SlideShare une entreprise Scribd logo
1  sur  30
Télécharger pour lire hors ligne
Diffusion models for inverse problems
ICLR 2023 (spotlight)
Background: Diffusion models and Stein score
http://yang-song.github.io/blog/2021/score/
𝑝𝑑𝑎𝑡𝑎(𝐱)
∇𝐱log 𝑝𝑑𝑎𝑡𝑎(𝐱)
Training data
𝑠𝜃 𝐱 : ℝ𝑑
↦ ℝ𝑑
≃
2 / 63
http://yang-song.github.io/blog/2021/score/
Background: Diffusion models and Stein score
Explicit score matching
Denoising score matching
Equivalent (Vincent et al. 2010)
𝜽∗
= argmin
𝜽
𝔼[‖∇𝐱log 𝑝 𝐱 − 𝐬𝜽(𝐱)‖2
2
]
𝜽∗
= argmin
𝜽
𝔼[‖∇𝐱log 𝑝 ෤
𝐱 | 𝐱 − 𝐬𝜽(෤
𝐱)‖2
2
]
3 / 63
• Once the score model is trained to optimality,
• i.e. 𝑠𝜃 𝐱 ≃ ∇𝐱𝑝(𝐱)
• Use Langevin dynamics to draw samples
𝐱𝑖+1 ← 𝐱𝑖 + 𝜖∇𝐱log 𝑝 𝐱 + 2𝜖𝐳𝑖
𝑖 = 0, 1, … , 𝐾
http://yang-song.github.io/blog/2021/score/
Background: Diffusion models and Stein score 4 / 63
http://yang-song.github.io/blog/2021/score/
Background: Diffusion models and Stein score 5 / 63
Background: Inverse problems
Imaging system
𝒜
𝐱
Ground truth image
𝜼
noise
𝐲
Measurement
• Problem: recover 𝐱 from noisy measurement 𝐲
• Ill-posed: Infinitely many solutions may exist
• We need to know the prior of the data distribution: how should the image look like?
6 / 63
Background: Examples of inverse problems
𝐲
𝐱
• Inpainting
⨀
• Deblurring (Deconvolution)
𝐱 𝐲
∗
• CS-MRI
𝓕
⨀
7 / 63
Posterior sampling for inverse problems
𝐱 ∼ 𝑝(𝐱|𝐲)
𝐲 = 𝒜(𝐱) + 𝜼
8 / 63
Posterior sampling for inverse problems
𝑝 𝐱 𝐲 =
𝑝 𝐲 𝐱 𝑝(𝐱)
𝑝(𝐲)
log 𝑝 𝐱 𝐲 = log 𝑝 𝐲 𝐱 + log 𝑝 𝐱 − log 𝑝(𝐲)
∇𝐱log 𝑝 𝐱 𝐲 = ∇𝐱log 𝑝 𝐲 𝐱 + ∇𝐱log 𝑝 𝐱
What we know
Measurement
process
9 / 63
Posterior sampling for inverse problems
Prior sampling
10 / 63
𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2
∇𝐱𝑖
log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈)
Posterior sampling for inverse problems
Prior sampling
𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2
∇𝐱𝑖
log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈)
Posterior sampling
𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2
[∇𝐱𝑖
log 𝑝 𝐱𝑖 + ∇𝐱𝑖
log 𝑝(𝐲|𝐱𝑖)] + 𝑔 𝑖 + 1 𝒛
11 / 63
Posterior sampling for inverse problems
Prior sampling
Posterior sampling
Intractable!
12 / 65
𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2
[∇𝐱𝑖
log 𝑝 𝐱𝑖 + ∇𝐱𝑖
log 𝑝(𝐲|𝐱𝑖)] + 𝑔 𝑖 + 1 𝒛
𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2
∇𝐱𝑖
log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈)
The devil is in the likelihood
∇𝐱𝑡
