SlideShare a Scribd company logo

Deep Convolutional GANs - meaning of latent space

Hansol Kang
Hansol Kang

DCGAN은 GAN에 단순히 conv net을 적용했을 뿐만 아니라, latent space에서도 의미를 찾음. DCGAN 논문 리뷰 및 PyTorch 기반의 구현. VAE 세미나 이슈 사항에 대한 리뷰. my github : https://github.com/messy-snail/GAN_PyTorch [참고] https://github.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN https://github.com/taeoh-kim/Pytorch_DCGAN Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015).

1 of 43
Downloaded 106 times
Deep Convolutional GANs ISL Lab Seminar Hansol Kang : Meaning of Latent Space
Contents Review of GAN DCGAN Experiment Summary 2018-10-05 2
Review of GAN 2018-10-05 3 • Adversarial nets 1) Global Optimality of datag pp  2) Convergence of Algorithm D GVs x )(xpdata “Generative Adversarial Networks” Goal Method
D DCGAN 2018-10-05 4 • Introduction * Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015). * “I have the strongest MLP army.” “I have too.” G
Deep Convolutional GANs - meaning of latent space
DCGAN 2018-10-05 6 • Introduction * D G “What are they doing?” “We have a better CNN than MLP” D “I have the strongest MLP army.” “I have too.” G Vanilla GAN DCGAN * Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015).
DCGAN 2018-10-05 7 • Contributions Generating Natural Image Deep Convolutional GANs Image Classification using D Filter Visualization Vector arithmetic properties
Z“I’m very Important” Who am I? Black box Real D A B C DCGAN 2018-10-05 8 • Contributions Generating Natural Image Deep Convolutional GANs Image Classification using D Filter Visualization Vector arithmetic properties
DCGAN 2018-10-05 9 • Approach and Model Architecture Replace any pooling layers with strided convolutions (discriminator) and fractional- strided convolutions (generator). Use batchnorm in both the generator and the discriminator. Remove fully connected hidden layers for deeper architectures. Use ReLU activation in generator for all layers except for the output, which uses Tanh. Use LeakyReLU activation in the discriminator for all layers.
DCGAN 2018-10-05 10 • Approach and Model Architecture Strided Convolution Fractional Convolution(Transposed Convolution)
DCGAN 2018-10-05 11 • Approach and Model Architecture Batch Normalization Except for these layers. Output layer of Generator Input layer of Discriminator
DCGAN 2018-10-05 12 • Approach and Model Architecture No fully connected layer Classical CNN GAP(Global Average Pooling) http://nmhkahn.github.io/Casestudy-CNN
DCGAN 2018-10-05 13 • Approach and Model Architecture No fully connected layer https://raw.githubusercontent.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN/master/pytorch_DCGAN.png
DCGAN 2018-10-05 14 • Approach and Model Architecture ReLU, Tanh, LeakyReLU http://gmelli.org/RKB/Rectified_Linear_Unit_(ReLU)_Activation_Function Generator : ReLU, Tanh Discriminator : LeakyReLu , Sigmoid
DCGAN 2018-10-05 15 • Details of Adversarial Training • Mini-batch stochastic gradient descent(SGD); mini-batch size of 128 • All weights initialized from a zero-centered Normal distribution with standard deviation 0.02 • Leaky slope 0.2 • Adam optimizer; lr =0.0002, beta1 = 0.9, beta2 = 0.5
DCGAN 2018-10-05 16 • Details of Adversarial Training LSUN dataset 1 epoch
DCGAN 2018-10-05 17 • Details of Adversarial Training LSUN dataset 5 epochs
DCGAN 2018-10-05 18 • Empirical Validation of DCGANs Capabilities • CIFAR-10 • Classification • Domain robustness
DCGAN 2018-10-05 19 • Empirical Validation of DCGANs Capabilities SVHN(Street View House Numbers) dataset
DCGAN 2018-10-05 20 • Investigating and Visualizing The Internals of The Networks Walking in the latent space
DCGAN 2018-10-05 21 • Investigating and Visualizing The Internals of The Networks(cont.) Visualizing the discriminator features
DCGAN 2018-10-05 22 • Investigating and Visualizing The Internals of The Networks(cont.) Forgetting to draw certain objects in charge of windows in charge of beds in charge of lamps in charge of doors … Latent code Filters(Conv) Generation 1 0 0 Noise(z)
DCGAN 2018-10-05 23 • Investigating and Visualizing The Internals of The Networks(cont.) Forgetting to draw certain objects
DCGAN 2018-10-05 24 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
DCGAN 2018-10-05 25 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
DCGAN 2018-10-05 26 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
DCGAN 2018-10-05 27 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
Experiment • Code 2018-10-05 28https://github.com/messy-snail/GAN_PyTorch
Experiment • Code 2018-10-05 29https://github.com/messy-snail/GAN_PyTorch
Experiment • Results#1 CelebA 2018-10-05 30 Ground Truth Vanilla GAN : DCGAN : Epoch 1 Epoch 5 Epoch 100 Epoch 1 Epoch 5 Epoch 30 Still have this sample Results are cherry picked
Experiment • Results#2 LSUN) 2018-10-05 31 Ground Truth Vanilla GAN : DCGAN : Epoch 1 Epoch 5 Epoch 12 Epoch 1 Epoch 2 Epoch 5 Results are cherry picked
Experiment • Results#3 Korean Idol – Transfer trial 2018-10-05 32 • I used weights and biases generated by celebA learning. • I wanted the effect of transfer learning but failed. Maybe these factors (Asian, cropping image) Ground Truth Epoch 1 Epoch 2 Epoch 3 Epoch 4 Epoch 5 Epoch 6
Experiment • Results#4 Korean Idol 2018-10-05 33 Ground Truth Epoch 1 Epoch 5 Epoch 30 Epoch 50 Epoch 100 Epoch 150 • 10000 images Insufficient data set
Summary 2018-10-05 34 • Stable set of architectures for training generative adversarial networks • Good representations of images for supervised learning and generative modeling • Sometimes collapse a subset of filters to a single oscillating mode • Latent code has a special meaning, not a simple noise component. [Instability of GAN]
Future work 2018-10-05 35 Paper Review Vanilla GAN DCGAN InfoGAN Unrolled GAN Wasserstein GAN LS GAN BEGAN Pix2Pix Cycle GAN Proposed Model SpyGAN Tips Document Programming Mathematical Study Information theory (working title)
&
Appendix • Issues at the VAE Seminar (18.07.23) 2018-10-05 37  Issue#1 Performance of VAE and GAN  Issue#2 Log likelihood  Issue#3 Dimension of latent code  Issue#4 Why manifold? Durk Kingma 1. Adam: A Method for Stochastic Optimization 2. Auto-Encoding Variational Bayes Machine Learning researcher at OpenAI - Mathematically very difficult papers Intuitive explanation https://www.youtube.com/watch?v=o_peo6U7IRM 오토인코더의 모든 것 : I refer to this video
Appendix • Issue #1 Performance of VAE and GAN 2018-10-05 38 “Compared to GAN, VAE is relatively blurred and I do not know why.” “Cost function”     )(||)|()(|log),,( )|( zpxzqKLzgxExL xzq    ),,(min xL VAE ),(maxmin DGV DG       zGDExDEDGV zdata pzpx  1log)(log),( ~~ GAN Intuition Reconstruction Error Regularization ≈ D Loss ≈ G Loss
Appendix • Issue #1 Performance of VAE and GAN 2018-10-05 39 VAE Loss= Recon. Error + Regularization GAN Loss= G_Loss + D_Loss E D Recon. Error D Real Fake 1. Optimize 2. Image Quality 3. Generalization VAE vs. GAN
Appendix • Issue #2 Log likelihood 2018-10-05 40 Question about log likelihood “Summation and monotonically increasing” MLE(Maximum Likelihood Estimation) : Unknown parameter estimation from observation )|(maxargˆ   yp eg. Gaussian Distribution Mean and Std  i iyp )|(maxarg          i i i i ypyp )|(logmaxarg)|(logmaxarg   : monotonically increasing function1 Log(x) cf.  i ixp )(logmaxarg   Generation model
Appendix • Issue #3 Dimension of latent code 2018-10-05 41 “Is the latent code dimension always small?” “Yes” AE, What’s this? Dimension reduction E D High Low Interested Sparse AE FAILED
Appendix • Issue #4 Why manifold? 2018-10-05 42 What’s the manifold and Why explain the manifold? “Concept of manifold and Difference of between AE and VAE” High Low Subspace =Manifold Concept of manifold D Purpose of AE : Manifold Learning Purpose of AE and VAE Assumption(manifold hypothesis) Uniform sampling E Unsupervised Learning D Purpose of VAE : Generative Model E Unsupervised Learning : Correlation between generation and manifold…
Appendix • PyTorch (Variable length inputs) 2018-10-05 43 Shape = {Size} torch.Size([128, 3, 32, 32]) Shape = {Size} torch.Size([128, 64, 16, 16]) Shape = {Size} torch.Size([128, 16384])3x32x32 CIFAR-10 Shape = {Size} torch.Size([128, 64, 109, 89]) Shape = {Size} torch.Size([128, 3, 218, 178]) Shape = {Size} torch.Size([128, 620864]) 3x178x218 CelebA Conv Input Pool FC Conv2d(in_ch, out_ch, k_size, s, p) Reshape(bat_sz,-1) Input size is not fixed.
Ad

