SlideShare a Scribd company logo
2020/12/14
1
Learning with Unpaired Data
Jiebo Luo
University of Rochester
Keynote at ICMLA 2020
December 14, 2020
Introduction
● The recent successes by deep learning have been largely driven by
annotated big data
2020/12/14
2
Introduction
● Learning from more or less data
Learning from
Unpaired Data
Introduction
● The mapping from an input to a structured output is desired in ML
French→English
Une fusillade a eu lieu a`
l’aeroport international de
Los ´ Angeles.
There was a shooting in Los
Angeles International Airport.
Horse→Zebra Image→Caption
a black and white cat sitting
in a bathroom sink
2020/12/14
3
Introduction
● The mapping from an input to a structured output is desired in ML & applications
Image Enhancement Image Dehazing Artifact Reduction
Introduction
● Usually, paired training data are collected to train such a mapping model
● Collecting paired data is expensive
Collecting paired data Possible Expensive
Machine Translation ✅ 💲
Horse→Zebra ❌
Image Captioning ✅ 💲
Image Enhancement ✅ 💲💲
Image Dehazing ❓ 💲💲💲
Artifact Reduction ❓ 💲💲💲
2020/12/14
4
Introduction
● Unpaired data are much easier to collect
○ No annotation needed, or
○ Light-weight annotation
Introduction
● The new challenge: how to align the unpaired data
○ Reconstruction loss
○ GAN loss
Domain 1 Domain 2
2020/12/14
5
Introduction
● Going into details
○ Machine Translation
○ Horse→Zebra
○ Image Captioning
○ Image Dehazing
○ Artifact Reduction
Unsupervised Neural Machine Translation (NMT)
● Key elements
○ Dual structure (same model for L1-L2 and L2-L1)
○ Shared encoder (same for L1/L2)
○ Fixed embeddings in the encoder (pre-trained cross-lingual embeddings vs. random initialization)
Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. "Unsupervised neural machine translation." arXiv preprint arXiv:1710.11041
(2017).
2020/12/14
6
On-the-fly back-translation
Guillaume Lample, Alexis Conneau, Ludovic Denoyer, and Marc'Aurelio Ranzato. "Unsupervised machine translation using monolingual corpora
only." arXiv preprint arXiv:1711.00043 (2017).
Results
● BLEU scores on newstest2014
Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. "Unsupervised neural machine translation." arXiv preprint arXiv:1710.11041
(2017).
2020/12/14
7
Cycle GAN
● It is (nearly) impossible to collect paired images in some cases
Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial
networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
Cycle GAN
Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial
networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
2020/12/14
8
Cycle GAN
Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial
networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
● Adversarial Loss
● Cycle Consistency Loss
● Full Objective
Cycle GAN
Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial
networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
2020/12/14
9
Unsupervised Image Captioning
a. Supervised
b. Novel object
c. Domain adaptation
d. Pivot
e. Semi-supervised
f. Unsupervised
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
Unsupervised Image Captioning
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
2020/12/14
10
Unsupervised Image Captioning
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
Unsupervised Image Captioning
● Images are from MSCOCO dataset
● We crawl a large-scale image description corpus consisting of 2 million
natural sentences from Shutterstock to facilitate unsupervised image
captioning
● Object detectors trained on the OpenImage dataset are used
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
2020/12/14
11
Unsupervised Image Captioning
● Results
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
Unsupervised Image Captioning
● Results
Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and
pattern recognition, pp. 4125-4134. 2019.
2020/12/14
12
Image Dehazing
Atmosphere scattering model:
I(x) is the observed hazy image, J(x) is the scene radiance (haze-free image),
