SlideShare a Scribd company logo
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Wasserstein GAN
Bar Vinograd
The First Original Independent Seminar No. 5
03.05.2017
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Introduction
Unsupervised Learning
unsupervised learning
For data {x(i)}m
i=1 and a family of densities {Pθ}θ∈Rd solve
max
θ∈Rd
1
m
m
i=1
log Pθ(x(i)
)
or, min KL(Pr Pθ)
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Introduction
GANs
Find gθ : Z → X s.t.
Z ∈ Z is a random variable e.g. a Gaussian distribution
X is the domain being modelled (e.g. images, texts,
audio)
gθ a network of some kind
the distribution induced by gθ i.e. Pg is close to Pr
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Introduction
GANs are hard to train
Problems
Saturated gradients
Loss is not correlated with convergence
Unstable
Mode collapse
In general Pr and Pθ unlikley to have non-negligible
intersection
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Introduction
GANs are hard to train
Problems
Saturated gradients
Loss is not correlated with convergence
Unstable
Mode collapse
In general Pr and Pθ unlikley to have non-negligible
intersection
Solutions
Balancing generator and discriminator. This gives a lower
bound on loss and avoids collapse.
Apply random noise to real samples - creates an
intesections
−log(D) trick for generator loss
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Introduction
GANs are hard to train
Goodfello
(2017)
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
Kullback-Leibler (KL) divergence
KL(Pr Pg ) = log
Pr (x)
Pg (x)
Pr (x)dµ(x)
Jensen-Shannon (JS) distance
JS(Pr Pg ) =
1
2
(KL(Pr M) + KL(Pg M))
where M = 1
2 (Pr + Pg )
Total Variation (TV) distance
δ(Pr , Pg ) = sup
A⊂ΣX
|Pr (A) − Pg (A)|
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
Earth Mover (EM) distance or Wasserstein-1
W (Pr , Pg ) = inf
γ∈Π(Pr ,Pg )
E(x,y)∼γ [ x − y ]
where Π(Pr , Pg ) is the set of all copulings of Pr and Pg .
Which are all the distributions on X2 with marginals Pr
and Pg .
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
EM Illustration
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
EM Illustration
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
EM Illustration
Now do it with high dimensional dirt piles
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Distances
Example
Let
Z ∼ U[0, 1]
P0 be the distribution of points (0, Z) ∈ R2
gθ(z) = (θ, z)
Then
W (P0, Pθ) = |θ|
JS(P0, Pθ) =
log(2) θ = 0
0 θ = 0
KL(P0 Pθ) =
∞ θ = 0
0 θ = 0
δ(P0, Pθ) =
1 θ = 0
0 θ = 0
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
All distances other than EM are not continuous and so are
their derivatives.
Disjoint or measure zero intesection between supports of
function family and real distribution is common
We would like our loss to have an informative gradient
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Summary
Wasserstein (or EM) loss for neural networks is continuous and
differentiable almost everywhere. Moreover, Convergence in KL
implies convergence in TV and JS which implies convergence in
EM.
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
Kantorovich-Rubinstein duality
W (Pr , Pθ) = sup
f L≤1
Ex∼Pr [f (x)] − Ex∼Pθ
[f (x)]
where f : X → R and is 1-Lipschitz.
A function f : X → Y is K-Lipschitz if there exists a
K ≥ 0 s.t.
f (x1) − f (x2) ≤ K x1 − x2
for all x1, x2 ∈ X
Unllike the defintion of EM, this duality provides us with a
tractable definition
This is a private case of a integral probability metric
(IPM). For example, it is also defined for TV with the
appropriate choice of function family.
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
which tells us how to train the generator
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
clipping keeps the discriminator Lipschitz
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
clipping keeps the critic Lipschitz
training the critic to optimality
gradients do not saturate
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Training
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Results
64x64x3 image generation on LSUN-Bedrooms dataset
Compared with DCGAN, train with the −log(D) trick
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Results
JS generator loss
loss is saturated
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Results
WGAN generator loss
loss is correlated with image quality