Recommended

Generative Adversarial Networks
Generative Adversarial NetworksGenerative Adversarial Networks
Generative Adversarial Networks
Mustafa Yagmur
 
1시간만에 GAN(Generative Adversarial Network) 완전 정복하기
1시간만에 GAN(Generative Adversarial Network) 완전 정복하기1시간만에 GAN(Generative Adversarial Network) 완전 정복하기
1시간만에 GAN(Generative Adversarial Network) 완전 정복하기
NAVER Engineering
 
Dcgan
DcganDcgan
Dcgan
Brian Kim
 
Style gan
Style ganStyle gan
Style gan
哲东 郑
 
Introduction to Generative Adversarial Networks (GANs)
Introduction to Generative Adversarial Networks (GANs)Introduction to Generative Adversarial Networks (GANs)
Introduction to Generative Adversarial Networks (GANs)
Appsilon Data Science
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
Yunjey Choi
 
Introduction to Diffusion Models
Introduction to Diffusion ModelsIntroduction to Diffusion Models
Introduction to Diffusion Models
Sangwoo Mo
 
Generative Adversarial Networks
Generative Adversarial NetworksGenerative Adversarial Networks
Generative Adversarial Networks
Mark Chang
 
Score based generative model
Score based generative modelScore based generative model
Score based generative model
sangyun lee
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
남주 김
 
Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)
Manohar Mukku
 
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
 
A Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial NetworksA Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial Networks
Jong Wook Kim
 
