could generate new photos of animals that look like real animals, while a discriminative model could tell a dog from a cat.
https://xzz201920.medium.com/generative-model-v-s-discriminative-model-e1586935d277
More formally, given a set of data instances X and a set of labels Y:
-
Generative models capture the joint probability p(X, Y), or just p(X) if there are no labels.
-
Discriminative models capture the conditional probability p(Y | X).
What is the relation between Deep learning and Deep Generative Models?
Deep generative models are neural networks with many hidden layers trained to approximate complicated, high-dimensional probability distributions.
When trained successfully, we can use the DGM to estimate the likelihood of a given sample and to create new samples that are similar to samples from the unknown distribution.
Three approaches to generative models:
Generative Adversarial Networks (GANs)
Variational Autoencoders (VAEs)
Auto-regressive models
GANs
2014
2014
ConditionalGANs
2015
LAPGANs
2015
DCGANs
2016
BiGANs
2016
SGAN
2016
InfoGANs
2016
EBGAN
Auxilary
Classifier
GANs
2017
2017
Progressive GANs
2018
BigGANs
2019
SAGANs
2019
NRGANs
2019
AutoGANs
2020
Your
Local GANs
2020
MSG-GANs
A generative adversarial network (GAN) is a class of Machine learning frameworks designed by Ian Goodfellow and his colleagues in 2014
The main idea behind it is "Two neural networks contest with each other in a game (in the form of a zero-sum game, where one agent's gain is another agent's loss)".
GANs and its History:
Ian Goodfellow
Variational Autoencoders(VAEs):
The main idea is "to learn a low-dimensional latent representation of the training data called latent variables (variables which are not directly observed but are rather inferred through a mathematical model) which we assume to have generated our actual training data".
Variational AutoEncoders is first introduced by Diederik P Kingma and Max Welling in 2013.
https://medium.com/@realityenginesai/understanding-variational-autoencoders-and-their-applications-81a4f99efc0d
Auto-regressive Models/Networks:
DARNs (Deep AutoRegressive Networks) are generative sequential models, and are therefore often compared to other generative networks like GANs or VAEs; however, they are also sequence models and show promise in traditional sequence challenges like language processing and audio generation.
First, applications. Research in autoregressive networks tends toward two large subject areas: image generation and sequence modelling with a particular emphasis on audio generation and text to speech. Like PixelCNN is one of the best known autoregressive networks.
And more over autoregressive models such as PixelRNN instead train a network that models the conditional distribution of every individual pixel given previous pixels (to the left and to the top)
Intutive Picture of Deep Generative Models
https://ml.berkeley.edu/blog/posts/AR_intro/
-
Text-to-Image Translation
-
Image to Image Translation
-
Generate Photographs of Human Faces
-
3D Object Generation
-
Semantic-Image-to-Photo Translation
-
Photo Inpainting
-
Video Prediction
-
Face Aging
-
Super Resolution
-
Clothing Translation