t(x) is the medium transmission map, A is the global atmospheric light.
Goal: recover J(x), t(x) and A from I(x).
Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In
AAAI, pp. 7485-7492. 2018.
Image Dehazing
● Uses synthetic hazy image pairs in
previous methods
● Generates realistic haze-free images
using only unpaired supervision in the
proposed method
Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In
AAAI, pp. 7485-7492. 2018.
2020/12/14
13
Image Dehazing
● Synthetic datasets
○ D-HAZY dataset
○ NYU-Depth dataset
● Real image datasets
○ HazyCity dataset
○ Crawled photos of Beijing
○ Annotating hazy or not
○ 845 natural hazy images
○ 1891 haze-free images
Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In
AAAI, pp. 7485-7492. 2018.
Image Dehazing
● Results
Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In
AAAI, pp. 7485-7492. 2018.
2020/12/14
14
Image Dehazing
● Results
Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In
AAAI, pp. 7485-7492. 2018.
Artifact Reduction
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
Model
LossUnpaired
clinical data
● Existing models require paired data
for training, which are synthesized
● When applied to clinical data,
models trained on synthesized data
generalize poorly
● We aim to learn metal artifact
reduction (MAR) directly from
clinical data without using paired
data
input output
artifact-free
2020/12/14
15
Artifact Disentanglement
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
decodingencoding
artifact reduced!
: artifact space
: content space
with artifact
bones, soft tissues,
lesions, implants, etc.
dark shadings,
streaks, noises, etc.
Artifact Disentanglement Network
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
2020/12/14
16
Artifact Disentanglement Network
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
content
encoder
content
decoder
Adversarial
Loss
artifact
reduced
with
artifact
without
artifact
Artifact Disentanglement Network
content
encoder
artifact
encoder
artifact &
content
decoder
Adversarial
Loss
with artifact
artifact
transferred
without
artifact
with
artifact
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
2020/12/14
17
Artifact Disentanglement Network
artifact &
content
decoder
self-
reconstruction
Reconstruction
Loss (L1)
artifact
encoder
content
encoder
with
artifact
with
artifact
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
Artifact Disentanglement Network
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
content
encoder
content
decoder
self-
reconstruction
Reconstruction
Loss (L1)
without
artifact
without
artifact
2020/12/14
18
Artifact Disentanglement Network
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
-
-
Artifact Consistency
Loss (L1)
Artifact Disentanglement Network
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
artifact
reduced
Self-reduction
Loss (L1)
artifact
transferred
without
artifact
with
artifact
without
artifact
2020/12/14
19
Quantitative Results on Synthetic Data
● Results
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
Qualitative Results on Clinical Data
Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE
Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
2020/12/14
20
Conclusions
● Learning with unpaired data can greatly reduce the effort in collecting data.
● Some level of explicit disentanglement of compounding components is the
key to improved performance beyond using cycle consistency.
● For machine translation and image captioning, the performance of unpaired
method is still lower than fully supervised methods.
● Compared with data synthesis, unpaired learning can achieve promising
results on many machine learning tasks.
Open Issues and Future Directions
● Learning with the combination of unpaired data and small paired data to
achieve better performance
● Going beyond text and image modalities, for example, to video and sound
modalities
2020/12/14
21
Acknowledgments
Yang Feng (Google Cloud) Haofu Liao (AWS)
https://www.yangfeng.name http://liaohaofu.com
Visual Intelligence and Social Multimedia Analytics