3D Gaussian Splatting
3D Gaussian Splatting3D Gaussian Splatting
3D Gaussian Splatting
taeseon ryu
 
Basic Generative Adversarial Networks
Basic Generative Adversarial NetworksBasic Generative Adversarial Networks
Basic Generative Adversarial Networks
Dong Heon Cho
 
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAIGenerative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
WithTheBest
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
Ding Li
 
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
Deep Learning JP
 
Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...
Universitat Politècnica de Catalunya
 
그림 그리는 AI
그림 그리는 AI그림 그리는 AI
그림 그리는 AI
NAVER Engineering
 
오토인코더의 모든 것
오토인코더의 모든 것오토인코더의 모든 것
오토인코더의 모든 것
NAVER Engineering
 
GAN - Theory and Applications
GAN - Theory and ApplicationsGAN - Theory and Applications
GAN - Theory and Applications
Emanuele Ghelfi
 
Generative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variantsGenerative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variants
ananth
 
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual RepresentationsPR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
Jinwon Lee
 
CNN Machine learning DeepLearning
CNN Machine learning DeepLearningCNN Machine learning DeepLearning
CNN Machine learning DeepLearning
Abhishek Sharma
 
モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019
Yusuke Uchida
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
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
 
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
MLconf
 
IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」
Preferred Networks
 
Ad

More Related Content

What's hot (20)

Score based generative model
Score based generative modelScore based generative model
Score based generative model
sangyun lee
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
남주 김
 
Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)
Manohar Mukku
 
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
 
A Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial NetworksA Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial Networks
Jong Wook Kim
 
3D Gaussian Splatting
3D Gaussian Splatting3D Gaussian Splatting
3D Gaussian Splatting
taeseon ryu
 
Basic Generative Adversarial Networks
Basic Generative Adversarial NetworksBasic Generative Adversarial Networks
Basic Generative Adversarial Networks
Dong Heon Cho
 
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAIGenerative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
WithTheBest
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
Ding Li
 
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
Deep Learning JP
 
Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...
Universitat Politècnica de Catalunya
 
그림 그리는 AI
그림 그리는 AI그림 그리는 AI
그림 그리는 AI
NAVER Engineering
 
오토인코더의 모든 것
오토인코더의 모든 것오토인코더의 모든 것
오토인코더의 모든 것
NAVER Engineering
 
GAN - Theory and Applications
GAN - Theory and ApplicationsGAN - Theory and Applications
GAN - Theory and Applications
Emanuele Ghelfi
 
Generative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variantsGenerative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variants
ananth
 
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual RepresentationsPR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
Jinwon Lee
 
CNN Machine learning DeepLearning
CNN Machine learning DeepLearningCNN Machine learning DeepLearning
CNN Machine learning DeepLearning
Abhishek Sharma
 
モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019
Yusuke Uchida
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
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
 
Score based generative model
Score based generative modelScore based generative model
Score based generative model
sangyun lee
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
남주 김
 
Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)Generative Adversarial Networks (GAN)
Generative Adversarial Networks (GAN)
Manohar Mukku
 
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
 
A Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial NetworksA Short Introduction to Generative Adversarial Networks
A Short Introduction to Generative Adversarial Networks
Jong Wook Kim
 
3D Gaussian Splatting
3D Gaussian Splatting3D Gaussian Splatting
3D Gaussian Splatting
taeseon ryu
 
Basic Generative Adversarial Networks
Basic Generative Adversarial NetworksBasic Generative Adversarial Networks
Basic Generative Adversarial Networks
Dong Heon Cho
 
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAIGenerative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
Generative Adversarial Networks (GANs) - Ian Goodfellow, OpenAI
WithTheBest
 
Generative adversarial networks
Generative adversarial networksGenerative adversarial networks
Generative adversarial networks
Ding Li
 
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing[DLHacks]StyleGANとBigGANのStyle mixing, morphing
[DLHacks]StyleGANとBigGANのStyle mixing, morphing
Deep Learning JP
 
Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...Deep Learning for Computer Vision: Generative models and adversarial training...
Deep Learning for Computer Vision: Generative models and adversarial training...
Universitat Politècnica de Catalunya
 
그림 그리는 AI
그림 그리는 AI그림 그리는 AI
그림 그리는 AI
NAVER Engineering
 
오토인코더의 모든 것
오토인코더의 모든 것오토인코더의 모든 것
오토인코더의 모든 것
NAVER Engineering
 
GAN - Theory and Applications
GAN - Theory and ApplicationsGAN - Theory and Applications
GAN - Theory and Applications
Emanuele Ghelfi
 
Generative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variantsGenerative Adversarial Networks : Basic architecture and variants
Generative Adversarial Networks : Basic architecture and variants
ananth
 
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual RepresentationsPR-231: A Simple Framework for Contrastive Learning of Visual Representations
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
Jinwon Lee
 
CNN Machine learning DeepLearning
CNN Machine learning DeepLearningCNN Machine learning DeepLearning
CNN Machine learning DeepLearning
Abhishek Sharma
 
モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019モデルアーキテクチャ観点からの高速化2019
モデルアーキテクチャ観点からの高速化2019
Yusuke Uchida
 
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Convolutional Neural Network - CNN | How CNN Works | Deep Learning Course | S...
Simplilearn
 
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
 

Similar to Deep Convolutional GANs - meaning of latent space (20)

Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
MLconf
 
IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」
Preferred Networks
 
Deep Learning in Computer Vision
Deep Learning in Computer VisionDeep Learning in Computer Vision
Deep Learning in Computer Vision
Sungjoon Choi
 
