SlideShare a Scribd company logo
Deep Generative Modelling
Recap
● Part 1: Introduction to Machine Learning (Ivaylo Strandjev)
● Part 2: Deep Learning (Teodor Radenkov)
● Part 3: Playing with Image models (Ivaylo Strandjev)
Playing with image models
● The convolution operator for images
● Deep CNNs like Inception Resnet V2
● Interpretability of Deep Neural Networks
● Adversarial Examples
● Motivation
● Generative and Discriminative Models
● Popular Deep Generative Models for Images
● Conditional Generative Models
Today
Generative Models
Generates new random observable data, models the joint distribution of all
variables.
Given some dataset D generate new samples like D, but not the same.
We need to adjust their hidden parameters
Considered as branch of unsupervised learning, but they can be used for tasks
like classification
“What I cannot create, I do not
understand.”
—Richard Feynman
Motivation
● Tremendous amount of information out there in the world
● Machines are good in solving specific tasks
● Better than humans in Object recognition, Go, Speech recognition
● Cannot build compact representations of the world
How can we overcome this intelligence
gap?
By forcing our models to learn very compact and
disentangled representations.
Disentangled factors
● P( X | Z), where X is an image, Z is a vector that causes (explains) X
● We would like the dimensions of Z to describe
real world factors
● Z which has a separate dimension for lighting, guitar,
bookshelf , rotation will be considered more
disentangled than the raw pixels of X
● P(guitar | Z) can be easily computed with
Disentangled representation.
Applications
Short term applications
● Image translation, denoising, super-resolution
● Domain Adaptation
● Music, Audio and Text Generation
Long term applications
● Understanding of the real world
Discriminative Models
● ImageNet. Here y would be the vector of 1000 labels and x some image from
the dataset.
● They are trying to maximize log P(y | x)
● Predictions obtained by argmax of yi
: P(yi
| x)
● Classification models are discriminative ones.
● During training maximize the probability log P(X)
● Generate new sampled images close to the ImageNet distribution P(X)
● During inference for some image X depending on the model you might be
able to estimate the probability of the image X under the model
Generative Models
p(y, w, β)p(y, β ; w)
Discriminative Generative
y - text categories
w - sequence of words
β - model parameters
Properties and Drawbacks of Discriminative Models
● Good at capturing statistical regularities of the data
● Find features invariant to characteristics you don’t care for the task
○ Object classification: Rotation, Translation, Lighting, Color
○ Segmentation: You care for Rotation, Translation
● Having difficulties to build disentangled representations
● Adversarial examples are good example for that
Generation from Discriminative Model (Example)
Handwriting Model This is regarding my friend, Kate Zack
Gradient ascent on the input image X
Generation from Discriminative Model (Example)
Handwriting Model
P E T K O
X
Maximize
Generative Models
● Gaussian mixture model
● Hidden Markov model
● Naive Bayes
● Latent Dirichlet allocation
● … many others
Deep Generative Models
● Restricted Boltzmann Machines
● Variational Autoencoders
● PixelRNN, PixelCNN
● Generative Adversarial Networks
● Neural Language Models
● WaveNet
Deep Generative Models
● Restricted Boltzmann Machines
● Variational Autoencoders
● PixelRNN, PixelCNN
● Generative Adversarial Networks
● Neural Language Models
● WaveNet
Deep Generative Models
Generator
Latent variables (code)
Autoencoders
Autoencoder
network
Loss = Pixelwise L2 or Softmax.
● Latent variables
● Lower dimensional than the input
Autoencoders
Encoder Decoder
Loss =
Autoencoders
● random latent code won’t get us anywhere
● Pass an image to the encoder to get “valid” code
Encoder Decoder
● Encoder-Decoder architecture
● Forcing the latent code to be Gaussian distributed
● Sample the latent code from the Gaussian and pass it to the decoder network
Variational Autoencoders
Variational Autoencoders
Encoder Decoder
Loss =
Mean
Std
Sampled code
● CIFAR-10
● Blurry images
● Good
approximation
of the likelihood
of the input
data
Variational Autoencoders - Samples
Input Output
Deep Recurrent Attentive Writer
● Generates the image sequentially
● On each step the model decides where to focus and draw
● Uses an attention mechanism to achieve it
○ A topic of different lecture :(
Deep Recurrent Attention Writer (DRAW)
● Google Street View Numbers
● The red rectangle is showing
where the model is attending on
the current step
● Impressive as DRAW is the first
successful model that generates
images sequentially
Deep Recurrent Attentive Writer
VAEs DRAW
Fully Convolutional Model
● Typically using pre-trained classification network as encoder
● Most often VGG-16, because it’s fast and has less parameters
● Using transposed convolution layers as decoder until we reach the desired
shape
● Often the architecture of the encoder is the transposed of the one of the
decoder
Fully Convolutional Decoder
Radford et al 2015
Transposed Convolution (Deconvolution)
● Kernel size = 3
● Stride = 1
Animation source: leonardoaraujosantos.gitbooks.io
Input layer
Output layer
Properties of Transposed Convolution
● During backpropagation a convolutional layer becomes transposed
convolution
● Checkerboard pattern might appear in the generated image (sensitive to
kernel and stride sizes)
Odena, et al., "Deconvolution and Checkerboard
Artifacts", Distill, 2016. http://doi.org/10.23915
● In practice, VAEs latent code dimensions are very interpretable
● To achieve this it collapses some latent dimensions and doesn’t use them
● Able to generate samples close to the data distribution
Pros of VAEs
Drawbacks of VAEs
● Pixels in the L2 loss function are independent, which leads to blurry images
● The exact probability of a generated image under the model is intractable to
compute
X - input image, Z latent code
Generative Adversarial Networks (GAN)
● A generative model invented by Ian Goodfellow in 2014
● Already widely adopted and an area of massive research
● New GAN paper is published every week
● Has many awesome applications. We’ll see some of them later on.
● GANs define the generative problem as an adversarial game between two
networks
GANs
Generator
Discriminator
Sample
Real
Images
Sample
Real Fake
Loss
GANs - Discriminator Training
Generator
Discriminator
Sample
Real
Images
Sample
Real Fake
Classification Loss
GANs - Generator Training
Generator
Discriminator
Sample
Real
Images
Sample
Real Fake
Maximize
Problems in GAN Training
● Instable during training
● Mode colapse
● Higher Log-likelihood != better samples
However, GAN training is getting easier. Checkout Wassterstein GANs and
LSGANs .
DCGAN
Radford et al 2015
Generative Adversarial Networks
Generative Adversarial Networks
Conditional Generative Adversarial Networks
Real
Data
● Consists of pairs (X,Y)
● P(Y|X) : generate Y given X
Image-to-Image translation (pix2pix)
pix2pix:Image-to-Image Translation with CANs, Isola et al 2016
Image-to-Image translation (example)
pix2pix:Image-to-Image Translation with CANs, Isola et al 2016
CycleGAN
X Y
● X and Y are unpaired collections of images
● different domains of the same world
● Learn to translate image X into Y
CycleGAN
Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs)
CycleGAN (failure case)
Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs)
CycleGAN
Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs)
● Cycle Consistency Loss
○ || F(G(X)) - X ||
○ || G(F(Y)) - Y ||
CycleGAN
Generating Images from Text
Generating Images from Text
References
● https://blog.openai.com/generative-models
● http://distill.pub/2016/deconv-checkerboard/
● https://github.com/junyanz/CycleGAN
● https://github.com/phillipi/pix2pix
● http://videolectures.net/deeplearning2015_bengio_generative_models/
● https://www.youtube.com/watch?v=P78QYjWh5sM&spfreload=1
● http://image-net.org/explore