More Related Content

What's hot

SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~
SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~
SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~SSII
 
論文紹介 "DARTS: Differentiable Architecture Search"
論文紹介 "DARTS: Differentiable Architecture Search"論文紹介 "DARTS: Differentiable Architecture Search"
論文紹介 "DARTS: Differentiable Architecture Search"Yuta Koreeda
 
【DL輪読会】ViT + Self Supervised Learningまとめ
【DL輪読会】ViT + Self Supervised Learningまとめ【DL輪読会】ViT + Self Supervised Learningまとめ
【DL輪読会】ViT + Self Supervised LearningまとめDeep Learning JP
 
メタスタディ (Vision and Language)
メタスタディ (Vision and Language)メタスタディ (Vision and Language)
メタスタディ (Vision and Language)Shintaro Yamamoto
 
ECCV2020 Oral論文 完全読破(1/2)
ECCV2020 Oral論文 完全読破(1/2)ECCV2020 Oral論文 完全読破(1/2)
ECCV2020 Oral論文 完全読破(1/2)cvpaper. challenge
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Hwa Pyung Kim
 
【DL輪読会】WIRE: Wavelet Implicit Neural Representations
【DL輪読会】WIRE: Wavelet Implicit Neural Representations【DL輪読会】WIRE: Wavelet Implicit Neural Representations
【DL輪読会】WIRE: Wavelet Implicit Neural RepresentationsDeep Learning JP
 
[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions
[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions
[DL輪読会]Glow: Generative Flow with Invertible 1×1 ConvolutionsDeep Learning JP
 
Anchor free object detection by deep learning
Anchor free object detection by deep learningAnchor free object detection by deep learning
Anchor free object detection by deep learningYu Huang
 
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"Deep Learning JP
 
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)Yasunori Ozaki
 
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
 
生成モデルの Deep Learning
生成モデルの Deep Learning生成モデルの Deep Learning
生成モデルの Deep LearningSeiya Tokui
 
[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...Deep Learning JP
 
FPGAをロボット(ROS)で「やわらかく」使うには
FPGAをロボット(ROS)で「やわらかく」使うにはFPGAをロボット(ROS)で「やわらかく」使うには
FPGAをロボット(ROS)で「やわらかく」使うにはHideki Takase
 
視覚と対話の融合研究
視覚と対話の融合研究視覚と対話の融合研究
視覚と対話の融合研究Yoshitaka Ushiku
 
文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding
文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding
文献紹介:TSM: Temporal Shift Module for Efficient Video UnderstandingToru Tamaki
 
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)Preferred Networks
 

What's hot (20)

SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~
SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~
SSII2019OS: 深層学習にかかる時間を短くしてみませんか? ~分散学習の勧め~
 
論文紹介 "DARTS: Differentiable Architecture Search"
論文紹介 "DARTS: Differentiable Architecture Search"論文紹介 "DARTS: Differentiable Architecture Search"
論文紹介 "DARTS: Differentiable Architecture Search"
 
【DL輪読会】ViT + Self Supervised Learningまとめ
【DL輪読会】ViT + Self Supervised Learningまとめ【DL輪読会】ViT + Self Supervised Learningまとめ
【DL輪読会】ViT + Self Supervised Learningまとめ
 
メタスタディ (Vision and Language)
メタスタディ (Vision and Language)メタスタディ (Vision and Language)
メタスタディ (Vision and Language)
 
ECCV2020 Oral論文 完全読破(1/2)
ECCV2020 Oral論文 完全読破(1/2)ECCV2020 Oral論文 完全読破(1/2)
ECCV2020 Oral論文 完全読破(1/2)
 
Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)Tutorial on Object Detection (Faster R-CNN)
Tutorial on Object Detection (Faster R-CNN)
 
【DL輪読会】WIRE: Wavelet Implicit Neural Representations
【DL輪読会】WIRE: Wavelet Implicit Neural Representations【DL輪読会】WIRE: Wavelet Implicit Neural Representations
【DL輪読会】WIRE: Wavelet Implicit Neural Representations
 
[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions
[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions
[DL輪読会]Glow: Generative Flow with Invertible 1×1 Convolutions
 
Anchor free object detection by deep learning
Anchor free object detection by deep learningAnchor free object detection by deep learning
Anchor free object detection by deep learning
 
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
【DL輪読会】"Instant Neural Graphics Primitives with a Multiresolution Hash Encoding"
 
BERT+XLNet+RoBERTa
BERT+XLNet+RoBERTaBERT+XLNet+RoBERTa
BERT+XLNet+RoBERTa
 
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)
第四回 全日本CV勉強会スライド(MOTS: Multi-Object Tracking and Segmentation)
 
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)
 
生成モデルの Deep Learning
生成モデルの Deep Learning生成モデルの Deep Learning
生成モデルの Deep Learning
 
[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...
 