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Summary
Benefits
Theortically sound critic loss
Stable training
Loss correlates with desired result
No mode collapse
Problems
Does not work well with momentum-based optimizers e.g.
Adam
Slower to converge than KL loss
Requires hyper-parameter tuning
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
Diminished capacity. Lots of Lipschitz functions are not
included in this family
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
Diminished capacity. Lots of Lipschitz functions are not
included in this family
Exploding or vanishing gradients (w.o. batchnorm)
depending on c
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
Diminished capacity. Lots of Lipschitz functions are not
included in this family
Exploding or vanishing gradients (w.o. batchnorm)
depending on c
Weights tend to saturate i.e. −c or c
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
Diminished capacity. Lots of Lipschitz functions are not
included in this family
Exploding or vanishing gradients (w.o. batchnorm)
depending on c
Weights tend to saturate i.e. −c or c
Unstable with momentum-based techniques e.g. Adam
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
The Kantorovich-Rubinstein duality states that
W (Pr , Pθ) = sup
f L≤1
Ex∼Pr [f (x)] − Ex∼Pθ
[f (x)]
where f : X → R and is 1-Lipschitz.
A differentiable function is 1-Lipchitz iff its graidents are
in the unit ball
The optimal solution for the duality is has gradients with
norm 1 almost everywhere i.e. on the unit ball
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Problems with weight clipping
Enforcing unit sphere gradients almost everywhere is not
tractable
Sampling random points in X and taking the gradient on
them
Gradient Penalty
L = Ex∼Pg [D(x)] − Ex∼Pr [D(x)]
critic loss
+λ Eˆx∼Pˆx
( ˆx D(ˆx) 2 − 1)2
gradient penalty
No batch norm in critic - penalizing the norm per sample
and not per batch. Use layer normaliztion instead
Use Adam
Sample uniformaly along lines between samples from Pr
and Pg
Sphere under 2
λ = 10
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Toy Datasets
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Toy Datasets
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Speed on CIFAR-10
but architecture must be stablized first. Generator and
discriminator balanced.
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Overfitting on MNIST
(a) LSUN (b) MNIST with penalty (left) and clipping (right)
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
LSUN
No BN and a constant number of filters in the generator,
as in Arjovsky et al. (2017)
4-layer 512-dim ReLU MLP generator, as in Arjovsky et al.
(2017)
No normalization in either the discriminator or generator
Gated multiplicative nonlinearities everywhere, as in van
den Oord et al. (2016)
tanh nonlinearities everywhere
101-layer ResNet generator and discriminator
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
LSUN
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
LSUN
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
LSUN
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Language Modeling
First general language model trained entirely adversarially
without a supervised maximum-likelihood loss. Here X simplex
of degree n and Pr can be thought of as a histogram on the
alphabet.
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Outline
1 Wasserstein GAN
Introduction
Distances
EM Properties
Training
Results
Summary
2 Improved Training of Wasserstein GANs
Theory
Algorithm
Results
Summary
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Summary
Improves and maintains benefits of original WGAN
Fixed problems with weight clipping: Adam, Hyper
parameter Tuning, Depth
Enables a richer critic family
Wasserstein
GAN
Bar Vinograd
Wasserstein
GAN
Introduction
Distances
EM Properties
Training
Results
Summary
Improved
Training of
Wasserstein
GANs
Theory
Algorithm
Results
Summary
Thank You
Wasserstein
GAN
Bar Vinograd
Appendix
For Further
Reading
For Further Reading I
M. Arjovsky and L. Bottou
Towards principled methods for training generative
adverserial networks.
under review for ICLR 2017, abs/1701.04862, 2017.
M. Arjovsky, S. Chintala, and L. Bottou
Wasserstein GAN.
abs/1701.07875, 2017.
I. Gulrajani, F, Ahmed, M. Arjovsky, V, Dumoulin, and A.
Courville
Improved Training of Wasserstein GANs.
abs/1704.00028, 2017.