# Can we trust ai. the dilemma of model adjustment
# Can we trust ai. the dilemma of model adjustment# Can we trust ai. the dilemma of model adjustment
# Can we trust ai. the dilemma of model adjustment
Terence Huang
 
Implementation of linear regression and logistic regression on Spark
Implementation of linear regression and logistic regression on SparkImplementation of linear regression and logistic regression on Spark
Implementation of linear regression and logistic regression on Spark
Dalei Li
 
Introduction to Genetic algorithm and its significance in VLSI design and aut...
Introduction to Genetic algorithm and its significance in VLSI design and aut...Introduction to Genetic algorithm and its significance in VLSI design and aut...
Introduction to Genetic algorithm and its significance in VLSI design and aut...
Centre for Electronics, Computer, Self development
 
Using Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning ModelsUsing Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning Models
SigOpt
 
Using Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning ModelsUsing Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning Models
Scott Clark
 
MLSEV. Logistic Regression, Deepnets, and Time Series
MLSEV. Logistic Regression, Deepnets, and Time Series MLSEV. Logistic Regression, Deepnets, and Time Series
MLSEV. Logistic Regression, Deepnets, and Time Series
BigML, Inc
 
BSSML17 - Deepnets
BSSML17 - DeepnetsBSSML17 - Deepnets
BSSML17 - Deepnets
BigML, Inc
 
consistency regularization for generative adversarial networks_review
consistency regularization for generative adversarial networks_reviewconsistency regularization for generative adversarial networks_review
consistency regularization for generative adversarial networks_review
Yoonho Na
 
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
Universitat Politècnica de Catalunya
 
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
Julián Urbano
 
Intro to Deep Reinforcement Learning
Intro to Deep Reinforcement LearningIntro to Deep Reinforcement Learning
Intro to Deep Reinforcement Learning
Khaled Saleh
 
Gradient Boosted Regression Trees in scikit-learn
Gradient Boosted Regression Trees in scikit-learnGradient Boosted Regression Trees in scikit-learn
Gradient Boosted Regression Trees in scikit-learn
DataRobot
 
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
MLconf
 
Computer vision for transportation
Computer vision for transportationComputer vision for transportation
Computer vision for transportation
Wanjin Yu
 
Computer Vision for Beginners
Computer Vision for BeginnersComputer Vision for Beginners
Computer Vision for Beginners
Sanghamitra Deb
 
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
Deep Learning JP
 
gan.pdf
gan.pdfgan.pdf
gan.pdf
Dr.rukmani Devi
 
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
Hanjun Dai, PhD Student, School of Computational Science and Engineering, Geo...
MLconf
 
IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」IIBMP2019 講演資料「オープンソースで始める深層学習」
IIBMP2019 講演資料「オープンソースで始める深層学習」
Preferred Networks
 
Deep Learning in Computer Vision
Deep Learning in Computer VisionDeep Learning in Computer Vision
Deep Learning in Computer Vision
Sungjoon Choi
 
# Can we trust ai. the dilemma of model adjustment
# Can we trust ai. the dilemma of model adjustment# Can we trust ai. the dilemma of model adjustment
# Can we trust ai. the dilemma of model adjustment
Terence Huang
 
Implementation of linear regression and logistic regression on Spark
Implementation of linear regression and logistic regression on SparkImplementation of linear regression and logistic regression on Spark
Implementation of linear regression and logistic regression on Spark
Dalei Li
 
Introduction to Genetic algorithm and its significance in VLSI design and aut...
Introduction to Genetic algorithm and its significance in VLSI design and aut...Introduction to Genetic algorithm and its significance in VLSI design and aut...
Introduction to Genetic algorithm and its significance in VLSI design and aut...
Centre for Electronics, Computer, Self development
 
Using Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning ModelsUsing Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning Models
SigOpt
 
Using Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning ModelsUsing Bayesian Optimization to Tune Machine Learning Models
Using Bayesian Optimization to Tune Machine Learning Models
Scott Clark
 
MLSEV. Logistic Regression, Deepnets, and Time Series
MLSEV. Logistic Regression, Deepnets, and Time Series MLSEV. Logistic Regression, Deepnets, and Time Series
MLSEV. Logistic Regression, Deepnets, and Time Series
BigML, Inc
 
BSSML17 - Deepnets
BSSML17 - DeepnetsBSSML17 - Deepnets
BSSML17 - Deepnets
BigML, Inc
 
consistency regularization for generative adversarial networks_review
consistency regularization for generative adversarial networks_reviewconsistency regularization for generative adversarial networks_review
consistency regularization for generative adversarial networks_review
Yoonho Na
 
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
D1L5 Visualization (D1L2 Insight@DCU Machine Learning Workshop 2017)
Universitat Politècnica de Catalunya
 
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
How Do Gain and Discount Functions Affect the Correlation between DCG and Use...
Julián Urbano
 
Intro to Deep Reinforcement Learning
Intro to Deep Reinforcement LearningIntro to Deep Reinforcement Learning
Intro to Deep Reinforcement Learning
Khaled Saleh
 
Gradient Boosted Regression Trees in scikit-learn
Gradient Boosted Regression Trees in scikit-learnGradient Boosted Regression Trees in scikit-learn
Gradient Boosted Regression Trees in scikit-learn
DataRobot
 
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
Ehtsham Elahi, Senior Research Engineer, Personalization Science and Engineer...
MLconf
 
Computer vision for transportation
Computer vision for transportationComputer vision for transportation
Computer vision for transportation
Wanjin Yu
 