More Related Content

Similar to Deep Generative Modelling

Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
Catalina Arango
 
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
 
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 Deep Learning Approaches to Image Processing Applications (1).pptx
GAN Deep Learning Approaches to Image Processing Applications (1).pptxGAN Deep Learning Approaches to Image Processing Applications (1).pptx
GAN Deep Learning Approaches to Image Processing Applications (1).pptx
RMDAcademicCoordinat
 
Volodymyr Lyubinets “Generative models for images”
Volodymyr Lyubinets  “Generative models for images”Volodymyr Lyubinets  “Generative models for images”
Volodymyr Lyubinets “Generative models for images”
Lviv Startup Club
 
brief Introduction to Different Kinds of GANs
brief Introduction to Different Kinds of GANsbrief Introduction to Different Kinds of GANs
brief Introduction to Different Kinds of GANs
Parham Zilouchian
 
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
Codemotion
 
Generative modeling with Convolutional Neural Networks
Generative modeling with Convolutional Neural NetworksGenerative modeling with Convolutional Neural Networks
Generative modeling with Convolutional Neural Networks
Denis Dus
 
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
Universitat Politècnica de Catalunya
 
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
Codiax
 
Learning visual representation without human label
Learning visual representation without human labelLearning visual representation without human label
Learning visual representation without human label
Kai-Wen Zhao
 