More Related Content

What's hot

[DL輪読会]Conditional Neural Processes
[DL輪読会]Conditional Neural Processes[DL輪読会]Conditional Neural Processes
[DL輪読会]Conditional Neural Processes
Deep Learning JP
 
Variational AutoEncoder
Variational AutoEncoderVariational AutoEncoder
Variational AutoEncoder
Kazuki Nitta
 
[DL輪読会]Disentangling by Factorising
[DL輪読会]Disentangling by Factorising[DL輪読会]Disentangling by Factorising
[DL輪読会]Disentangling by Factorising
Deep Learning JP
 
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
GAN in medical imaging
GAN in medical imagingGAN in medical imaging
GAN in medical imaging
Cheng-Bin Jin
 
Diffusion models beat gans on image synthesis
Diffusion models beat gans on image synthesisDiffusion models beat gans on image synthesis
Diffusion models beat gans on image synthesis
BeerenSahu
 
(DL hacks輪読) Deep Kernel Learning
(DL hacks輪読) Deep Kernel Learning(DL hacks輪読) Deep Kernel Learning
(DL hacks輪読) Deep Kernel Learning
Masahiro Suzuki
 
PRML学習者から入る深層生成モデル入門
PRML学習者から入る深層生成モデル入門PRML学習者から入る深層生成モデル入門
PRML学習者から入る深層生成モデル入門
tmtm otm
 
Mean Teacher
Mean TeacherMean Teacher
Mean Teacher
harmonylab
 
PR-409: Denoising Diffusion Probabilistic Models
PR-409: Denoising Diffusion Probabilistic ModelsPR-409: Denoising Diffusion Probabilistic Models
PR-409: Denoising Diffusion Probabilistic Models
Hyeongmin Lee
 
Journal Club: VQ-VAE2
Journal Club: VQ-VAE2Journal Club: VQ-VAE2
Journal Club: VQ-VAE2
Takuya Koumura
 
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
Deep Learning JP
 
Latent Dirichlet Allocation
Latent Dirichlet AllocationLatent Dirichlet Allocation
Latent Dirichlet Allocation
Sangwoo Mo
 
猫でも分かるVariational AutoEncoder
猫でも分かるVariational AutoEncoder猫でも分かるVariational AutoEncoder
猫でも分かるVariational AutoEncoder
Sho Tatsuno
 
最近(2020/09/13)のarxivの分布外検知の論文を紹介
最近(2020/09/13)のarxivの分布外検知の論文を紹介最近(2020/09/13)のarxivの分布外検知の論文を紹介
最近(2020/09/13)のarxivの分布外検知の論文を紹介
ぱんいち すみもと
 
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
Masahiro Suzuki
 
Generating Diverse High-Fidelity Images with VQ-VAE-2
Generating Diverse High-Fidelity Images with VQ-VAE-2Generating Diverse High-Fidelity Images with VQ-VAE-2
Generating Diverse High-Fidelity Images with VQ-VAE-2
harmonylab
 
変分推論と Normalizing Flow
変分推論と Normalizing Flow変分推論と Normalizing Flow
変分推論と Normalizing FlowAkihiro Nitta
 
Finding connections among images using CycleGAN
Finding connections among images using CycleGANFinding connections among images using CycleGAN
Finding connections among images using CycleGAN
NAVER Engineering
 
GAN - Generative Adversarial Nets
GAN - Generative Adversarial NetsGAN - Generative Adversarial Nets
GAN - Generative Adversarial Nets
KyeongUkJang
 

What's hot (20)

[DL輪読会]Conditional Neural Processes
[DL輪読会]Conditional Neural Processes[DL輪読会]Conditional Neural Processes
[DL輪読会]Conditional Neural Processes
 
