This PDF 1.5 document has been generated by LaTeX with hyperref package / pdfTeX-1.40.16, and has been sent on pdf-archive.com on 19/08/2017 at 02:13, from IP address 173.247.x.x.
The current document download page has been viewed 244 times.
File size: 2.98 MB (5 pages).
Privacy: public file
Montreal AI Sympossium Supplementary
Materials
August 18, 2017
1
GAN cost functions
1.1
Vanilla GAN
Cost Function
E
[log D(x)] +
1.2
E
[log(1 − D(G(z)))]
z∼pz (z)
x∼pdata (x)
Wasserstein GAN
Earth-Mover/Wasserstein Distance
W (Pr , Pg ) =
inf
E
[||x − y||]
γ∈Π(Pr ,Pg ) (x,y)∼γ
where Π(Pr , Pg ) is the set of all joint distributions whose marginals are Pr and
Pg .
Cost Function
sup E [D(x)] − E [D(x)]
D x∼Pr
x∼Pθ
where D is some discriminator that is constrained to be K-Lipschitz and Pθ is
the distribution of a generator parameterized by θ.
1.3
Consensus Optimization
Consensus Term
1
||v(x)||2
2
∇f
v(x) =
∇g
L(x) =
where
where f and g are the utility functions of the generator and discriminator in
vanilla GANs, respectively.
1
New Utility Functions
fe = f − γL
ge = g − γL
where fe and ge are the new utility functions under the Consensus Optimization
regime.
2
2
Figures
0.6
0.4
0.3
0.4
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0.6
0.4
0.3
0.4
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0.6
0.5
0.4
0.4
0.3
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0.6
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00
−0.05
−0.10
0.4
0.2
0.0
−0.2
0.6
0.5
0.4
0.4
0.3
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0.6
0.4
0.3
0.4
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0.6
0.4
0.3
0.4
0.2
0.2
0.1
0.0
0.0
−0.2
−0.1
0.6
0.4
0.3
0.4
0.2
0.2
0.1
0.0
0.0
−0.1
−0.2
−0.2
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
0
50 100150200 0
(a) Iteration 10000
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(b) Iteration 20000
0.5
0.5
0.4
0.3
0.0
0.2
−0.5
0.1
0.0
−1.0
−0.1
0.6
0.5
0.4
0.3
0.2
0.1
0.0
−0.1
−0.2
0.5
0.0
−0.5
−1.0
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
−0.1
0.5
0.0
−0.5
−1.0
0.4
0.5
0.3
0.0
0.2
0.1
−0.5
0.0
−1.0
−0.1
0.5
0.5
0.4
0.3
0.0
0.2
−0.5
0.1
0.0
−1.0
−0.1
0.6
0.5
0.4
0.3
0.2
0.1
0.0
−0.1
−0.2
0.5
0.0
−0.5
−1.0
0.6
0.5
0.5
0.4
0.0
0.3
0.2
−0.5
0.1
0.0
−1.0
−0.1
0.6
0.5
0.4
0.3
0.2
0.1
0.0
−0.1
−0.2
0.5
0.0
−0.5
−1.0
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
0
(c) Iteration 30000
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(d) Iteration 40000
Figure 1: Illustration of Mode Collapse: Each image is a random sample of 64
different walk cycles from the distribution generated by a vanilla GAN, from top
left around clockwise each image is 10000 iterations further into training than
the last. One can see that during training the GAN is jumping from mode to
mode rather than learning to spread out probability mass across the modes.
3
1.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
1.0
0.5
0.5
0.0
0.0
−0.5
−0.5
−1.0
−1.0
−1.5
−1.5
−2.0
−2.0
−2.5
−2.5
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
−3.5
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
1.0
0.5
0.5
0.0
0.0
−0.5
−0.5
−1.0
−1.0
−1.5
−1.5
−2.0
−2.0
−2.5
−2.5
1.0
1.0
0.5
0.5
0.0
0.0
−0.5
−0.5
−1.0
−1.0
−1.5
−1.5
−2.0
−2.0
−2.5
1.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(a) Vanilla GAN sample
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(b) Consensus Optimized GAN sample
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
−3.5
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
0
50
100
150
200
(d) Conditional GAN sample
(c) Wasserstein GAN sample
4
0.8
1.0
0.6
0.5
0.4
0.0
0.2
−0.5
0.0
−1.0
−0.2
−1.5
−0.4
−2.0
−0.6
−2.5
0.8
1.0
0.6
0.5
0.4
0.0
0.2
−0.5
0.0
−1.0
−0.2
−1.5
−0.4
−2.0
−0.6
−2.5
0.8
1.0
0.6
0.5
0.4
0.0
0.2
−0.5
0.0
−1.0
−0.2
−1.5
−0.4
−2.0
−0.6
−2.5
0.8
1.0
0.6
0.5
0.4
0.0
0.2
−0.5
0.0
−1.0
−0.2
−0.4
−1.5
−0.6
−2.0
0.8
1.0
0.6
0.5
0.4
0.0
0.2
−0.5
0.0
−1.0
−0.2
−1.5
−0.4
−2.0
−0.6
−2.5
0.8
4
0.6
2
0.4
0
0.2
−2
0.0
−4
−0.2
−6
−0.4
−8
−0.6
−10
0.8
4
0.6
2
0.4
0
0.2
−2
0.0
−4
−0.2
−6
−0.4
−8
−0.6
−10
0.8
20
0.6
15
0.4
10
0.2
5
0.0
0
−0.2
−5
−0.4
−10
−0.6
−15
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
0
50 100150200 0
(e) Recurrent GAN sample
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(f) AdaGAN sample
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
1.0
0.5
0.0
−0.5
−1.0
−1.5
−2.0
−2.5
0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200 0
50 100150200
(g) GMM sample
(h) Real sample
Figure 2: Random Samples from generated and real distributions. Random
samples from six different GAN techniques, a GMM baseline, and the real
distribution.
5
supp.pdf (PDF, 2.98 MB)
Use the permanent link to the download page to share your document on Facebook, Twitter, LinkedIn, or directly with a contact by e-Mail, Messenger, Whatsapp, Line..
Use the short link to share your document on Twitter or by text message (SMS)
Copy the following HTML code to share your document on a Website or Blog