Computer Vision for Beginners
Computer Vision for BeginnersComputer Vision for Beginners
Computer Vision for Beginners
Sanghamitra Deb
 
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
【DL輪読会】Toward Fast and Stabilized GAN Training for Highfidelity Few-shot Imag...
Deep Learning JP
 
gan.pdf
gan.pdfgan.pdf
gan.pdf
Dr.rukmani Devi
 
Ad

More from Hansol Kang (20)

이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
Hansol Kang
 
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
Support Vector Machine - 기본 이해와 OpenCV 실습.pdfSupport Vector Machine - 기본 이해와 OpenCV 실습.pdf
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
Hansol Kang
 
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
Hansol Kang
 
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
Hansol Kang
 
알아두면 쓸모있는 깃허브 2
알아두면 쓸모있는 깃허브 2알아두면 쓸모있는 깃허브 2
알아두면 쓸모있는 깃허브 2
Hansol Kang
 
알아두면 쓸모있는 깃허브 1
알아두면 쓸모있는 깃허브 1알아두면 쓸모있는 깃허브 1
알아두면 쓸모있는 깃허브 1
Hansol Kang
 
FPN 리뷰
FPN 리뷰FPN 리뷰
FPN 리뷰
Hansol Kang
 
R-FCN 리뷰
R-FCN 리뷰R-FCN 리뷰
R-FCN 리뷰
Hansol Kang
 
basic of deep learning
basic of deep learningbasic of deep learning
basic of deep learning
Hansol Kang
 
파이썬 제대로 활용하기
파이썬 제대로 활용하기파이썬 제대로 활용하기
파이썬 제대로 활용하기
Hansol Kang
 
모던 C++ 정리
모던 C++ 정리모던 C++ 정리
모던 C++ 정리
Hansol Kang
 
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Hansol Kang
 
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
Hansol Kang
 
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
Hansol Kang
 
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
Hansol Kang
 
PyTorch 튜토리얼 (Touch to PyTorch)
PyTorch 튜토리얼 (Touch to PyTorch)PyTorch 튜토리얼 (Touch to PyTorch)
PyTorch 튜토리얼 (Touch to PyTorch)
Hansol Kang
 
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
Hansol Kang
 
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
Hansol Kang
 
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
Hansol Kang
 
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
Hansol Kang
 
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
Hansol Kang
 
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
Support Vector Machine - 기본 이해와 OpenCV 실습.pdfSupport Vector Machine - 기본 이해와 OpenCV 실습.pdf
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
Hansol Kang
 
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
Hansol Kang
 
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
Hansol Kang
 
알아두면 쓸모있는 깃허브 2
알아두면 쓸모있는 깃허브 2알아두면 쓸모있는 깃허브 2
알아두면 쓸모있는 깃허브 2
Hansol Kang
 
알아두면 쓸모있는 깃허브 1
알아두면 쓸모있는 깃허브 1알아두면 쓸모있는 깃허브 1
알아두면 쓸모있는 깃허브 1
Hansol Kang
 
FPN 리뷰
FPN 리뷰FPN 리뷰
FPN 리뷰
Hansol Kang
 
R-FCN 리뷰
R-FCN 리뷰R-FCN 리뷰
R-FCN 리뷰
Hansol Kang
 
basic of deep learning
basic of deep learningbasic of deep learning
basic of deep learning
Hansol Kang
 
파이썬 제대로 활용하기
파이썬 제대로 활용하기파이썬 제대로 활용하기
파이썬 제대로 활용하기
Hansol Kang
 
모던 C++ 정리
모던 C++ 정리모던 C++ 정리
모던 C++ 정리
Hansol Kang
 
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
Hansol Kang
 
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
Hansol Kang
 
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
Hansol Kang
 
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
Hansol Kang
 
PyTorch 튜토리얼 (Touch to PyTorch)
PyTorch 튜토리얼 (Touch to PyTorch)PyTorch 튜토리얼 (Touch to PyTorch)
PyTorch 튜토리얼 (Touch to PyTorch)
Hansol Kang
 
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
Hansol Kang
 
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
Hansol Kang
 
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
Hansol Kang
 
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
Hansol Kang
 
Ad

Recently uploaded (20)

AI and Data Privacy in 2025: Global Trends
AI and Data Privacy in 2025: Global TrendsAI and Data Privacy in 2025: Global Trends
AI and Data Privacy in 2025: Global Trends
InData Labs
 
HCL Nomad Web – Best Practices and Managing Multiuser Environments
HCL Nomad Web – Best Practices and Managing Multiuser EnvironmentsHCL Nomad Web – Best Practices and Managing Multiuser Environments
HCL Nomad Web – Best Practices and Managing Multiuser Environments
panagenda
 
Into The Box Conference Keynote Day 1 (ITB2025)
Into The Box Conference Keynote Day 1 (ITB2025)Into The Box Conference Keynote Day 1 (ITB2025)
Into The Box Conference Keynote Day 1 (ITB2025)
Ortus Solutions, Corp
 
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In France
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In FranceManifest Pre-Seed Update | A Humanoid OEM Deeptech In France
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In France
chb3
 
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdfAre Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Telecoms Supermarket
 
Rusty Waters: Elevating Lakehouses Beyond Spark
Rusty Waters: Elevating Lakehouses Beyond SparkRusty Waters: Elevating Lakehouses Beyond Spark
Rusty Waters: Elevating Lakehouses Beyond Spark
carlyakerly1
 
