SlideShare a Scribd company logo
1 of 27
2020 격변하는 SSL
After SimCLR
ImageNet Supervised를 이긴 이후에
> Self Supervised 2020년부터
SimCLR, MOCO : Imagenet SL에 근접한 Self Supervised 할아버지들 (2020 초)
BYOL, SWAV : 할배들 이겨보자~ 방법론 이것저것 써본다 (2020 중)
BT, SimSiam : SSL 하는데 너무 트레이닝 트릭들이 많아서 귀찮고! Computing Power 너무 많이 쓰는 듯? 못
살겠다 줄여보자! (2020 말, 2021 초)
PAWS : 겟 서거라 Semi Supervised도 나가신다~
LeCun, Kaming He, Hinton etc
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
1. Supervised를 따라잡기 시작
2. Batch 내에서 Positive-Negative Pair
3. Nonlinear Projection Head
의의
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
방법론
Contrastive Learning
- Data Augmentation ( Generate Pair )
- Similarity Loss
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
방법론
Contrastive Learning - Data Augmentation ( Generate Pair )
1. multiple transform이
single transform 보다
acc이 높음
2. color distortion은
self-supervised에서 긍
정적인 영향
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
방법론
Contrastive Learning - Data Augmentation ( Generate Pair )
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
방법론
Contrastive Learning - Similarity Loss (NT-Xent loss)
SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020)
한계
1. 엄청난 Computation ( Batch Size )
2. Negative Pair가 항상 Negative Pair를 보장하지 않음 (Cat1
- Cat2 페어도 Negative로 계산)
3. 비교적 적은 데이터셋 (CIFAR)에서는 유사하거나 낮음 -
Linear Protocol
4. Resnet-50를 (width) 4x 키워 큰 모델에서 사용
MOCO : A Simple Framework for Contrastive Learning of Visual Representations (2020)
의의
1. Supervised에는 못 미쳤지만… FB도 놀고 있지 않았다!
+ Segmentation/Detection COCO, VOC는 이겼다!
MOCO : A Simple Framework for Contrastive Learning of Visual Representations (2020)
방법론
Positive/Negative Pair
- Batch : Batch Size에 종속
- Memory Bank : pre-calculated feature라서… inconsistency 문제 있음
=> Dynamic Dictionary
- Dictionary as a queue
: Memory Bank Queue를 Mini Batch Size보다는 크게 유지하며
Feature를 저장해둔다
- Momentum update
: Key Encoder가 Query Encoder보다 천천히 업데이트 되도록 함
MOCO… updates
방법론
v2
1. Feature Head를 SimCLR처럼 MLP non linear 추가해서 학습
2. Blur Augmentation 추가
3. Cosine Learning Rate decay 추가
v3 (2021)
1. Dynamic Dictionary를 큰 Batch Size로 대체
~ batch size에 따라 적절한 lr을 찾고, AdamW~LAMB Opt 사용
2. Resnet 말고 Transformer (ViT)에 사용함
3. ViT의 학습 불안정성을 Fixed Random patch projection으로 해결
BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020)
의의
DeepMind도 안 잔다!
1. Contrastive Learning에서는 충분한 Negative Pair가 성능을
좌우하여… Negative Pair을 안쓰는 방법을 고민해봄
2. SimCLR에 비교해 BatchSize와 Transform에 “덜” 민감함
https://hoya012.github.io/blog/byol/
BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020)
방법론
https://hoya012.github.io/blog/byol/
data aug
encoder mlp
mlp
Online
Network
Target
Network
Target Network는 Online Network를 Momentum update를 통해
Following 함 == Exponential Moving Average
BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020)
방법론
https://hoya012.github.io/blog/byol/
BYOL Moving Average
MOCO Momentum
BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020)
방법론
https://hoya012.github.io/blog/byol/
Loss == MSE Loss
Online-Target Input을 교체한
Symmetric Loss 추가
BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020)
한계
https://hoya012.github.io/blog/byol/
1. 그래도 큰 Batch Size (4096)와 모델
2. 여전히 작은 데이터셋에서는 비교적 SL보다 떨어짐 -
Linear Protocal
SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020)
의의
1. 일단… 또 SOTA
2. 비교적 작은 배치 (256)에서 SimCLR를 이김
3. Single Image Random Crop이 가지고 있던 View 수가 2개 뿐
이 였던 한계를 Multi-Crop으로 성능 향상
Swapping으로 Clustering을 통해 Feature를 학습하는 시도
SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020)
방법론
1. Feature to Prototype vector
2. Swap Assignment
SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020)
방법론
Feature to Prototype vector & Loss
-> z는 feature vector, q는 prototype vector
SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020)
방법론
Multi Crop
- SimCLR 같은 논문에서는 224
Random Crop으로 1 Pair만을 만
들었지만, 96 Random Crop을 한
이미지에서 N회 추가로 하여 더 많
은 Pair를 만들어낼 수 있다
https://wandb.ai/authors/swav-tf/reports/Unsupervised-Visual-Representation-Learning-with-SwAV--VmlldzoyMjg3Mzg
SimSiam: Exploring Simple Siamese Representation Learning (2020)
의의
1. SimCLR, MOCO 같이 자꾸 negative sample 쓰는건 너무 힘
들어! - 클러스터링, moving average 다 없이 할거야!
= Stop-Grad + 샴 구조로 SSL 하기
2. Batch Size도 안 커도됨! (64부터 실험해봄 ㅋ), epoch도 작음
https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
SimSiam: Exploring Simple Siamese Representation Learning (2020)
방법론
https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
cosine sim이 loss
data augmentation
- RandomResized Crop 224
- Flip, ColorJitter, GrayScale
SimSiam: Exploring Simple Siamese Representation Learning (2020)
방법론
https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
Stop Gradient 효과
SimSiam: Exploring Simple Siamese Representation Learning (2020)
한계
https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021)
의의
LeCun의 논문
1. Feature Cross Correlation Matrix를 통해 Loss 계산
2. Batch 사이즈가 비교적 작고, moving average, 추가 테크닉들
이 없이 high-dimension vector 학습에 성공함
3. 음 이것도 negative가 없어도 되네요!
간단한데 잘되네….?
페북 코드보면 걍 main.py 하나에 다 때려박음 300줄 ㄷ
BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021)
방법론
Feature를 0~1 Normalize 후에 Cross-Corr 계산해서 Identical Vector Target으로 Loss 계산함
Transform은 Flip, Color Jitter, Grayscale, Blurring, Solarization
BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021)
방법론