Generative adversarial network_Ayadi_Alaeddine
Generative adversarial network_Ayadi_AlaeddineGenerative adversarial network_Ayadi_Alaeddine
Generative adversarial network_Ayadi_Alaeddine
Deep Learning Italia
 
物件偵測與辨識技術
物件偵測與辨識技術物件偵測與辨識技術
物件偵測與辨識技術
CHENHuiMei
 
Generational Adversarial Neural Networks - Essential Reference
Generational Adversarial Neural Networks - Essential ReferenceGenerational Adversarial Neural Networks - Essential Reference
Generational Adversarial Neural Networks - Essential Reference
Gokul Alex
 
Reading group gan - 20170417
Reading group   gan - 20170417Reading group   gan - 20170417
Reading group gan - 20170417
Shuai Zhang
 
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
Universitat Politècnica de Catalunya
 
Deep Learning Tutorial
Deep Learning Tutorial Deep Learning Tutorial
Deep Learning Tutorial
Ligeng Zhu
 
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
Universitat Politècnica de Catalunya
 

Similar to Deep Generative Modelling (20)

Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
Let's paint a Picasso - A Look at Generative Adversarial Networks (GAN) and i...
 
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...
 
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 Deep Learning Approaches to Image Processing Applications (1).pptx
GAN Deep Learning Approaches to Image Processing Applications (1).pptxGAN Deep Learning Approaches to Image Processing Applications (1).pptx
GAN Deep Learning Approaches to Image Processing Applications (1).pptx
 
Volodymyr Lyubinets “Generative models for images”
Volodymyr Lyubinets  “Generative models for images”Volodymyr Lyubinets  “Generative models for images”
Volodymyr Lyubinets “Generative models for images”
 
brief Introduction to Different Kinds of GANs
brief Introduction to Different Kinds of GANsbrief Introduction to Different Kinds of GANs
brief Introduction to Different Kinds of GANs
 
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
Alberto Massidda - Scenes from a memory - Codemotion Rome 2019
 
Generative modeling with Convolutional Neural Networks
Generative modeling with Convolutional Neural NetworksGenerative modeling with Convolutional Neural Networks
Generative modeling with Convolutional Neural Networks
 
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
Deep Generative Models - Kevin McGuinness - UPC Barcelona 2018
 
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
Generative Models and Adversarial Training (D2L3 Insight@DCU Machine Learning...
 
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
Jakub Langr (University of Oxford) - Overview of Generative Adversarial Netwo...
 
Learning visual representation without human label
Learning visual representation without human labelLearning visual representation without human label
Learning visual representation without human label
 
Generative adversarial network_Ayadi_Alaeddine
Generative adversarial network_Ayadi_AlaeddineGenerative adversarial network_Ayadi_Alaeddine
Generative adversarial network_Ayadi_Alaeddine
 
物件偵測與辨識技術
物件偵測與辨識技術物件偵測與辨識技術
物件偵測與辨識技術
 
Generational Adversarial Neural Networks - Essential Reference
Generational Adversarial Neural Networks - Essential ReferenceGenerational Adversarial Neural Networks - Essential Reference
Generational Adversarial Neural Networks - Essential Reference
 
Reading group gan - 20170417
Reading group   gan - 20170417Reading group   gan - 20170417
Reading group gan - 20170417
 
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
Generative Adversarial Networks GAN - Santiago Pascual - UPC Barcelona 2018
 
Deep Learning Tutorial
Deep Learning Tutorial Deep Learning Tutorial
Deep Learning Tutorial
 
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
Deep Learning for Computer Vision: Unsupervised Learning (UPC 2016)
 
gan.pdf
gan.pdfgan.pdf
gan.pdf
 

Recently uploaded

在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
obonagu
 
Final project report on grocery store management system..pdf
Final project report on grocery store management system..pdfFinal project report on grocery store management system..pdf
Final project report on grocery store management system..pdf
Kamal Acharya
 
Water Industry Process Automation and Control Monthly - May 2024.pdf
Water Industry Process Automation and Control Monthly - May 2024.pdfWater Industry Process Automation and Control Monthly - May 2024.pdf
Water Industry Process Automation and Control Monthly - May 2024.pdf
Water Industry Process Automation & Control
 
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
bakpo1
 
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERSCW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
veerababupersonal22
 