FPGAをロボット(ROS)で「やわらかく」使うには
FPGAをロボット(ROS)で「やわらかく」使うにはFPGAをロボット(ROS)で「やわらかく」使うには
FPGAをロボット(ROS)で「やわらかく」使うには
 
視覚と対話の融合研究
視覚と対話の融合研究視覚と対話の融合研究
視覚と対話の融合研究
 
文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding
文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding
文献紹介:TSM: Temporal Shift Module for Efficient Video Understanding
 
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)
Deep learningの発展と化学反応への応用 - 日本化学会第101春季大会(2021)
 
実装レベルで学ぶVQVAE
実装レベルで学ぶVQVAE実装レベルで学ぶVQVAE
実装レベルで学ぶVQVAE
 

Similar to Learning with Unpaired Data

Jia-Bin Huang's Curriculum Vitae
Jia-Bin Huang's Curriculum VitaeJia-Bin Huang's Curriculum Vitae
Jia-Bin Huang's Curriculum VitaeJia-Bin Huang
 
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...Tomoyuki Suzuki
 
Deep Learning Hardware: Past, Present, & Future
Deep Learning Hardware: Past, Present, & FutureDeep Learning Hardware: Past, Present, & Future
Deep Learning Hardware: Past, Present, & FutureRouyun Pan
 
Real Time Object Detection with Audio Feedback using Yolo v3
Real Time Object Detection with Audio Feedback using Yolo v3Real Time Object Detection with Audio Feedback using Yolo v3
Real Time Object Detection with Audio Feedback using Yolo v3ijtsrd
 
Generative Adversarial Networksの基礎と応用について
Generative Adversarial Networksの基礎と応用についてGenerative Adversarial Networksの基礎と応用について
Generative Adversarial Networksの基礎と応用についてSo Hasegawa
 
IRJET - Object Detection using Deep Learning with OpenCV and Python
IRJET - Object Detection using Deep Learning with OpenCV and PythonIRJET - Object Detection using Deep Learning with OpenCV and Python
IRJET - Object Detection using Deep Learning with OpenCV and PythonIRJET Journal
 
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...Universitat Politècnica de Catalunya
 
AI&BigData Lab. Артем Чернодуб "Распознавание изображений методом Lazy Deep ...
AI&BigData Lab. Артем Чернодуб  "Распознавание изображений методом Lazy Deep ...AI&BigData Lab. Артем Чернодуб  "Распознавание изображений методом Lazy Deep ...
AI&BigData Lab. Артем Чернодуб "Распознавание изображений методом Lazy Deep ...GeeksLab Odessa
 
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...IRJET Journal
 
CV_Lisongnan2015
CV_Lisongnan2015CV_Lisongnan2015
CV_Lisongnan2015Songnan Li
 
Visualizing Networks
Visualizing NetworksVisualizing Networks
Visualizing Networksfreshdatabos
 
Deep Learning for X ray Image to Text Generation
Deep Learning for X ray Image to Text GenerationDeep Learning for X ray Image to Text Generation
Deep Learning for X ray Image to Text Generationijtsrd
 
Visual geometry with deep learning
Visual geometry with deep learningVisual geometry with deep learning
Visual geometry with deep learningNAVER Engineering
 
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"SMART Infrastructure Facility
 
The power of_deep_learning_models_applications
The power of_deep_learning_models_applicationsThe power of_deep_learning_models_applications
The power of_deep_learning_models_applicationsDrjabez
 
PR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsPR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsjaewon lee
 
Content Based Image Retrieval (CBIR)
Content Based Image Retrieval (CBIR)Content Based Image Retrieval (CBIR)
Content Based Image Retrieval (CBIR)Behzad Shomali
 
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...GeeksLab Odessa
 

Similar to Learning with Unpaired Data (20)

Jia-Bin Huang's Curriculum Vitae
Jia-Bin Huang's Curriculum VitaeJia-Bin Huang's Curriculum Vitae
Jia-Bin Huang's Curriculum Vitae
 
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...
AET vs. AED: Unsupervised Representation Learning by Auto-Encoding Transforma...
 