More Related Content

What's hot

0から理解するニューラルネットアーキテクチャサーチ(NAS)
0から理解するニューラルネットアーキテクチャサーチ(NAS)0から理解するニューラルネットアーキテクチャサーチ(NAS)
0から理解するニューラルネットアーキテクチャサーチ(NAS)
MasanoriSuganuma
 

What's hot (20)

상상을 현실로 만드는, 이미지 생성 모델을 위한 엔지니어링
상상을 현실로 만드는, 이미지 생성 모델을 위한 엔지니어링상상을 현실로 만드는, 이미지 생성 모델을 위한 엔지니어링
상상을 현실로 만드는, 이미지 생성 모델을 위한 엔지니어링
 
Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩Deep Learningによる超解像の進歩
Deep Learningによる超解像の進歩
 
[DL輪読会]Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Ima...
[DL輪読会]Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Ima...[DL輪読会]Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Ima...
[DL輪読会]Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Ima...
 
SIGNATE 鰹節コンペ2nd Place Solution
SIGNATE 鰹節コンペ2nd Place SolutionSIGNATE 鰹節コンペ2nd Place Solution
SIGNATE 鰹節コンペ2nd Place Solution
 
[DL輪読会]NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
[DL輪読会]NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis[DL輪読会]NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
[DL輪読会]NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
 
[解説スライド] NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
[解説スライド] NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis[解説スライド] NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
[解説スライド] NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
 
Deep-Learning Based Stereo Super-Resolution
Deep-Learning Based Stereo Super-ResolutionDeep-Learning Based Stereo Super-Resolution
Deep-Learning Based Stereo Super-Resolution
 
십분딥러닝_17_DIM(Deep InfoMax)
십분딥러닝_17_DIM(Deep InfoMax)십분딥러닝_17_DIM(Deep InfoMax)
십분딥러닝_17_DIM(Deep InfoMax)
 
Lucas kanade法について
Lucas kanade法についてLucas kanade法について
Lucas kanade法について
 
【DL輪読会】Efficiently Modeling Long Sequences with Structured State Spaces
【DL輪読会】Efficiently Modeling Long Sequences with Structured State Spaces【DL輪読会】Efficiently Modeling Long Sequences with Structured State Spaces
【DL輪読会】Efficiently Modeling Long Sequences with Structured State Spaces
 