Dev Dives: Automate and orchestrate your processes with UiPath Maestro
Dev Dives: Automate and orchestrate your processes with UiPath MaestroDev Dives: Automate and orchestrate your processes with UiPath Maestro
Dev Dives: Automate and orchestrate your processes with UiPath Maestro
UiPathCommunity
 
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep DiveDesigning Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
ScyllaDB
 
Splunk Security Update | Public Sector Summit Germany 2025
Splunk Security Update | Public Sector Summit Germany 2025Splunk Security Update | Public Sector Summit Germany 2025
Splunk Security Update | Public Sector Summit Germany 2025
Splunk
 
TrsLabs - Fintech Product & Business Consulting
TrsLabs - Fintech Product & Business ConsultingTrsLabs - Fintech Product & Business Consulting
TrsLabs - Fintech Product & Business Consulting
Trs Labs
 
Vaibhav Gupta BAML: AI work flows without Hallucinations
Vaibhav Gupta BAML: AI work flows without HallucinationsVaibhav Gupta BAML: AI work flows without Hallucinations
Vaibhav Gupta BAML: AI work flows without Hallucinations
john409870
 
Big Data Analytics Quick Research Guide by Arthur Morgan
Big Data Analytics Quick Research Guide by Arthur MorganBig Data Analytics Quick Research Guide by Arthur Morgan
Big Data Analytics Quick Research Guide by Arthur Morgan
Arthur Morgan
 
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptxIncreasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Anoop Ashok
 
Technology Trends in 2025: AI and Big Data Analytics
Technology Trends in 2025: AI and Big Data AnalyticsTechnology Trends in 2025: AI and Big Data Analytics
Technology Trends in 2025: AI and Big Data Analytics
InData Labs
 
ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes Partner Innovation Updates for May 2025ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes
 
Unlocking the Power of IVR: A Comprehensive Guide
Unlocking the Power of IVR: A Comprehensive GuideUnlocking the Power of IVR: A Comprehensive Guide
Unlocking the Power of IVR: A Comprehensive Guide
vikasascentbpo
 
MINDCTI revenue release Quarter 1 2025 PR
MINDCTI revenue release Quarter 1 2025 PRMINDCTI revenue release Quarter 1 2025 PR
MINDCTI revenue release Quarter 1 2025 PR
MIND CTI
 
Quantum Computing Quick Research Guide by Arthur Morgan
Quantum Computing Quick Research Guide by Arthur MorganQuantum Computing Quick Research Guide by Arthur Morgan
Quantum Computing Quick Research Guide by Arthur Morgan
Arthur Morgan
 
Build Your Own Copilot & Agents For Devs
Build Your Own Copilot & Agents For DevsBuild Your Own Copilot & Agents For Devs
Build Your Own Copilot & Agents For Devs
Brian McKeiver
 
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
BookNet Canada
 
AI and Data Privacy in 2025: Global Trends
AI and Data Privacy in 2025: Global TrendsAI and Data Privacy in 2025: Global Trends
AI and Data Privacy in 2025: Global Trends
InData Labs
 
HCL Nomad Web – Best Practices and Managing Multiuser Environments
HCL Nomad Web – Best Practices and Managing Multiuser EnvironmentsHCL Nomad Web – Best Practices and Managing Multiuser Environments
HCL Nomad Web – Best Practices and Managing Multiuser Environments
panagenda
 
Into The Box Conference Keynote Day 1 (ITB2025)
Into The Box Conference Keynote Day 1 (ITB2025)Into The Box Conference Keynote Day 1 (ITB2025)
Into The Box Conference Keynote Day 1 (ITB2025)
Ortus Solutions, Corp
 
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In France
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In FranceManifest Pre-Seed Update | A Humanoid OEM Deeptech In France
Manifest Pre-Seed Update | A Humanoid OEM Deeptech In France
chb3
 
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdfAre Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Are Cloud PBX Providers in India Reliable for Small Businesses (1).pdf
Telecoms Supermarket
 
Rusty Waters: Elevating Lakehouses Beyond Spark
Rusty Waters: Elevating Lakehouses Beyond SparkRusty Waters: Elevating Lakehouses Beyond Spark
Rusty Waters: Elevating Lakehouses Beyond Spark
carlyakerly1
 
Dev Dives: Automate and orchestrate your processes with UiPath Maestro
Dev Dives: Automate and orchestrate your processes with UiPath MaestroDev Dives: Automate and orchestrate your processes with UiPath Maestro
Dev Dives: Automate and orchestrate your processes with UiPath Maestro
UiPathCommunity
 
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep DiveDesigning Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
Designing Low-Latency Systems with Rust and ScyllaDB: An Architectural Deep Dive
ScyllaDB
 
Splunk Security Update | Public Sector Summit Germany 2025
Splunk Security Update | Public Sector Summit Germany 2025Splunk Security Update | Public Sector Summit Germany 2025
Splunk Security Update | Public Sector Summit Germany 2025
Splunk
 
TrsLabs - Fintech Product & Business Consulting
TrsLabs - Fintech Product & Business ConsultingTrsLabs - Fintech Product & Business Consulting
TrsLabs - Fintech Product & Business Consulting
Trs Labs
 
Vaibhav Gupta BAML: AI work flows without Hallucinations
Vaibhav Gupta BAML: AI work flows without HallucinationsVaibhav Gupta BAML: AI work flows without Hallucinations
Vaibhav Gupta BAML: AI work flows without Hallucinations
john409870
 