Variational AutoEncoder
Variational AutoEncoderVariational AutoEncoder
Variational AutoEncoder
 
[DL輪読会]Disentangling by Factorising
[DL輪読会]Disentangling by Factorising[DL輪読会]Disentangling by Factorising
[DL輪読会]Disentangling by Factorising
 
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
Variational Autoencoders VAE - Santiago Pascual - UPC Barcelona 2018
 
GAN in medical imaging
GAN in medical imagingGAN in medical imaging
GAN in medical imaging
 
Diffusion models beat gans on image synthesis
Diffusion models beat gans on image synthesisDiffusion models beat gans on image synthesis
Diffusion models beat gans on image synthesis
 
(DL hacks輪読) Deep Kernel Learning
(DL hacks輪読) Deep Kernel Learning(DL hacks輪読) Deep Kernel Learning
(DL hacks輪読) Deep Kernel Learning
 
PRML学習者から入る深層生成モデル入門
PRML学習者から入る深層生成モデル入門PRML学習者から入る深層生成モデル入門
PRML学習者から入る深層生成モデル入門
 
Mean Teacher
Mean TeacherMean Teacher
Mean Teacher
 
PR-409: Denoising Diffusion Probabilistic Models
PR-409: Denoising Diffusion Probabilistic ModelsPR-409: Denoising Diffusion Probabilistic Models
PR-409: Denoising Diffusion Probabilistic Models
 
Journal Club: VQ-VAE2
Journal Club: VQ-VAE2Journal Club: VQ-VAE2
Journal Club: VQ-VAE2
 
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
[DL輪読会]Graph Convolutional Policy Network for Goal-Directed Molecular Graph G...
 
Latent Dirichlet Allocation
Latent Dirichlet AllocationLatent Dirichlet Allocation
Latent Dirichlet Allocation
 
猫でも分かるVariational AutoEncoder
猫でも分かるVariational AutoEncoder猫でも分かるVariational AutoEncoder
猫でも分かるVariational AutoEncoder
 
最近(2020/09/13)のarxivの分布外検知の論文を紹介
最近(2020/09/13)のarxivの分布外検知の論文を紹介最近(2020/09/13)のarxivの分布外検知の論文を紹介
最近(2020/09/13)のarxivの分布外検知の論文を紹介
 
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
(DL hacks輪読) How to Train Deep Variational Autoencoders and Probabilistic Lad...
 
Generating Diverse High-Fidelity Images with VQ-VAE-2
Generating Diverse High-Fidelity Images with VQ-VAE-2Generating Diverse High-Fidelity Images with VQ-VAE-2
Generating Diverse High-Fidelity Images with VQ-VAE-2
 
変分推論と Normalizing Flow
変分推論と Normalizing Flow変分推論と Normalizing Flow
変分推論と Normalizing Flow
 
Finding connections among images using CycleGAN
Finding connections among images using CycleGANFinding connections among images using CycleGAN
Finding connections among images using CycleGAN
 
GAN - Generative Adversarial Nets
GAN - Generative Adversarial NetsGAN - Generative Adversarial Nets
GAN - Generative Adversarial Nets
 

Similar to Wasserstein GAN

Gaussian Dictionary for Compressive Sensing of the ECG Signal
Gaussian Dictionary for Compressive Sensing of the ECG SignalGaussian Dictionary for Compressive Sensing of the ECG Signal
Gaussian Dictionary for Compressive Sensing of the ECG Signal
Riccardo Bernardini
 
Value Function Geometry and Gradient TD
Value Function Geometry and Gradient TDValue Function Geometry and Gradient TD
Value Function Geometry and Gradient TD
Ashwin Rao
 
Secure Domination in graphs
Secure Domination in graphsSecure Domination in graphs
Secure Domination in graphs
Mahesh Gadhwal
 