log 𝑝(𝐲|𝐱𝑡)
𝑝 𝐲 𝐱0 = 𝒩(𝐲|𝒜𝐱𝟎, 𝜎2
𝑰)
∇𝐱0
log 𝑝 𝐲 𝐱0 = − 𝒚 − 𝒜𝐱𝟎 𝟐
𝟐
/𝜎2
Intractable!
13 / 63
Key 1. Factorization of the probabilistic graph
𝑝 𝐲 𝐱𝑡 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
14 / 63
𝑝 𝐲 𝐱t = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
= න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0
Key 1. Factorization of the probabilistic graph 15 / 63
𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
= න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0
known partially
known
Key 1. Factorization of the probabilistic graph 16 / 63
𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
= න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0
Key 2. Tweedie’s formula 17 / 63
= 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)]
Key 1. Tweedie denoising
𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
= න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0
Key 2. Tweedie’s formula 18 / 63
= 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)]
≃ 𝑝(𝐲|𝔼[𝐱0|𝐱𝑡])]
Push
expectation
inside
𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0
= න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0
Key 2. Tweedie’s formula 19 / 63
= 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)]
≃ 𝑝(𝐲|𝔼[𝐱0|𝐱𝑡])]
= 𝑝(𝐲|ො
𝐱0)
Key 3. Jensen Bound 20 / 63
∇𝐱𝑡
log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡
log 𝑝(𝐱𝑡) + ∇𝐱𝑡
𝑝 𝐲 𝐱𝑡
Diffusion Posterior Sampling (DPS) 21 / 63
∇𝐱𝑡
log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡
log 𝑝(𝐱𝑡) + ∇𝐱𝑡
𝑝 𝐲 𝐱𝑡
Diffusion Posterior Sampling (DPS)
≃ ∇𝐱𝑡
log 𝑝(𝐱𝑡) + ∇𝐱𝑡
𝑝 𝐲 ො
𝐱0
Theorem 1.
22 / 63
∇𝐱𝑡
log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡
log 𝑝(𝐱𝑡) + ∇𝐱𝑡
log 𝑝 𝐲 𝐱𝑡
Diffusion Posterior Sampling (DPS)
≃ ∇𝐱𝑡
log 𝑝 𝐱𝑡 + ∇𝐱𝑡
log 𝑝 𝐲 ො
𝐱0
≃ 𝑠𝜃∗ 𝐱𝑡, 𝑡 + ∇𝐱𝑡
log 𝑝 𝐲 ො
𝐱0
∇𝐱t
log 𝑝 𝐲 ො
𝐱0 = −𝜌∇𝐱t
‖𝐲 − 𝓐(ො
𝐱0)‖2
2
1. Gaussian
2. Poisson ∇𝐱t
log 𝑝 𝐲 ො
𝐱0 ≃ −𝜌∇𝐱t
‖𝐲 − 𝓐(ො
𝐱0)‖Λ
2
23 / 63
Diffusion Posterior Sampling (DPS) - Algorithm 24 / 63
Results: Super-resolution x16
Measurement (Gauss, 𝜎 = 0.05) DPS
25 / 63
Results: Inpainting (92%)
Measurement (Gauss, 𝜎 = 0.05) DPS
26 / 63
Results: Gaussian deblurring
Measurement (Gauss, 𝜎 = 0.05) DPS
27 / 63
Results: Motion deblurring
Measurement (Gauss, 𝜎 = 0.05) DPS
28 / 63
Results: Phase retrieval
Measurement (Gauss, 𝜎 = 0.05) DPS
29 / 63
Results: Non-uniform deblurring
Measurement (Gauss, 𝜎 = 0.05) DPS
30 / 63