6th International Conference on Machine Learning & Applications (CMLA 2024)
6th International Conference on Machine Learning & Applications (CMLA 2024)6th International Conference on Machine Learning & Applications (CMLA 2024)
6th International Conference on Machine Learning & Applications (CMLA 2024)
ClaraZara1
 
Understanding Inductive Bias in Machine Learning
Understanding Inductive Bias in Machine LearningUnderstanding Inductive Bias in Machine Learning
Understanding Inductive Bias in Machine Learning
SUTEJAS
 
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&BDesign and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
Sreedhar Chowdam
 
Student information management system project report ii.pdf
Student information management system project report ii.pdfStudent information management system project report ii.pdf
Student information management system project report ii.pdf
Kamal Acharya
 
Steel & Timber Design according to British Standard
Steel & Timber Design according to British StandardSteel & Timber Design according to British Standard
Steel & Timber Design according to British Standard
AkolbilaEmmanuel1
 
The Role of Electrical and Electronics Engineers in IOT Technology.pdf
The Role of Electrical and Electronics Engineers in IOT Technology.pdfThe Role of Electrical and Electronics Engineers in IOT Technology.pdf
The Role of Electrical and Electronics Engineers in IOT Technology.pdf
Nettur Technical Training Foundation
 
Investor-Presentation-Q1FY2024 investor presentation document.pptx
Investor-Presentation-Q1FY2024 investor presentation document.pptxInvestor-Presentation-Q1FY2024 investor presentation document.pptx
Investor-Presentation-Q1FY2024 investor presentation document.pptx
AmarGB2
 
Recycled Concrete Aggregate in Construction Part III
Recycled Concrete Aggregate in Construction Part IIIRecycled Concrete Aggregate in Construction Part III
Recycled Concrete Aggregate in Construction Part III
Aditya Rajan Patra
 
Hierarchical Digital Twin of a Naval Power System
Hierarchical Digital Twin of a Naval Power SystemHierarchical Digital Twin of a Naval Power System
Hierarchical Digital Twin of a Naval Power System
Kerry Sado
 
road safety engineering r s e unit 3.pdf
road safety engineering  r s e unit 3.pdfroad safety engineering  r s e unit 3.pdf
road safety engineering r s e unit 3.pdf
VENKATESHvenky89705
 
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
ydteq
 
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdfHybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
fxintegritypublishin
 
Technical Drawings introduction to drawing of prisms
Technical Drawings introduction to drawing of prismsTechnical Drawings introduction to drawing of prisms
Technical Drawings introduction to drawing of prisms
heavyhaig
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
zwunae
 
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdfAKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
SamSarthak3
 

Recently uploaded (20)

在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
在线办理(ANU毕业证书)澳洲国立大学毕业证录取通知书一模一样
 
Final project report on grocery store management system..pdf
Final project report on grocery store management system..pdfFinal project report on grocery store management system..pdf
Final project report on grocery store management system..pdf
 
Water Industry Process Automation and Control Monthly - May 2024.pdf
Water Industry Process Automation and Control Monthly - May 2024.pdfWater Industry Process Automation and Control Monthly - May 2024.pdf
Water Industry Process Automation and Control Monthly - May 2024.pdf
 
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
一比一原版(SFU毕业证)西蒙菲莎大学毕业证成绩单如何办理
 
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERSCW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
CW RADAR, FMCW RADAR, FMCW ALTIMETER, AND THEIR PARAMETERS
 
6th International Conference on Machine Learning & Applications (CMLA 2024)
6th International Conference on Machine Learning & Applications (CMLA 2024)6th International Conference on Machine Learning & Applications (CMLA 2024)
6th International Conference on Machine Learning & Applications (CMLA 2024)
 
Understanding Inductive Bias in Machine Learning
Understanding Inductive Bias in Machine LearningUnderstanding Inductive Bias in Machine Learning
Understanding Inductive Bias in Machine Learning
 
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&BDesign and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
Design and Analysis of Algorithms-DP,Backtracking,Graphs,B&B
 
Student information management system project report ii.pdf
Student information management system project report ii.pdfStudent information management system project report ii.pdf
Student information management system project report ii.pdf
 
Steel & Timber Design according to British Standard
Steel & Timber Design according to British StandardSteel & Timber Design according to British Standard
Steel & Timber Design according to British Standard
 
The Role of Electrical and Electronics Engineers in IOT Technology.pdf
The Role of Electrical and Electronics Engineers in IOT Technology.pdfThe Role of Electrical and Electronics Engineers in IOT Technology.pdf
The Role of Electrical and Electronics Engineers in IOT Technology.pdf
 