129966863283913778[1]
129966863283913778[1]129966863283913778[1]
129966863283913778[1]威華 王
 
Distributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUsDistributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUs
Pantelis Sopasakis
 
Metrics for generativemodels
Metrics for generativemodelsMetrics for generativemodels
Metrics for generativemodels
Dai-Hai Nguyen
 
EE402B Radio Systems and Personal Communication Networks-Formula sheet
EE402B Radio Systems and Personal Communication Networks-Formula sheetEE402B Radio Systems and Personal Communication Networks-Formula sheet
EE402B Radio Systems and Personal Communication Networks-Formula sheet
Haris Hassan
 
Daa chpater14
Daa chpater14Daa chpater14
Daa chpater14
B.Kirron Reddi
 
New approaches for boosting to uniformity
New approaches for boosting to uniformityNew approaches for boosting to uniformity
New approaches for boosting to uniformity
Nikita Kazeev
 
Intelligent fault diagnosis for power distribution systemcomparative studies
Intelligent fault diagnosis for power distribution systemcomparative studiesIntelligent fault diagnosis for power distribution systemcomparative studies
Intelligent fault diagnosis for power distribution systemcomparative studies
nooriasukmaningtyas
 
Wasserstein gan
Wasserstein ganWasserstein gan
Wasserstein gan
Jinho Lee
 
Bayesian modelling and computation for Raman spectroscopy
Bayesian modelling and computation for Raman spectroscopyBayesian modelling and computation for Raman spectroscopy
Bayesian modelling and computation for Raman spectroscopy
Matt Moores
 
20 Single Source Shorthest Path
20 Single Source Shorthest Path20 Single Source Shorthest Path
20 Single Source Shorthest Path
Andres Mendez-Vazquez
 
Introducing Zap Q-Learning
Introducing Zap Q-Learning   Introducing Zap Q-Learning
Introducing Zap Q-Learning
Sean Meyn
 
Chapter 25 aoa
Chapter 25 aoaChapter 25 aoa
Chapter 25 aoa
Hanif Durad
 
New tools from the bandit literature to improve A/B Testing
New tools from the bandit literature to improve A/B TestingNew tools from the bandit literature to improve A/B Testing
New tools from the bandit literature to improve A/B Testing
recsysfr
 
Concentration inequality in Machine Learning
Concentration inequality in Machine LearningConcentration inequality in Machine Learning
Concentration inequality in Machine Learning
VARUN KUMAR
 
NP-Completeness - II
NP-Completeness - IINP-Completeness - II
NP-Completeness - II
Amrinder Arora
 
Da36615618
Da36615618Da36615618
Da36615618
IJERA Editor
 
Minimum spanning tree
Minimum spanning treeMinimum spanning tree
Minimum spanning tree
AhmedMalik74
 

Similar to Wasserstein GAN (20)

Gaussian Dictionary for Compressive Sensing of the ECG Signal
Gaussian Dictionary for Compressive Sensing of the ECG SignalGaussian Dictionary for Compressive Sensing of the ECG Signal
Gaussian Dictionary for Compressive Sensing of the ECG Signal
 
Value Function Geometry and Gradient TD
Value Function Geometry and Gradient TDValue Function Geometry and Gradient TD
Value Function Geometry and Gradient TD
 
Secure Domination in graphs
Secure Domination in graphsSecure Domination in graphs
Secure Domination in graphs
 
129966863283913778[1]
129966863283913778[1]129966863283913778[1]
129966863283913778[1]
 
Distributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUsDistributed solution of stochastic optimal control problem on GPUs
Distributed solution of stochastic optimal control problem on GPUs
 
Metrics for generativemodels
Metrics for generativemodelsMetrics for generativemodels
Metrics for generativemodels
 
EE402B Radio Systems and Personal Communication Networks-Formula sheet
EE402B Radio Systems and Personal Communication Networks-Formula sheetEE402B Radio Systems and Personal Communication Networks-Formula sheet
EE402B Radio Systems and Personal Communication Networks-Formula sheet
 