Contenu connexe

Similaire à diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf

Sampling method : MCMC
Sampling method : MCMCSampling method : MCMC
Sampling method : MCMCSEMINARGROOT
 
BlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxBlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxChung Hyung Jin
 
[GAN by Hung-yi Lee]Part 1: General introduction of GAN
[GAN by Hung-yi Lee]Part 1: General introduction of GAN[GAN by Hung-yi Lee]Part 1: General introduction of GAN
[GAN by Hung-yi Lee]Part 1: General introduction of GANNAVER Engineering
 
Semana 11 numeros complejos ii álgebra-uni ccesa007
Semana 11   numeros complejos ii   álgebra-uni ccesa007Semana 11   numeros complejos ii   álgebra-uni ccesa007
Semana 11 numeros complejos ii álgebra-uni ccesa007Demetrio Ccesa Rayme
 
research on journaling
research on journalingresearch on journaling
research on journalingrikaseorika
 
ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022
 ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022 ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022
ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022anasKhalaf4
 
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 32018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3Yosia Adi Setiawan
 
B.tech ii unit-5 material vector integration
B.tech ii unit-5 material vector integrationB.tech ii unit-5 material vector integration
B.tech ii unit-5 material vector integrationRai University
 
Review of generative adversarial nets
Review of generative adversarial netsReview of generative adversarial nets
Review of generative adversarial netsSungminYou
 
function power point presentation for class 11 and 12 for jee
function power point presentation for class 11 and 12 for jeefunction power point presentation for class 11 and 12 for jee
function power point presentation for class 11 and 12 for jeeMohanSonawane
 
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاول
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاولملزمة الرياضيات للصف السادس الاحيائي الفصل الاول
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاولanasKhalaf4
 
Functions ppt Dr Frost Maths Mixed questions
Functions ppt Dr Frost Maths Mixed questionsFunctions ppt Dr Frost Maths Mixed questions
Functions ppt Dr Frost Maths Mixed questionsgcutbill
 
Detailed Description on Cross Entropy Loss Function
Detailed Description on Cross Entropy Loss FunctionDetailed Description on Cross Entropy Loss Function
Detailed Description on Cross Entropy Loss Function범준 김
 
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 32018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3Yosia Adi Setiawan
 
Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Dimas Ruliandi
 
Penjelasan Integral Lipat dua dan Penerapan pada momen inersia
Penjelasan Integral Lipat dua dan Penerapan pada momen inersiaPenjelasan Integral Lipat dua dan Penerapan pada momen inersia
Penjelasan Integral Lipat dua dan Penerapan pada momen inersiabisma samudra
 
GAN in_kakao
GAN in_kakaoGAN in_kakao
GAN in_kakaoJunho Kim
 
서포트 벡터 머신(Support Vector Machine, SVM)
서포트 벡터 머신(Support Vector Machine, SVM)서포트 벡터 머신(Support Vector Machine, SVM)
서포트 벡터 머신(Support Vector Machine, SVM)Hansol Kang
 

Similaire à diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf (20)

Sampling method : MCMC
Sampling method : MCMCSampling method : MCMC
Sampling method : MCMC
 
BlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptxBlindDPS_presentation_silde.pptx
BlindDPS_presentation_silde.pptx
 
[GAN by Hung-yi Lee]Part 1: General introduction of GAN
[GAN by Hung-yi Lee]Part 1: General introduction of GAN[GAN by Hung-yi Lee]Part 1: General introduction of GAN
[GAN by Hung-yi Lee]Part 1: General introduction of GAN
 
Semana 11 numeros complejos ii álgebra-uni ccesa007
Semana 11   numeros complejos ii   álgebra-uni ccesa007Semana 11   numeros complejos ii   álgebra-uni ccesa007
Semana 11 numeros complejos ii álgebra-uni ccesa007
 
research on journaling
research on journalingresearch on journaling
research on journaling
 
ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022
 ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022 ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022
ملزمة الرياضيات للصف السادس التطبيقي الفصل الاول الاعداد المركبة 2022
 
Z transforms
Z transformsZ transforms
Z transforms
 
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 32018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 7 Rombel 3
 
B.tech ii unit-5 material vector integration
B.tech ii unit-5 material vector integrationB.tech ii unit-5 material vector integration
B.tech ii unit-5 material vector integration
 