Deep Learning Hardware: Past, Present, & Future
Deep Learning Hardware: Past, Present, & FutureDeep Learning Hardware: Past, Present, & Future
Deep Learning Hardware: Past, Present, & Future
 
Real Time Object Detection with Audio Feedback using Yolo v3
Real Time Object Detection with Audio Feedback using Yolo v3Real Time Object Detection with Audio Feedback using Yolo v3
Real Time Object Detection with Audio Feedback using Yolo v3
 
Generative Adversarial Networksの基礎と応用について
Generative Adversarial Networksの基礎と応用についてGenerative Adversarial Networksの基礎と応用について
Generative Adversarial Networksの基礎と応用について
 
Road Object Detection
Road Object DetectionRoad Object Detection
Road Object Detection
 
IRJET - Object Detection using Deep Learning with OpenCV and Python
IRJET - Object Detection using Deep Learning with OpenCV and PythonIRJET - Object Detection using Deep Learning with OpenCV and Python
IRJET - Object Detection using Deep Learning with OpenCV and Python
 
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...
Video Analysis with Convolutional Neural Networks (Master Computer Vision Bar...
 
Year-End Seminar 2018
Year-End Seminar 2018Year-End Seminar 2018
Year-End Seminar 2018
 
AI&BigData Lab. Артем Чернодуб "Распознавание изображений методом Lazy Deep ...
AI&BigData Lab. Артем Чернодуб  "Распознавание изображений методом Lazy Deep ...AI&BigData Lab. Артем Чернодуб  "Распознавание изображений методом Lazy Deep ...
AI&BigData Lab. Артем Чернодуб "Распознавание изображений методом Lazy Deep ...
 
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...
IRJET- A Real Time Yolo Human Detection in Flood Affected Areas based on Vide...
 
CV_Lisongnan2015
CV_Lisongnan2015CV_Lisongnan2015
CV_Lisongnan2015
 
Visualizing Networks
Visualizing NetworksVisualizing Networks
Visualizing Networks
 
Deep Learning for X ray Image to Text Generation
Deep Learning for X ray Image to Text GenerationDeep Learning for X ray Image to Text Generation
Deep Learning for X ray Image to Text Generation
 
Visual geometry with deep learning
Visual geometry with deep learningVisual geometry with deep learning
Visual geometry with deep learning
 
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"
SMART Seminar Series: "Smart Cities: The Good, The Bad & The Ugly"
 
The power of_deep_learning_models_applications
The power of_deep_learning_models_applicationsThe power of_deep_learning_models_applications
The power of_deep_learning_models_applications
 
PR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypointsPR-146: CornerNet detecting objects as paired keypoints
PR-146: CornerNet detecting objects as paired keypoints
 
Content Based Image Retrieval (CBIR)
Content Based Image Retrieval (CBIR)Content Based Image Retrieval (CBIR)
Content Based Image Retrieval (CBIR)
 
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...
AI&BigData Lab 2016. Артем Чернодуб: Обучение глубоких, очень глубоких и реку...
 

More from Goergen Institute for Data Science

When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...
When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...
When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...Goergen Institute for Data Science
 
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...Goergen Institute for Data Science
 
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...Goergen Institute for Data Science
 
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...Goergen Institute for Data Science
 

More from Goergen Institute for Data Science (14)

Vision and Language: Past, Present and Future
Vision and Language: Past, Present and FutureVision and Language: Past, Present and Future
Vision and Language: Past, Present and Future
 
How to properly review AI papers?
How to properly review AI papers?How to properly review AI papers?
How to properly review AI papers?
 
Video + Language 2019
Video + Language 2019Video + Language 2019
Video + Language 2019
 
Video + Language
Video + LanguageVideo + Language
Video + Language
 
When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...
When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...
When E-commerce Meets Social Media: Identifying Business on WeChat Moment Usi...
 
Computer Vision++: Where Do We Go from Here?
Computer Vision++: Where Do We Go from Here?Computer Vision++: Where Do We Go from Here?
Computer Vision++: Where Do We Go from Here?
 