RoFormer: Enhanced Transformer with Rotary Position Embedding
RoFormer: Enhanced Transformer with Rotary Position EmbeddingRoFormer: Enhanced Transformer with Rotary Position Embedding
RoFormer: Enhanced Transformer with Rotary Position Embedding
 
0から理解するニューラルネットアーキテクチャサーチ(NAS)
0から理解するニューラルネットアーキテクチャサーチ(NAS)0から理解するニューラルネットアーキテクチャサーチ(NAS)
0から理解するニューラルネットアーキテクチャサーチ(NAS)
 
動画認識サーベイv1(メタサーベイ )
動画認識サーベイv1(メタサーベイ )動画認識サーベイv1(メタサーベイ )
動画認識サーベイv1(メタサーベイ )
 
Brief intro : Invariance and Equivariance
Brief intro : Invariance and EquivarianceBrief intro : Invariance and Equivariance
Brief intro : Invariance and Equivariance
 
[DL輪読会] Spectral Norm Regularization for Improving the Generalizability of De...
[DL輪読会] Spectral Norm Regularization for Improving the Generalizability of De...[DL輪読会] Spectral Norm Regularization for Improving the Generalizability of De...
[DL輪読会] Spectral Norm Regularization for Improving the Generalizability of De...
 
Data-Centric AIの紹介
Data-Centric AIの紹介Data-Centric AIの紹介
Data-Centric AIの紹介
 
論文紹介 Pixel Recurrent Neural Networks
論文紹介 Pixel Recurrent Neural Networks論文紹介 Pixel Recurrent Neural Networks
論文紹介 Pixel Recurrent Neural Networks
 
Domain Adaptive Faster R-CNN for Object Detection in the Wild 論文紹介
Domain Adaptive Faster R-CNN for Object Detection in the Wild 論文紹介Domain Adaptive Faster R-CNN for Object Detection in the Wild 論文紹介
Domain Adaptive Faster R-CNN for Object Detection in the Wild 論文紹介
 
畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向畳み込みニューラルネットワークの研究動向
畳み込みニューラルネットワークの研究動向
 
論文要約:AUGMIX: A SIMPLE DATA PROCESSING METHOD TO IMPROVE ROBUSTNESS AND UNCERT...
論文要約:AUGMIX: A SIMPLE DATA PROCESSING METHOD TO IMPROVE ROBUSTNESS AND UNCERT...論文要約:AUGMIX: A SIMPLE DATA PROCESSING METHOD TO IMPROVE ROBUSTNESS AND UNCERT...
論文要約:AUGMIX: A SIMPLE DATA PROCESSING METHOD TO IMPROVE ROBUSTNESS AND UNCERT...
 

Similar to 2020 > Self supervised learning

[Paper Review] Image captioning with semantic attention
[Paper Review] Image captioning with semantic attention[Paper Review] Image captioning with semantic attention
[Paper Review] Image captioning with semantic attention
Hyeongmin Lee
 
임태현, MMO 서버 개발 포스트 모템, NDC2012
임태현, MMO 서버 개발 포스트 모템, NDC2012임태현, MMO 서버 개발 포스트 모템, NDC2012
임태현, MMO 서버 개발 포스트 모템, NDC2012
devCAT Studio, NEXON
 

Similar to 2020 > Self supervised learning (20)

비동기 어플리케이션 모니터링과 밀당하기
비동기 어플리케이션 모니터링과 밀당하기비동기 어플리케이션 모니터링과 밀당하기
비동기 어플리케이션 모니터링과 밀당하기
 
Segment Anything
Segment AnythingSegment Anything
Segment Anything
 
Deep learning overview
Deep learning overviewDeep learning overview
Deep learning overview
 
Self Supervised Learning 세미나.pptx
Self Supervised Learning 세미나.pptxSelf Supervised Learning 세미나.pptx
Self Supervised Learning 세미나.pptx
 
2019 5-5-week-i-learned-generative model
2019 5-5-week-i-learned-generative model2019 5-5-week-i-learned-generative model
2019 5-5-week-i-learned-generative model
 
