Robustness of Conditional GANs to Noisy Labels Spotlight - - PowerPoint PPT Presentation

▶
robustness of conditional gans to noisy labels
SMART_READER_LITE
LIVE PREVIEW

Robustness of Conditional GANs to Noisy Labels Spotlight - - PowerPoint PPT Presentation

Sep 02, 2022 548 likes •770 views

Robustness of Conditional GANs to Noisy Labels Spotlight presentation, NeurIPS 2018 Kiran K. Thekumparampil 1 Ashish Khetan 1 Zinan Lin 2 Sewoong Oh 1 1 University of Illinois at Urbana-Champaign 2 Carnegie Mellon University Poster #5, Tue, Dec 4

slide-1
SLIDE 1

Robustness of Conditional GANs to Noisy Labels

Spotlight presentation, NeurIPS 2018 Kiran K. Thekumparampil1 Ashish Khetan1 Zinan Lin2 Sewoong Oh1

1University of Illinois at Urbana-Champaign 2Carnegie Mellon University

Poster #5, Tue, Dec 4 2018

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 1 / 14

slide-2
SLIDE 2

Conditional GAN (cGAN) is vital for achieving high quality

Input: Labeled real samples (X, Y ) Output: Fake samples for label Y

cGAN “Cat” Latent Code

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 2 / 14

slide-3
SLIDE 3

Conditional GAN (cGAN) is vital for achieving high quality

Input: Labeled real samples (X, Y ) Output: Fake samples for label Y

cGAN “Cat” Latent Code

[Brock et al. 2018] Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 2 / 14

slide-4
SLIDE 4

Conditional GAN (cGAN) is vital for achieving high quality

Input: Labeled real samples (X, Y ) Output: Fake samples for label Y

cGAN “Cat” Latent Code

[Brock et al. 2018]

Visual quality: cGAN >> GAN

[https://github.com/tensorflow/models/tree/master/research/gan] Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 2 / 14

slide-5
SLIDE 5

Conditional GAN is sensitive to noise in labels

cGAN trained with noisy labels produces samples that are biased, generating examples from wrong classes, and,

  • f lower quality (red boxes).

noisy real data label

1 2 3 4 5 6 7 8 9

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 3 / 14

slide-6
SLIDE 6

Conditional GAN is sensitive to noise in labels

cGAN trained with noisy labels produces samples that are biased, generating examples from wrong classes, and,

  • f lower quality (red boxes).

noisy real data label

1 2 3 4 5 6 7 8 9

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 3 / 14

slide-7
SLIDE 7

Conditional GAN is sensitive to noise in labels

cGAN trained with noisy labels produces samples that are biased, generating examples from wrong classes, and,

  • f lower quality (red boxes).

noisy real data label

1 2 3 4 5 6 7 8 9

standard cGAN

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 3 / 14

slide-8
SLIDE 8

Conditional GAN is sensitive to noise in labels

cGAN trained with noisy labels produces samples that are biased, generating examples from wrong classes, and,

  • f lower quality (red boxes).

noisy real data label

1 2 3 4 5 6 7 8 9

standard cGAN

  • ur RCGAN

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 3 / 14

slide-9
SLIDE 9

Conditional GAN (cGAN)

G D z x ˜ yreal xreal

adversarial loss

EIGHT EIGHT

y

EIGHT

y

P Q minQ JS(P || Q)

[Bora et al. 2018, Miyato et al. 2018, Sukhbaatar et al. 2015]

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 4 / 14

slide-10
SLIDE 10

Conditional GAN under noisy labeled data

G D z y x ˜ yreal xreal

adversarial loss

C ˜ yreal

EIGHT

y

EIGHT EIGHT

  • P

Q minQ JS( P || Q)

[Bora et al. 2018, Miyato et al. 2018, Sukhbaatar et al. 2015]

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 5 / 14

slide-11
SLIDE 11

Robust Conditional GAN (RCGAN) Architecture

EIGHT

G D z y x ˜ yreal xreal

adversarial loss

C ˜ yreal C ˜ yreal

EIGHT

  • P
  • Q

Projection Discriminator

minQ JS( P || Q)

[Bora et al. 2018, Miyato et al. 2018, Sukhbaatar et al. 2015]

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 6 / 14

slide-12
SLIDE 12

Minimizing noisy divergence minimizes true divergence

Let P & Q be the noisy labeled versions of P & Q.