Review of generative adversarial nets
Review of generative adversarial netsReview of generative adversarial nets
Review of generative adversarial nets
 
function power point presentation for class 11 and 12 for jee
function power point presentation for class 11 and 12 for jeefunction power point presentation for class 11 and 12 for jee
function power point presentation for class 11 and 12 for jee
 
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاول
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاولملزمة الرياضيات للصف السادس الاحيائي الفصل الاول
ملزمة الرياضيات للصف السادس الاحيائي الفصل الاول
 
Functions ppt Dr Frost Maths Mixed questions
Functions ppt Dr Frost Maths Mixed questionsFunctions ppt Dr Frost Maths Mixed questions
Functions ppt Dr Frost Maths Mixed questions
 
Detailed Description on Cross Entropy Loss Function
Detailed Description on Cross Entropy Loss FunctionDetailed Description on Cross Entropy Loss Function
Detailed Description on Cross Entropy Loss Function
 
On Cubic Graceful Labeling
On Cubic Graceful LabelingOn Cubic Graceful Labeling
On Cubic Graceful Labeling
 
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 32018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3
2018 Geometri Transformasi Perkalian 5 Isometri Kelompok 8 Rombel 3
 
Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...Direct solution of sparse network equations by optimally ordered triangular f...
Direct solution of sparse network equations by optimally ordered triangular f...
 
Penjelasan Integral Lipat dua dan Penerapan pada momen inersia
Penjelasan Integral Lipat dua dan Penerapan pada momen inersiaPenjelasan Integral Lipat dua dan Penerapan pada momen inersia
Penjelasan Integral Lipat dua dan Penerapan pada momen inersia
 
GAN in_kakao
GAN in_kakaoGAN in_kakao
GAN in_kakao
 
서포트 벡터 머신(Support Vector Machine, SVM)
서포트 벡터 머신(Support Vector Machine, SVM)서포트 벡터 머신(Support Vector Machine, SVM)
서포트 벡터 머신(Support Vector Machine, SVM)
 

Plus de Chung Hyung Jin

mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxmr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxChung Hyung Jin
 
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Chung Hyung Jin
 
Score-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxScore-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxChung Hyung Jin
 
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Chung Hyung Jin
 
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Chung Hyung Jin
 
Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Chung Hyung Jin
 
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...Chung Hyung Jin
 
Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Chung Hyung Jin
 
Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Chung Hyung Jin
 

Plus de Chung Hyung Jin (9)

mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptxmr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
mr-denoising-and-super-resolution-using-regularized-reverse-diffusion.pptx
 
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
Low-dose sparse-view HAADF-STEM-EDX tomography of nanocrystals using unsuperv...
 
Score-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptxScore-based diffusion models for accelerated MRI.pptx
Score-based diffusion models for accelerated MRI.pptx
 
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
Come-Closer-Diffuse-Faster Accelerating Conditional Diffusion Models for Inve...
 
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
Two stage deep learning for accelerated 3 d time-of-flight mra without matche...
 
Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...Missing cone artifact removal in odt using unsupervised deep learning in the ...
Missing cone artifact removal in odt using unsupervised deep learning in the ...
 
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
A deep learning model for diagnosing gastric mucosal lesions using endoscopic...
 
Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...Feature disentanglement in generating a three dimensional structure from a tw...
Feature disentanglement in generating a three dimensional structure from a tw...
 
Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...Unsupervised deep learning methods for biological image reconstruction and en...
Unsupervised deep learning methods for biological image reconstruction and en...
 