Create App Easier With SVC Pattern - DroidKnights 2019 @Seoul
Create App Easier With SVC Pattern - DroidKnights 2019 @SeoulCreate App Easier With SVC Pattern - DroidKnights 2019 @Seoul
Create App Easier With SVC Pattern - DroidKnights 2019 @Seoul
 
2021-11-16 모두콘 딥러닝 경량화 발표
2021-11-16 모두콘 딥러닝 경량화 발표2021-11-16 모두콘 딥러닝 경량화 발표
2021-11-16 모두콘 딥러닝 경량화 발표
 
머신러닝으로 쏟아지는 유저 CS 답변하기 DEVIEW 2017
머신러닝으로 쏟아지는 유저 CS 답변하기 DEVIEW 2017머신러닝으로 쏟아지는 유저 CS 답변하기 DEVIEW 2017
머신러닝으로 쏟아지는 유저 CS 답변하기 DEVIEW 2017
 
[Paper Review] Image captioning with semantic attention
[Paper Review] Image captioning with semantic attention[Paper Review] Image captioning with semantic attention
[Paper Review] Image captioning with semantic attention
 
CS231n chap12_Visualization and Understand Summary
CS231n chap12_Visualization and Understand SummaryCS231n chap12_Visualization and Understand Summary
CS231n chap12_Visualization and Understand Summary
 
Vip detection sensor
Vip detection sensorVip detection sensor
Vip detection sensor
 
딥러닝 논문읽기 efficient netv2 논문리뷰
딥러닝 논문읽기 efficient netv2  논문리뷰딥러닝 논문읽기 efficient netv2  논문리뷰
딥러닝 논문읽기 efficient netv2 논문리뷰
 
PYCON KR 2017 - 구름이 하늘의 일이라면 (윤상웅)
PYCON KR 2017 - 구름이 하늘의 일이라면 (윤상웅)PYCON KR 2017 - 구름이 하늘의 일이라면 (윤상웅)
PYCON KR 2017 - 구름이 하늘의 일이라면 (윤상웅)
 
임태현, MMO 서버 개발 포스트 모템, NDC2012
임태현, MMO 서버 개발 포스트 모템, NDC2012임태현, MMO 서버 개발 포스트 모템, NDC2012
임태현, MMO 서버 개발 포스트 모템, NDC2012
 
[NEXT] Android Profiler 사용법
[NEXT] Android Profiler 사용법 [NEXT] Android Profiler 사용법
[NEXT] Android Profiler 사용법
 
AnoGAN을 이용한 철강 소재 결함 검출 AI
AnoGAN을 이용한 철강 소재 결함 검출 AIAnoGAN을 이용한 철강 소재 결함 검출 AI
AnoGAN을 이용한 철강 소재 결함 검출 AI
 
210801 hierarchical long term video frame prediction without supervision
210801 hierarchical long term video frame prediction without supervision210801 hierarchical long term video frame prediction without supervision
210801 hierarchical long term video frame prediction without supervision
 
Meteor를 통해서 개발하는 웹어플리케이션 서비스
Meteor를 통해서 개발하는 웹어플리케이션 서비스Meteor를 통해서 개발하는 웹어플리케이션 서비스
Meteor를 통해서 개발하는 웹어플리케이션 서비스
 
20210131deit-210204074124.pdf
20210131deit-210204074124.pdf20210131deit-210204074124.pdf
20210131deit-210204074124.pdf
 
Training data-efficient image transformers & distillation through attention
Training data-efficient image transformers & distillation through attentionTraining data-efficient image transformers & distillation through attention
Training data-efficient image transformers & distillation through attention
 

More from Dong Heon Cho

More from Dong Heon Cho (20)

Forward-Forward Algorithm
Forward-Forward AlgorithmForward-Forward Algorithm
Forward-Forward Algorithm
 
What is Texture.pdf
What is Texture.pdfWhat is Texture.pdf
What is Texture.pdf
 
BADGE
BADGEBADGE
BADGE
 
Neural Radiance Field
Neural Radiance FieldNeural Radiance Field
Neural Radiance Field
 
All about that pooling
All about that poolingAll about that pooling
All about that pooling
 
Background elimination review
Background elimination reviewBackground elimination review
Background elimination review
 
Transparent Latent GAN
Transparent Latent GANTransparent Latent GAN
Transparent Latent GAN
 
Image matting atoc
Image matting atocImage matting atoc
Image matting atoc
 
Multi object Deep reinforcement learning
Multi object Deep reinforcement learningMulti object Deep reinforcement learning
Multi object Deep reinforcement learning
 