Big Data Analytics Quick Research Guide by Arthur Morgan
Big Data Analytics Quick Research Guide by Arthur MorganBig Data Analytics Quick Research Guide by Arthur Morgan
Big Data Analytics Quick Research Guide by Arthur Morgan
Arthur Morgan
 
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptxIncreasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Increasing Retail Store Efficiency How can Planograms Save Time and Money.pptx
Anoop Ashok
 
Technology Trends in 2025: AI and Big Data Analytics
Technology Trends in 2025: AI and Big Data AnalyticsTechnology Trends in 2025: AI and Big Data Analytics
Technology Trends in 2025: AI and Big Data Analytics
InData Labs
 
ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes Partner Innovation Updates for May 2025ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes Partner Innovation Updates for May 2025
ThousandEyes
 
Unlocking the Power of IVR: A Comprehensive Guide
Unlocking the Power of IVR: A Comprehensive GuideUnlocking the Power of IVR: A Comprehensive Guide
Unlocking the Power of IVR: A Comprehensive Guide
vikasascentbpo
 
MINDCTI revenue release Quarter 1 2025 PR
MINDCTI revenue release Quarter 1 2025 PRMINDCTI revenue release Quarter 1 2025 PR
MINDCTI revenue release Quarter 1 2025 PR
MIND CTI
 
Quantum Computing Quick Research Guide by Arthur Morgan
Quantum Computing Quick Research Guide by Arthur MorganQuantum Computing Quick Research Guide by Arthur Morgan
Quantum Computing Quick Research Guide by Arthur Morgan
Arthur Morgan
 
Build Your Own Copilot & Agents For Devs
Build Your Own Copilot & Agents For DevsBuild Your Own Copilot & Agents For Devs
Build Your Own Copilot & Agents For Devs
Brian McKeiver
 
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
#StandardsGoals for 2025: Standards & certification roundup - Tech Forum 2025
BookNet Canada
 