Dernier

Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...Christo Ananth
 
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Call Girls in Nagpur High Profile
 
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Standamitlee9823
 
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...Christo Ananth
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfJiananWang21
 
Thermal Engineering Unit - I & II . ppt
Thermal Engineering  Unit - I & II . pptThermal Engineering  Unit - I & II . ppt
Thermal Engineering Unit - I & II . pptDineshKumar4165
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Bookingdharasingh5698
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Christo Ananth
 
Bhosari ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For ...
Bhosari ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For ...Bhosari ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For ...
Bhosari ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For ...tanu pandey
 
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Wakad Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance Bookingroncy bisnoi
 
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Bookingdharasingh5698
 
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptx
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptxBSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptx
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptxfenichawla
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdfankushspencer015
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756dollysharma2066
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performancesivaprakash250
 
Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01KreezheaRecto
 

Dernier (20)

(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
(INDIRA) Call Girl Aurangabad Call Now 8617697112 Aurangabad Escorts 24x7
 
Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...Call for Papers - International Journal of Intelligent Systems and Applicatio...
Call for Papers - International Journal of Intelligent Systems and Applicatio...
 
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...Top Rated  Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
Top Rated Pune Call Girls Budhwar Peth ⟟ 6297143586 ⟟ Call Me For Genuine Se...
 
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bangalore ☎ 7737669865 🥵 Book Your One night Stand
 
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
Call for Papers - Educational Administration: Theory and Practice, E-ISSN: 21...
 
data_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdfdata_management_and _data_science_cheat_sheet.pdf
data_management_and _data_science_cheat_sheet.pdf
 
Thermal Engineering Unit - I & II . ppt
Thermal Engineering  Unit - I & II . pptThermal Engineering  Unit - I & II . ppt
Thermal Engineering Unit - I & II . ppt
 
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Ankleshwar 7001035870 Whatsapp Number, 24/07 Booking
 
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
Call for Papers - African Journal of Biological Sciences, E-ISSN: 2663-2187, ...
 
Bhosari ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For ...
Bhosari ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For ...Bhosari ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready For ...
Bhosari ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready For ...
 
Call Now ≽ 9953056974 ≼🔝 Call Girls In New Ashok Nagar ≼🔝 Delhi door step de...
Call Now ≽ 9953056974 ≼🔝 Call Girls In New Ashok Nagar  ≼🔝 Delhi door step de...Call Now ≽ 9953056974 ≼🔝 Call Girls In New Ashok Nagar  ≼🔝 Delhi door step de...
Call Now ≽ 9953056974 ≼🔝 Call Girls In New Ashok Nagar ≼🔝 Delhi door step de...
 
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Wakad Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Wakad Call Me 7737669865 Budget Friendly No Advance Booking
 
NFPA 5000 2024 standard .
NFPA 5000 2024 standard                                  .NFPA 5000 2024 standard                                  .
NFPA 5000 2024 standard .
 
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Palanpur 7001035870 Whatsapp Number, 24/07 Booking
 
Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024Water Industry Process Automation & Control Monthly - April 2024
Water Industry Process Automation & Control Monthly - April 2024
 
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptx
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptxBSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptx
BSides Seattle 2024 - Stopping Ethan Hunt From Taking Your Data.pptx
 
AKTU Computer Networks notes --- Unit 3.pdf
AKTU Computer Networks notes ---  Unit 3.pdfAKTU Computer Networks notes ---  Unit 3.pdf
AKTU Computer Networks notes --- Unit 3.pdf
 
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
FULL ENJOY Call Girls In Mahipalpur Delhi Contact Us 8377877756
 
UNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its PerformanceUNIT - IV - Air Compressors and its Performance
UNIT - IV - Air Compressors and its Performance
 
Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01Double rodded leveling 1 pdf activity 01
Double rodded leveling 1 pdf activity 01
 