Multi agent reinforcement learning for sequential social dilemmas
Multi agent reinforcement learning for sequential social dilemmasMulti agent reinforcement learning for sequential social dilemmas
Multi agent reinforcement learning for sequential social dilemmas
 
Multi agent System
Multi agent SystemMulti agent System
Multi agent System
 
Hybrid reward architecture
Hybrid reward architectureHybrid reward architecture
Hybrid reward architecture
 
Use Jupyter notebook guide in 5 minutes
Use Jupyter notebook guide in 5 minutesUse Jupyter notebook guide in 5 minutes
Use Jupyter notebook guide in 5 minutes
 
AlexNet and so on...
AlexNet and so on...AlexNet and so on...
AlexNet and so on...
 
Deep Learning AtoC with Image Perspective
Deep Learning AtoC with Image PerspectiveDeep Learning AtoC with Image Perspective
Deep Learning AtoC with Image Perspective
 
LOL win prediction
LOL win predictionLOL win prediction
LOL win prediction
 
How can we train with few data
How can we train with few dataHow can we train with few data
How can we train with few data
 
Domain adaptation gan
Domain adaptation ganDomain adaptation gan
Domain adaptation gan
 
Dense sparse-dense training for dnn and Other Models
Dense sparse-dense training for dnn and Other ModelsDense sparse-dense training for dnn and Other Models
Dense sparse-dense training for dnn and Other Models
 