Investor-Presentation-Q1FY2024 investor presentation document.pptx
Investor-Presentation-Q1FY2024 investor presentation document.pptxInvestor-Presentation-Q1FY2024 investor presentation document.pptx
Investor-Presentation-Q1FY2024 investor presentation document.pptx
 
Recycled Concrete Aggregate in Construction Part III
Recycled Concrete Aggregate in Construction Part IIIRecycled Concrete Aggregate in Construction Part III
Recycled Concrete Aggregate in Construction Part III
 
Hierarchical Digital Twin of a Naval Power System
Hierarchical Digital Twin of a Naval Power SystemHierarchical Digital Twin of a Naval Power System
Hierarchical Digital Twin of a Naval Power System
 
road safety engineering r s e unit 3.pdf
road safety engineering  r s e unit 3.pdfroad safety engineering  r s e unit 3.pdf
road safety engineering r s e unit 3.pdf
 
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
一比一原版(UofT毕业证)多伦多大学毕业证成绩单如何办理
 
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdfHybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
Hybrid optimization of pumped hydro system and solar- Engr. Abdul-Azeez.pdf
 
Technical Drawings introduction to drawing of prisms
Technical Drawings introduction to drawing of prismsTechnical Drawings introduction to drawing of prisms
Technical Drawings introduction to drawing of prisms
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单专业办理
 
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdfAKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
AKS UNIVERSITY Satna Final Year Project By OM Hardaha.pdf
 