diffusion_posterior_sampling_for_general_noisy_inverse_problems_slideshare.pdf

  • 1. Diffusion models for inverse problems ICLR 2023 (spotlight)
  • 2. Background: Diffusion models and Stein score http://yang-song.github.io/blog/2021/score/ 𝑝𝑑𝑎𝑡𝑎(𝐱) ∇𝐱log 𝑝𝑑𝑎𝑡𝑎(𝐱) Training data 𝑠𝜃 𝐱 : ℝ𝑑 ↦ ℝ𝑑 ≃ 2 / 63
  • 3. http://yang-song.github.io/blog/2021/score/ Background: Diffusion models and Stein score Explicit score matching Denoising score matching Equivalent (Vincent et al. 2010) 𝜽∗ = argmin 𝜽 𝔼[‖∇𝐱log 𝑝 𝐱 − 𝐬𝜽(𝐱)‖2 2 ] 𝜽∗ = argmin 𝜽 𝔼[‖∇𝐱log 𝑝 ෤ 𝐱 | 𝐱 − 𝐬𝜽(෤ 𝐱)‖2 2 ] 3 / 63
  • 4. • Once the score model is trained to optimality, • i.e. 𝑠𝜃 𝐱 ≃ ∇𝐱𝑝(𝐱) • Use Langevin dynamics to draw samples 𝐱𝑖+1 ← 𝐱𝑖 + 𝜖∇𝐱log 𝑝 𝐱 + 2𝜖𝐳𝑖 𝑖 = 0, 1, … , 𝐾 http://yang-song.github.io/blog/2021/score/ Background: Diffusion models and Stein score 4 / 63
  • 6. Background: Inverse problems Imaging system 𝒜 𝐱 Ground truth image 𝜼 noise 𝐲 Measurement • Problem: recover 𝐱 from noisy measurement 𝐲 • Ill-posed: Infinitely many solutions may exist • We need to know the prior of the data distribution: how should the image look like? 6 / 63
  • 7. Background: Examples of inverse problems 𝐲 𝐱 • Inpainting ⨀ • Deblurring (Deconvolution) 𝐱 𝐲 ∗ • CS-MRI 𝓕 ⨀ 7 / 63
  • 8. Posterior sampling for inverse problems 𝐱 ∼ 𝑝(𝐱|𝐲) 𝐲 = 𝒜(𝐱) + 𝜼 8 / 63
  • 9. Posterior sampling for inverse problems 𝑝 𝐱 𝐲 = 𝑝 𝐲 𝐱 𝑝(𝐱) 𝑝(𝐲) log 𝑝 𝐱 𝐲 = log 𝑝 𝐲 𝐱 + log 𝑝 𝐱 − log 𝑝(𝐲) ∇𝐱log 𝑝 𝐱 𝐲 = ∇𝐱log 𝑝 𝐲 𝐱 + ∇𝐱log 𝑝 𝐱 What we know Measurement process 9 / 63
  • 10. Posterior sampling for inverse problems Prior sampling 10 / 63 𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2 ∇𝐱𝑖 log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈)
  • 11. Posterior sampling for inverse problems Prior sampling 𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2 ∇𝐱𝑖 log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈) Posterior sampling 𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2 [∇𝐱𝑖 log 𝑝 𝐱𝑖 + ∇𝐱𝑖 log 𝑝(𝐲|𝐱𝑖)] + 𝑔 𝑖 + 1 𝒛 11 / 63
  • 12. Posterior sampling for inverse problems Prior sampling Posterior sampling Intractable! 12 / 65 𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2 [∇𝐱𝑖 log 𝑝 𝐱𝑖 + ∇𝐱𝑖 log 𝑝(𝐲|𝐱𝑖)] + 𝑔 𝑖 + 1 𝒛 𝐱𝑖 ← 𝐱𝑖+1 + 𝑓 𝐱𝑖+1, 𝑖 + 1 − 𝑔 𝑖 + 1 2 ∇𝐱𝑖 log 𝑝 𝐱𝑖 + 𝑔 𝑖 + 1 𝒛, 𝒛 ∼ 𝒩(𝟎, 𝐈)
  • 13. The devil is in the likelihood ∇𝐱𝑡 log 𝑝(𝐲|𝐱𝑡) 𝑝 𝐲 𝐱0 = 𝒩(𝐲|𝒜𝐱𝟎, 𝜎2 𝑰) ∇𝐱0 log 𝑝 𝐲 𝐱0 = − 𝒚 − 𝒜𝐱𝟎 𝟐 𝟐 /𝜎2 Intractable! 13 / 63