Squeeeze models
Squeeeze modelsSqueeeze models
Squeeeze models
 

2020 > Self supervised learning

  • 1. 2020 격변하는 SSL After SimCLR ImageNet Supervised를 이긴 이후에
  • 2. > Self Supervised 2020년부터 SimCLR, MOCO : Imagenet SL에 근접한 Self Supervised 할아버지들 (2020 초) BYOL, SWAV : 할배들 이겨보자~ 방법론 이것저것 써본다 (2020 중) BT, SimSiam : SSL 하는데 너무 트레이닝 트릭들이 많아서 귀찮고! Computing Power 너무 많이 쓰는 듯? 못 살겠다 줄여보자! (2020 말, 2021 초) PAWS : 겟 서거라 Semi Supervised도 나가신다~ LeCun, Kaming He, Hinton etc
  • 3. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 1. Supervised를 따라잡기 시작 2. Batch 내에서 Positive-Negative Pair 3. Nonlinear Projection Head 의의
  • 4. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 방법론 Contrastive Learning - Data Augmentation ( Generate Pair ) - Similarity Loss
  • 5. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 방법론 Contrastive Learning - Data Augmentation ( Generate Pair ) 1. multiple transform이 single transform 보다 acc이 높음 2. color distortion은 self-supervised에서 긍 정적인 영향
  • 6. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 방법론 Contrastive Learning - Data Augmentation ( Generate Pair )
  • 7. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 방법론 Contrastive Learning - Similarity Loss (NT-Xent loss)
  • 8. SimCLR: A Simple Framework for Contrastive Learning of Visual Representations (2020) 한계 1. 엄청난 Computation ( Batch Size ) 2. Negative Pair가 항상 Negative Pair를 보장하지 않음 (Cat1 - Cat2 페어도 Negative로 계산) 3. 비교적 적은 데이터셋 (CIFAR)에서는 유사하거나 낮음 - Linear Protocol 4. Resnet-50를 (width) 4x 키워 큰 모델에서 사용
  • 9. MOCO : A Simple Framework for Contrastive Learning of Visual Representations (2020) 의의 1. Supervised에는 못 미쳤지만… FB도 놀고 있지 않았다! + Segmentation/Detection COCO, VOC는 이겼다!
  • 10. MOCO : A Simple Framework for Contrastive Learning of Visual Representations (2020) 방법론 Positive/Negative Pair - Batch : Batch Size에 종속 - Memory Bank : pre-calculated feature라서… inconsistency 문제 있음 => Dynamic Dictionary - Dictionary as a queue : Memory Bank Queue를 Mini Batch Size보다는 크게 유지하며 Feature를 저장해둔다 - Momentum update : Key Encoder가 Query Encoder보다 천천히 업데이트 되도록 함
  • 11. MOCO… updates 방법론 v2 1. Feature Head를 SimCLR처럼 MLP non linear 추가해서 학습 2. Blur Augmentation 추가 3. Cosine Learning Rate decay 추가 v3 (2021) 1. Dynamic Dictionary를 큰 Batch Size로 대체 ~ batch size에 따라 적절한 lr을 찾고, AdamW~LAMB Opt 사용 2. Resnet 말고 Transformer (ViT)에 사용함 3. ViT의 학습 불안정성을 Fixed Random patch projection으로 해결
  • 12. BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020) 의의 DeepMind도 안 잔다! 1. Contrastive Learning에서는 충분한 Negative Pair가 성능을 좌우하여… Negative Pair을 안쓰는 방법을 고민해봄 2. SimCLR에 비교해 BatchSize와 Transform에 “덜” 민감함 https://hoya012.github.io/blog/byol/
  • 13. BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020) 방법론 https://hoya012.github.io/blog/byol/ data aug encoder mlp mlp Online Network Target Network Target Network는 Online Network를 Momentum update를 통해 Following 함 == Exponential Moving Average
  • 14. BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020) 방법론 https://hoya012.github.io/blog/byol/ BYOL Moving Average MOCO Momentum
  • 15. BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020) 방법론 https://hoya012.github.io/blog/byol/ Loss == MSE Loss Online-Target Input을 교체한 Symmetric Loss 추가
  • 16. BYOL: Bootstrap Your Own Latent A New Approach to Self-Supervised Learning (2020) 한계 https://hoya012.github.io/blog/byol/ 1. 그래도 큰 Batch Size (4096)와 모델 2. 여전히 작은 데이터셋에서는 비교적 SL보다 떨어짐 - Linear Protocal
  • 17. SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020) 의의 1. 일단… 또 SOTA 2. 비교적 작은 배치 (256)에서 SimCLR를 이김 3. Single Image Random Crop이 가지고 있던 View 수가 2개 뿐 이 였던 한계를 Multi-Crop으로 성능 향상 Swapping으로 Clustering을 통해 Feature를 학습하는 시도
  • 18. SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020) 방법론 1. Feature to Prototype vector 2. Swap Assignment
  • 19. SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020) 방법론 Feature to Prototype vector & Loss -> z는 feature vector, q는 prototype vector
  • 20. SWAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments (2020) 방법론 Multi Crop - SimCLR 같은 논문에서는 224 Random Crop으로 1 Pair만을 만 들었지만, 96 Random Crop을 한 이미지에서 N회 추가로 하여 더 많 은 Pair를 만들어낼 수 있다 https://wandb.ai/authors/swav-tf/reports/Unsupervised-Visual-Representation-Learning-with-SwAV--VmlldzoyMjg3Mzg
  • 21. SimSiam: Exploring Simple Siamese Representation Learning (2020) 의의 1. SimCLR, MOCO 같이 자꾸 negative sample 쓰는건 너무 힘 들어! - 클러스터링, moving average 다 없이 할거야! = Stop-Grad + 샴 구조로 SSL 하기 2. Batch Size도 안 커도됨! (64부터 실험해봄 ㅋ), epoch도 작음 https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
  • 22. SimSiam: Exploring Simple Siamese Representation Learning (2020) 방법론 https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e cosine sim이 loss data augmentation - RandomResized Crop 224 - Flip, ColorJitter, GrayScale
  • 23. SimSiam: Exploring Simple Siamese Representation Learning (2020) 방법론 https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e Stop Gradient 효과
  • 24. SimSiam: Exploring Simple Siamese Representation Learning (2020) 한계 https://medium.com/analytics-vidhya/simsiam-paper-explained-df3577cab49e
  • 25. BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021) 의의 LeCun의 논문 1. Feature Cross Correlation Matrix를 통해 Loss 계산 2. Batch 사이즈가 비교적 작고, moving average, 추가 테크닉들 이 없이 high-dimension vector 학습에 성공함 3. 음 이것도 negative가 없어도 되네요! 간단한데 잘되네….? 페북 코드보면 걍 main.py 하나에 다 때려박음 300줄 ㄷ
  • 26. BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021) 방법론 Feature를 0~1 Normalize 후에 Cross-Corr 계산해서 Identical Vector Target으로 Loss 계산함 Transform은 Flip, Color Jitter, Grayscale, Blurring, Solarization
  • 27. BT: Barlow Twins: Self-Supervised Learning via Redundancy Reduction (2021) 방법론