Deep Convolutional GANs - meaning of latent space

  • 1. Deep Convolutional GANs ISL Lab Seminar Hansol Kang : Meaning of Latent Space
  • 3. Review of GAN 2018-10-05 3 • Adversarial nets 1) Global Optimality of datag pp  2) Convergence of Algorithm D GVs x )(xpdata “Generative Adversarial Networks” Goal Method
  • 4. D DCGAN 2018-10-05 4 • Introduction * Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015). * “I have the strongest MLP army.” “I have too.” G
  • 6. DCGAN 2018-10-05 6 • Introduction * D G “What are they doing?” “We have a better CNN than MLP” D “I have the strongest MLP army.” “I have too.” G Vanilla GAN DCGAN * Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015).
  • 7. DCGAN 2018-10-05 7 • Contributions Generating Natural Image Deep Convolutional GANs Image Classification using D Filter Visualization Vector arithmetic properties
  • 8. Z“I’m very Important” Who am I? Black box Real D A B C DCGAN 2018-10-05 8 • Contributions Generating Natural Image Deep Convolutional GANs Image Classification using D Filter Visualization Vector arithmetic properties
  • 9. DCGAN 2018-10-05 9 • Approach and Model Architecture Replace any pooling layers with strided convolutions (discriminator) and fractional- strided convolutions (generator). Use batchnorm in both the generator and the discriminator. Remove fully connected hidden layers for deeper architectures. Use ReLU activation in generator for all layers except for the output, which uses Tanh. Use LeakyReLU activation in the discriminator for all layers.
  • 10. DCGAN 2018-10-05 10 • Approach and Model Architecture Strided Convolution Fractional Convolution(Transposed Convolution)
  • 11. DCGAN 2018-10-05 11 • Approach and Model Architecture Batch Normalization Except for these layers. Output layer of Generator Input layer of Discriminator
  • 12. DCGAN 2018-10-05 12 • Approach and Model Architecture No fully connected layer Classical CNN GAP(Global Average Pooling) http://nmhkahn.github.io/Casestudy-CNN
  • 13. DCGAN 2018-10-05 13 • Approach and Model Architecture No fully connected layer https://raw.githubusercontent.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN/master/pytorch_DCGAN.png
  • 14. DCGAN 2018-10-05 14 • Approach and Model Architecture ReLU, Tanh, LeakyReLU http://gmelli.org/RKB/Rectified_Linear_Unit_(ReLU)_Activation_Function Generator : ReLU, Tanh Discriminator : LeakyReLu , Sigmoid
  • 15. DCGAN 2018-10-05 15 • Details of Adversarial Training • Mini-batch stochastic gradient descent(SGD); mini-batch size of 128 • All weights initialized from a zero-centered Normal distribution with standard deviation 0.02 • Leaky slope 0.2 • Adam optimizer; lr =0.0002, beta1 = 0.9, beta2 = 0.5
  • 16. DCGAN 2018-10-05 16 • Details of Adversarial Training LSUN dataset 1 epoch
  • 17. DCGAN 2018-10-05 17 • Details of Adversarial Training LSUN dataset 5 epochs
  • 18. DCGAN 2018-10-05 18 • Empirical Validation of DCGANs Capabilities • CIFAR-10 • Classification • Domain robustness
  • 19. DCGAN 2018-10-05 19 • Empirical Validation of DCGANs Capabilities SVHN(Street View House Numbers) dataset
  • 20. DCGAN 2018-10-05 20 • Investigating and Visualizing The Internals of The Networks Walking in the latent space
  • 21. DCGAN 2018-10-05 21 • Investigating and Visualizing The Internals of The Networks(cont.) Visualizing the discriminator features
  • 22. DCGAN 2018-10-05 22 • Investigating and Visualizing The Internals of The Networks(cont.) Forgetting to draw certain objects in charge of windows in charge of beds in charge of lamps in charge of doors … Latent code Filters(Conv) Generation 1 0 0 Noise(z)
  • 23. DCGAN 2018-10-05 23 • Investigating and Visualizing The Internals of The Networks(cont.) Forgetting to draw certain objects
  • 24. DCGAN 2018-10-05 24 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
  • 25. DCGAN 2018-10-05 25 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
  • 26. DCGAN 2018-10-05 26 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
  • 27. DCGAN 2018-10-05 27 • Investigating and Visualizing The Internals of The Networks(cont.) Vector arithmetic on face samples
  • 30. Experiment • Results#1 CelebA 2018-10-05 30 Ground Truth Vanilla GAN : DCGAN : Epoch 1 Epoch 5 Epoch 100 Epoch 1 Epoch 5 Epoch 30 Still have this sample Results are cherry picked
  • 31. Experiment • Results#2 LSUN) 2018-10-05 31 Ground Truth Vanilla GAN : DCGAN : Epoch 1 Epoch 5 Epoch 12 Epoch 1 Epoch 2 Epoch 5 Results are cherry picked
  • 32. Experiment • Results#3 Korean Idol – Transfer trial 2018-10-05 32 • I used weights and biases generated by celebA learning. • I wanted the effect of transfer learning but failed. Maybe these factors (Asian, cropping image) Ground Truth Epoch 1 Epoch 2 Epoch 3 Epoch 4 Epoch 5 Epoch 6
  • 33. Experiment • Results#4 Korean Idol 2018-10-05 33 Ground Truth Epoch 1 Epoch 5 Epoch 30 Epoch 50 Epoch 100 Epoch 150 • 10000 images Insufficient data set
  • 34. Summary 2018-10-05 34 • Stable set of architectures for training generative adversarial networks • Good representations of images for supervised learning and generative modeling • Sometimes collapse a subset of filters to a single oscillating mode • Latent code has a special meaning, not a simple noise component. [Instability of GAN]
  • 35. Future work 2018-10-05 35 Paper Review Vanilla GAN DCGAN InfoGAN Unrolled GAN Wasserstein GAN LS GAN BEGAN Pix2Pix Cycle GAN Proposed Model SpyGAN Tips Document Programming Mathematical Study Information theory (working title)
  • 36. &
  • 37. Appendix • Issues at the VAE Seminar (18.07.23) 2018-10-05 37  Issue#1 Performance of VAE and GAN  Issue#2 Log likelihood  Issue#3 Dimension of latent code  Issue#4 Why manifold? Durk Kingma 1. Adam: A Method for Stochastic Optimization 2. Auto-Encoding Variational Bayes Machine Learning researcher at OpenAI - Mathematically very difficult papers Intuitive explanation https://www.youtube.com/watch?v=o_peo6U7IRM 오토인코더의 모든 것 : I refer to this video
  • 38. Appendix • Issue #1 Performance of VAE and GAN 2018-10-05 38 “Compared to GAN, VAE is relatively blurred and I do not know why.” “Cost function”     )(||)|()(|log),,( )|( zpxzqKLzgxExL xzq    ),,(min xL VAE ),(maxmin DGV DG       zGDExDEDGV zdata pzpx  1log)(log),( ~~ GAN Intuition Reconstruction Error Regularization ≈ D Loss ≈ G Loss
  • 39. Appendix • Issue #1 Performance of VAE and GAN 2018-10-05 39 VAE Loss= Recon. Error + Regularization GAN Loss= G_Loss + D_Loss E D Recon. Error D Real Fake 1. Optimize 2. Image Quality 3. Generalization VAE vs. GAN
  • 40. Appendix • Issue #2 Log likelihood 2018-10-05 40 Question about log likelihood “Summation and monotonically increasing” MLE(Maximum Likelihood Estimation) : Unknown parameter estimation from observation )|(maxargˆ   yp eg. Gaussian Distribution Mean and Std  i iyp )|(maxarg          i i i i ypyp )|(logmaxarg)|(logmaxarg   : monotonically increasing function1 Log(x) cf.  i ixp )(logmaxarg   Generation model
  • 41. Appendix • Issue #3 Dimension of latent code 2018-10-05 41 “Is the latent code dimension always small?” “Yes” AE, What’s this? Dimension reduction E D High Low Interested Sparse AE FAILED
  • 42. Appendix • Issue #4 Why manifold? 2018-10-05 42 What’s the manifold and Why explain the manifold? “Concept of manifold and Difference of between AE and VAE” High Low Subspace =Manifold Concept of manifold D Purpose of AE : Manifold Learning Purpose of AE and VAE Assumption(manifold hypothesis) Uniform sampling E Unsupervised Learning D Purpose of VAE : Generative Model E Unsupervised Learning : Correlation between generation and manifold…
  • 43. Appendix • PyTorch (Variable length inputs) 2018-10-05 43 Shape = {Size} torch.Size([128, 3, 32, 32]) Shape = {Size} torch.Size([128, 64, 16, 16]) Shape = {Size} torch.Size([128, 16384])3x32x32 CIFAR-10 Shape = {Size} torch.Size([128, 64, 109, 89]) Shape = {Size} torch.Size([128, 3, 218, 178]) Shape = {Size} torch.Size([128, 620864]) 3x178x218 CelebA Conv Input Pool FC Conv2d(in_ch, out_ch, k_size, s, p) Reshape(bat_sz,-1) Input size is not fixed.