  • 14. Key 1. Factorization of the probabilistic graph 𝑝 𝐲 𝐱𝑡 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 14 / 63
  • 15. 𝑝 𝐲 𝐱t = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 = න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0 Key 1. Factorization of the probabilistic graph 15 / 63
  • 16. 𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 = න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0 known partially known Key 1. Factorization of the probabilistic graph 16 / 63
  • 17. 𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 = න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0 Key 2. Tweedie’s formula 17 / 63 = 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)] Key 1. Tweedie denoising
  • 18. 𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 = න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0 Key 2. Tweedie’s formula 18 / 63 = 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)] ≃ 𝑝(𝐲|𝔼[𝐱0|𝐱𝑡])] Push expectation inside
  • 19. 𝑝 𝐲 𝐱0 = න 𝑝 𝐲 𝐱0, 𝐱t 𝑝 𝐱0 𝐱t 𝑑𝐱0 = න 𝑝 𝐲 𝐱0 𝑝 𝐱0 𝐱t 𝑑𝐱0 Key 2. Tweedie’s formula 19 / 63 = 𝔼𝑝(𝐱𝟎|𝐱𝒕)[𝑝(𝐲|𝐱𝟎)] ≃ 𝑝(𝐲|𝔼[𝐱0|𝐱𝑡])] = 𝑝(𝐲|ො 𝐱0)
  • 20. Key 3. Jensen Bound 20 / 63
  • 21. ∇𝐱𝑡 log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡 log 𝑝(𝐱𝑡) + ∇𝐱𝑡 𝑝 𝐲 𝐱𝑡 Diffusion Posterior Sampling (DPS) 21 / 63
  • 22. ∇𝐱𝑡 log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡 log 𝑝(𝐱𝑡) + ∇𝐱𝑡 𝑝 𝐲 𝐱𝑡 Diffusion Posterior Sampling (DPS) ≃ ∇𝐱𝑡 log 𝑝(𝐱𝑡) + ∇𝐱𝑡 𝑝 𝐲 ො 𝐱0 Theorem 1. 22 / 63
  • 23. ∇𝐱𝑡 log 𝑝(𝐱𝑡|𝐲) = ∇𝐱𝑡 log 𝑝(𝐱𝑡) + ∇𝐱𝑡 log 𝑝 𝐲 𝐱𝑡 Diffusion Posterior Sampling (DPS) ≃ ∇𝐱𝑡 log 𝑝 𝐱𝑡 + ∇𝐱𝑡 log 𝑝 𝐲 ො 𝐱0 ≃ 𝑠𝜃∗ 𝐱𝑡, 𝑡 + ∇𝐱𝑡 log 𝑝 𝐲 ො 𝐱0 ∇𝐱t log 𝑝 𝐲 ො 𝐱0 = −𝜌∇𝐱t ‖𝐲 − 𝓐(ො 𝐱0)‖2 2 1. Gaussian 2. Poisson ∇𝐱t log 𝑝 𝐲 ො 𝐱0 ≃ −𝜌∇𝐱t ‖𝐲 − 𝓐(ො 𝐱0)‖Λ 2 23 / 63
  • 24. Diffusion Posterior Sampling (DPS) - Algorithm 24 / 63
  • 25. Results: Super-resolution x16 Measurement (Gauss, 𝜎 = 0.05) DPS 25 / 63
  • 26. Results: Inpainting (92%) Measurement (Gauss, 𝜎 = 0.05) DPS 26 / 63
  • 27. Results: Gaussian deblurring Measurement (Gauss, 𝜎 = 0.05) DPS 27 / 63
  • 28. Results: Motion deblurring Measurement (Gauss, 𝜎 = 0.05) DPS 28 / 63
  • 29. Results: Phase retrieval Measurement (Gauss, 𝜎 = 0.05) DPS 29 / 63
  • 30. Results: Non-uniform deblurring Measurement (Gauss, 𝜎 = 0.05) DPS 30 / 63