Theorem 1 (Population-level Analysis)

TV

  • P,

Q

  • ≤ TV (P, Q) ≤ MC TV
  • P,

Q

  • JS
  • P
  • Q
  • ≤ JS(P Q) ≤ MC
  • 8 JS
  • P
  • Q

    = ⇒ Q = P ⇒ Q = P where TV: Total Variation, JS: Jensen-Shannon divergence and MC maxi

  • j
  • (C −1)ij
  • .

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 7 / 14

slide-13
SLIDE 13

Minimizing noisy divergence minimizes true divergence

Let P & Q be the noisy labeled versions of P & Q.

Theorem 1 (Population-level Analysis)

TV

  • P,

Q

  • ≤ TV (P, Q) ≤ MC TV
  • P,

Q

  • JS
  • P
  • Q
  • ≤ JS(P Q) ≤ MC
  • 8 JS
  • P
  • Q

    = ⇒ Q = P ⇒ Q = P where TV: Total Variation, JS: Jensen-Shannon divergence and MC maxi

  • j
  • (C −1)ij
  • .

Neural Network Distance (dF) w.r.t a class of parametric discriminator functions F is known to generalize [Arora et al. 2017]

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 7 / 14

slide-14
SLIDE 14

Minimizing noisy divergence minimizes true divergence

Let Pn & Qn be the empirical noisy real and generated distributions.

Theorem 2 (Finite Sample Analysis)

If F satisfies inclusion condition, then ∃c > 0 such that dF( Pn, Qn) − ǫ ≤ dF(P, Q) ≤ MC

  • dF(

Pn, Qn) + ǫ

  • with probability at least 1 − e−p for any ε > 0 and n ≥ cp log (pL/ǫ) /ǫ2

when F is L-Lipschitz in p parameters Projection Discriminator satisfies inclusion condition

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 8 / 14

slide-15
SLIDE 15

RCGAN generates correct class (MNIST)

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.0 0.2 0.4 0.6 0.8 1.0

Noise Level Generator Label Accuracy − → cGAN

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 9 / 14

slide-16
SLIDE 16

RCGAN generates correct class (MNIST)

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.0 0.2 0.4 0.6 0.8 1.0

Noise Level Generator Label Accuracy − → cGAN − → RCGAN

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 10 / 14

slide-17
SLIDE 17

RCGAN generates correct class (MNIST)

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.0 0.2 0.4 0.6 0.8 1.0

Noise Level Generator Label Accuracy − → cGAN − → RCGAN − → RCGAN-U

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 11 / 14

slide-18
SLIDE 18

RCGAN improves quality of samples (CIFAR-10)

0.0 0.2 0.4 0.6 0.8 7.5 7.6 7.7 7.8 7.9 8.0 8.1 8.2

Noise Level Inception Score − → cGAN − → RCGAN − → RCGAN-U

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 12 / 14

slide-19
SLIDE 19

RCGAN can correct noisy training labels (MNIST)

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.0 0.2 0.4 0.6 0.8 1.0

Noise Level Label Recovery Accuracy − → cGAN − → RCGAN − → RCGAN-U

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 13 / 14

slide-20
SLIDE 20

Thank you

Poster #5, Tue, Dec 04

https://github.com/POLane16/Robust-Conditional-GAN

[Arora 2015] S. Arora, R. Ge, Y. Liang, T. Ma, and Y. Zhang. Generalization and equilibrium in generative adversarial nets (GANs), ICML 2018. [Bora 2018] A. Bora, E. Price, and A. G. Dimakis. AmbientGAN: Generative models from lossy measurements, ICLR, 2018. [Brock 2018] A. Brock, J. Donahue, and K. Simonyan. Large scale gan training for high fidelity natural image synthesis, arXiv preprint arXiv:1809.11096. [Miyato 2018] T. Miyato, and M. Koyama. cGANs with projection discriminator. ICLR, 2018. [Sukhbaatar 2015] S. Sukhbaatar, J. Bruna, M. Paluri, L. Bourdev, and R. Fergus. Training convolutional networks with noisy labels. In ICLR, Workshop, 2015.

Thekumparampil (UIUC) Robust Conditional GAN NeurIPS 2018 14 / 14