Daa chpater14
Daa chpater14Daa chpater14
Daa chpater14
 
New approaches for boosting to uniformity
New approaches for boosting to uniformityNew approaches for boosting to uniformity
New approaches for boosting to uniformity
 
Intelligent fault diagnosis for power distribution systemcomparative studies
Intelligent fault diagnosis for power distribution systemcomparative studiesIntelligent fault diagnosis for power distribution systemcomparative studies
Intelligent fault diagnosis for power distribution systemcomparative studies
 
Wasserstein gan
Wasserstein ganWasserstein gan
Wasserstein gan
 
Bayesian modelling and computation for Raman spectroscopy
Bayesian modelling and computation for Raman spectroscopyBayesian modelling and computation for Raman spectroscopy
Bayesian modelling and computation for Raman spectroscopy
 
20 Single Source Shorthest Path
20 Single Source Shorthest Path20 Single Source Shorthest Path
20 Single Source Shorthest Path
 
Introducing Zap Q-Learning
Introducing Zap Q-Learning   Introducing Zap Q-Learning
Introducing Zap Q-Learning
 
Chapter 25 aoa
Chapter 25 aoaChapter 25 aoa
Chapter 25 aoa
 
New tools from the bandit literature to improve A/B Testing
New tools from the bandit literature to improve A/B TestingNew tools from the bandit literature to improve A/B Testing
New tools from the bandit literature to improve A/B Testing
 
Concentration inequality in Machine Learning
Concentration inequality in Machine LearningConcentration inequality in Machine Learning
Concentration inequality in Machine Learning
 
NP-Completeness - II
NP-Completeness - IINP-Completeness - II
NP-Completeness - II
 
Da36615618
Da36615618Da36615618
Da36615618
 
Minimum spanning tree
Minimum spanning treeMinimum spanning tree
Minimum spanning tree
 

Recently uploaded

Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Sérgio Sacani
 
What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.
moosaasad1975
 
Seminar of U.V. Spectroscopy by SAMIR PANDA
 Seminar of U.V. Spectroscopy by SAMIR PANDA Seminar of U.V. Spectroscopy by SAMIR PANDA
Seminar of U.V. Spectroscopy by SAMIR PANDA
SAMIR PANDA
 
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Ana Luísa Pinho
 
Unveiling the Energy Potential of Marshmallow Deposits.pdf
Unveiling the Energy Potential of Marshmallow Deposits.pdfUnveiling the Energy Potential of Marshmallow Deposits.pdf
Unveiling the Energy Potential of Marshmallow Deposits.pdf
Erdal Coalmaker
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
kejapriya1
 
S.1 chemistry scheme term 2 for ordinary level
S.1 chemistry scheme term 2 for ordinary levelS.1 chemistry scheme term 2 for ordinary level
S.1 chemistry scheme term 2 for ordinary level
ronaldlakony0
 
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Studia Poinsotiana
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
yqqaatn0
 
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
Wasswaderrick3
 
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATIONPRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
ChetanK57
 
GBSN- Microbiology (Lab 3) Gram Staining
GBSN- Microbiology (Lab 3) Gram StainingGBSN- Microbiology (Lab 3) Gram Staining
GBSN- Microbiology (Lab 3) Gram Staining
Areesha Ahmad
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
David Osipyan
 
Chapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisisChapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisis
tonzsalvador2222
 
BLOOD AND BLOOD COMPONENT- introduction to blood physiology
BLOOD AND BLOOD COMPONENT- introduction to blood physiologyBLOOD AND BLOOD COMPONENT- introduction to blood physiology
BLOOD AND BLOOD COMPONENT- introduction to blood physiology
NoelManyise1
 