Deep Generative Modelling

  • 2. Recap ● Part 1: Introduction to Machine Learning (Ivaylo Strandjev) ● Part 2: Deep Learning (Teodor Radenkov) ● Part 3: Playing with Image models (Ivaylo Strandjev)
  • 3. Playing with image models ● The convolution operator for images ● Deep CNNs like Inception Resnet V2 ● Interpretability of Deep Neural Networks ● Adversarial Examples
  • 4. ● Motivation ● Generative and Discriminative Models ● Popular Deep Generative Models for Images ● Conditional Generative Models Today
  • 5. Generative Models Generates new random observable data, models the joint distribution of all variables. Given some dataset D generate new samples like D, but not the same. We need to adjust their hidden parameters Considered as branch of unsupervised learning, but they can be used for tasks like classification
  • 6. “What I cannot create, I do not understand.” —Richard Feynman
  • 7. Motivation ● Tremendous amount of information out there in the world ● Machines are good in solving specific tasks ● Better than humans in Object recognition, Go, Speech recognition ● Cannot build compact representations of the world
  • 8. How can we overcome this intelligence gap? By forcing our models to learn very compact and disentangled representations.
  • 9. Disentangled factors ● P( X | Z), where X is an image, Z is a vector that causes (explains) X ● We would like the dimensions of Z to describe real world factors ● Z which has a separate dimension for lighting, guitar, bookshelf , rotation will be considered more disentangled than the raw pixels of X ● P(guitar | Z) can be easily computed with Disentangled representation.
  • 10. Applications Short term applications ● Image translation, denoising, super-resolution ● Domain Adaptation ● Music, Audio and Text Generation Long term applications ● Understanding of the real world
  • 11. Discriminative Models ● ImageNet. Here y would be the vector of 1000 labels and x some image from the dataset. ● They are trying to maximize log P(y | x) ● Predictions obtained by argmax of yi : P(yi | x) ● Classification models are discriminative ones.
  • 12. ● During training maximize the probability log P(X) ● Generate new sampled images close to the ImageNet distribution P(X) ● During inference for some image X depending on the model you might be able to estimate the probability of the image X under the model Generative Models
  • 13. p(y, w, β)p(y, β ; w) Discriminative Generative y - text categories w - sequence of words β - model parameters
  • 14. Properties and Drawbacks of Discriminative Models ● Good at capturing statistical regularities of the data ● Find features invariant to characteristics you don’t care for the task ○ Object classification: Rotation, Translation, Lighting, Color ○ Segmentation: You care for Rotation, Translation ● Having difficulties to build disentangled representations ● Adversarial examples are good example for that
  • 15. Generation from Discriminative Model (Example) Handwriting Model This is regarding my friend, Kate Zack Gradient ascent on the input image X
  • 16. Generation from Discriminative Model (Example) Handwriting Model P E T K O X Maximize
  • 17. Generative Models ● Gaussian mixture model ● Hidden Markov model ● Naive Bayes ● Latent Dirichlet allocation ● … many others
  • 18. Deep Generative Models ● Restricted Boltzmann Machines ● Variational Autoencoders ● PixelRNN, PixelCNN ● Generative Adversarial Networks ● Neural Language Models ● WaveNet
  • 19. Deep Generative Models ● Restricted Boltzmann Machines ● Variational Autoencoders ● PixelRNN, PixelCNN ● Generative Adversarial Networks ● Neural Language Models ● WaveNet
  • 22. ● Latent variables ● Lower dimensional than the input Autoencoders Encoder Decoder Loss =
  • 23. Autoencoders ● random latent code won’t get us anywhere ● Pass an image to the encoder to get “valid” code Encoder Decoder
  • 24. ● Encoder-Decoder architecture ● Forcing the latent code to be Gaussian distributed ● Sample the latent code from the Gaussian and pass it to the decoder network Variational Autoencoders
  • 26. ● CIFAR-10 ● Blurry images ● Good approximation of the likelihood of the input data Variational Autoencoders - Samples Input Output
  • 27. Deep Recurrent Attentive Writer ● Generates the image sequentially ● On each step the model decides where to focus and draw ● Uses an attention mechanism to achieve it ○ A topic of different lecture :(
  • 28. Deep Recurrent Attention Writer (DRAW) ● Google Street View Numbers ● The red rectangle is showing where the model is attending on the current step ● Impressive as DRAW is the first successful model that generates images sequentially
  • 29. Deep Recurrent Attentive Writer VAEs DRAW
  • 30. Fully Convolutional Model ● Typically using pre-trained classification network as encoder ● Most often VGG-16, because it’s fast and has less parameters ● Using transposed convolution layers as decoder until we reach the desired shape ● Often the architecture of the encoder is the transposed of the one of the decoder
  • 32. Transposed Convolution (Deconvolution) ● Kernel size = 3 ● Stride = 1 Animation source: leonardoaraujosantos.gitbooks.io Input layer Output layer
  • 33. Properties of Transposed Convolution ● During backpropagation a convolutional layer becomes transposed convolution ● Checkerboard pattern might appear in the generated image (sensitive to kernel and stride sizes) Odena, et al., "Deconvolution and Checkerboard Artifacts", Distill, 2016. http://doi.org/10.23915
  • 34. ● In practice, VAEs latent code dimensions are very interpretable ● To achieve this it collapses some latent dimensions and doesn’t use them ● Able to generate samples close to the data distribution Pros of VAEs
  • 35. Drawbacks of VAEs ● Pixels in the L2 loss function are independent, which leads to blurry images ● The exact probability of a generated image under the model is intractable to compute X - input image, Z latent code
  • 36. Generative Adversarial Networks (GAN) ● A generative model invented by Ian Goodfellow in 2014 ● Already widely adopted and an area of massive research ● New GAN paper is published every week ● Has many awesome applications. We’ll see some of them later on. ● GANs define the generative problem as an adversarial game between two networks
  • 38. GANs - Discriminator Training Generator Discriminator Sample Real Images Sample Real Fake Classification Loss
  • 39. GANs - Generator Training Generator Discriminator Sample Real Images Sample Real Fake Maximize
  • 40. Problems in GAN Training ● Instable during training ● Mode colapse ● Higher Log-likelihood != better samples However, GAN training is getting easier. Checkout Wassterstein GANs and LSGANs .
  • 44. Conditional Generative Adversarial Networks Real Data ● Consists of pairs (X,Y) ● P(Y|X) : generate Y given X
  • 45. Image-to-Image translation (pix2pix) pix2pix:Image-to-Image Translation with CANs, Isola et al 2016
  • 46. Image-to-Image translation (example) pix2pix:Image-to-Image Translation with CANs, Isola et al 2016
  • 47. CycleGAN X Y ● X and Y are unpaired collections of images ● different domains of the same world ● Learn to translate image X into Y
  • 48. CycleGAN Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs)
  • 49. CycleGAN (failure case) Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs)
  • 50. CycleGAN Zhu et al 2017 (Unpaired Image-to-Image Translation using Cycle-Consisten GANs) ● Cycle Consistency Loss ○ || F(G(X)) - X || ○ || G(F(Y)) - Y ||
  • 54. References ● https://blog.openai.com/generative-models ● http://distill.pub/2016/deconv-checkerboard/ ● https://github.com/junyanz/CycleGAN ● https://github.com/phillipi/pix2pix ● http://videolectures.net/deeplearning2015_bengio_generative_models/ ● https://www.youtube.com/watch?v=P78QYjWh5sM&spfreload=1 ● http://image-net.org/explore