Hemostasis_importance& clinical significance.pptx
Hemostasis_importance& clinical significance.pptxHemostasis_importance& clinical significance.pptx
Hemostasis_importance& clinical significance.pptx
muralinath2
 
In silico drugs analogue design: novobiocin analogues.pptx
In silico drugs analogue design: novobiocin analogues.pptxIn silico drugs analogue design: novobiocin analogues.pptx
In silico drugs analogue design: novobiocin analogues.pptx
AlaminAfendy1
 
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
University of Maribor
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
University of Maribor
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
University of Rennes, INSA Rennes, Inria/IRISA, CNRS
 

Recently uploaded (20)

Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
Observation of Io’s Resurfacing via Plume Deposition Using Ground-based Adapt...
 
What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.What is greenhouse gasses and how many gasses are there to affect the Earth.
What is greenhouse gasses and how many gasses are there to affect the Earth.
 
Seminar of U.V. Spectroscopy by SAMIR PANDA
 Seminar of U.V. Spectroscopy by SAMIR PANDA Seminar of U.V. Spectroscopy by SAMIR PANDA
Seminar of U.V. Spectroscopy by SAMIR PANDA
 
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
Deep Behavioral Phenotyping in Systems Neuroscience for Functional Atlasing a...
 
Unveiling the Energy Potential of Marshmallow Deposits.pdf
Unveiling the Energy Potential of Marshmallow Deposits.pdfUnveiling the Energy Potential of Marshmallow Deposits.pdf
Unveiling the Energy Potential of Marshmallow Deposits.pdf
 
bordetella pertussis.................................ppt
bordetella pertussis.................................pptbordetella pertussis.................................ppt
bordetella pertussis.................................ppt
 
S.1 chemistry scheme term 2 for ordinary level
S.1 chemistry scheme term 2 for ordinary levelS.1 chemistry scheme term 2 for ordinary level
S.1 chemistry scheme term 2 for ordinary level
 
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
Salas, V. (2024) "John of St. Thomas (Poinsot) on the Science of Sacred Theol...
 
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
原版制作(carleton毕业证书)卡尔顿大学毕业证硕士文凭原版一模一样
 
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
DERIVATION OF MODIFIED BERNOULLI EQUATION WITH VISCOUS EFFECTS AND TERMINAL V...
 
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATIONPRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
PRESENTATION ABOUT PRINCIPLE OF COSMATIC EVALUATION
 
GBSN- Microbiology (Lab 3) Gram Staining
GBSN- Microbiology (Lab 3) Gram StainingGBSN- Microbiology (Lab 3) Gram Staining
GBSN- Microbiology (Lab 3) Gram Staining
 
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
3D Hybrid PIC simulation of the plasma expansion (ISSS-14)
 
Chapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisisChapter 12 - climate change and the energy crisis
Chapter 12 - climate change and the energy crisis
 
BLOOD AND BLOOD COMPONENT- introduction to blood physiology
BLOOD AND BLOOD COMPONENT- introduction to blood physiologyBLOOD AND BLOOD COMPONENT- introduction to blood physiology
BLOOD AND BLOOD COMPONENT- introduction to blood physiology
 
Hemostasis_importance& clinical significance.pptx
Hemostasis_importance& clinical significance.pptxHemostasis_importance& clinical significance.pptx
Hemostasis_importance& clinical significance.pptx
 
In silico drugs analogue design: novobiocin analogues.pptx
In silico drugs analogue design: novobiocin analogues.pptxIn silico drugs analogue design: novobiocin analogues.pptx
In silico drugs analogue design: novobiocin analogues.pptx
 
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
Remote Sensing and Computational, Evolutionary, Supercomputing, and Intellige...
 
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
Comparing Evolved Extractive Text Summary Scores of Bidirectional Encoder Rep...
 
Deep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless ReproducibilityDeep Software Variability and Frictionless Reproducibility
Deep Software Variability and Frictionless Reproducibility
 

Wasserstein GAN