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
 
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
 
Big Data Better Life
Big Data Better LifeBig Data Better Life
Big Data Better Life
 
Social Multimedia as Sensors
Social Multimedia as SensorsSocial Multimedia as Sensors
Social Multimedia as Sensors
 
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
 
Video+Language: From Classification to Description
Video+Language: From Classification to DescriptionVideo+Language: From Classification to Description
Video+Language: From Classification to Description
 
Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?
 
Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?
 

Recently uploaded

10 Differences between Sales Cloud and CPQ, Blanka Doktorová
10 Differences between Sales Cloud and CPQ, Blanka Doktorová10 Differences between Sales Cloud and CPQ, Blanka Doktorová
10 Differences between Sales Cloud and CPQ, Blanka DoktorováCzechDreamin
 
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya Halder
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya HalderCustom Approval Process: A New Perspective, Pavel Hrbacek & Anindya Halder
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya HalderCzechDreamin
 
AI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří KarpíšekAI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří KarpíšekCzechDreamin
 
Essentials of Automations: Optimizing FME Workflows with Parameters
Essentials of Automations: Optimizing FME Workflows with ParametersEssentials of Automations: Optimizing FME Workflows with Parameters
Essentials of Automations: Optimizing FME Workflows with ParametersSafe Software
 
"Impact of front-end architecture on development cost", Viktor Turskyi
"Impact of front-end architecture on development cost", Viktor Turskyi"Impact of front-end architecture on development cost", Viktor Turskyi
"Impact of front-end architecture on development cost", Viktor TurskyiFwdays
 
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...Product School
 
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...Sri Ambati
 
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered QualitySoftware Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered QualityInflectra
 
Salesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone KomSalesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone KomCzechDreamin
 
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)Julian Hyde
 
IESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIES VE
 
Mission to Decommission: Importance of Decommissioning Products to Increase E...
Mission to Decommission: Importance of Decommissioning Products to Increase E...Mission to Decommission: Importance of Decommissioning Products to Increase E...
Mission to Decommission: Importance of Decommissioning Products to Increase E...Product School
 
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...CzechDreamin
 
Powerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaPowerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaCzechDreamin
 
UiPath Test Automation using UiPath Test Suite series, part 2
UiPath Test Automation using UiPath Test Suite series, part 2UiPath Test Automation using UiPath Test Suite series, part 2
UiPath Test Automation using UiPath Test Suite series, part 2DianaGray10
 
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...CzechDreamin
 
Knowledge engineering: from people to machines and back
Knowledge engineering: from people to machines and backKnowledge engineering: from people to machines and back
Knowledge engineering: from people to machines and backElena Simperl
 
IoT Analytics Company Presentation May 2024
IoT Analytics Company Presentation May 2024IoT Analytics Company Presentation May 2024
IoT Analytics Company Presentation May 2024IoTAnalytics
 
Assuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyesAssuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyesThousandEyes
 
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Jeffrey Haguewood
 

Recently uploaded (20)

10 Differences between Sales Cloud and CPQ, Blanka Doktorová
10 Differences between Sales Cloud and CPQ, Blanka Doktorová10 Differences between Sales Cloud and CPQ, Blanka Doktorová
10 Differences between Sales Cloud and CPQ, Blanka Doktorová
 
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya Halder
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya HalderCustom Approval Process: A New Perspective, Pavel Hrbacek & Anindya Halder
Custom Approval Process: A New Perspective, Pavel Hrbacek & Anindya Halder
 
AI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří KarpíšekAI revolution and Salesforce, Jiří Karpíšek
AI revolution and Salesforce, Jiří Karpíšek
 
Essentials of Automations: Optimizing FME Workflows with Parameters
Essentials of Automations: Optimizing FME Workflows with ParametersEssentials of Automations: Optimizing FME Workflows with Parameters
Essentials of Automations: Optimizing FME Workflows with Parameters
 
"Impact of front-end architecture on development cost", Viktor Turskyi
"Impact of front-end architecture on development cost", Viktor Turskyi"Impact of front-end architecture on development cost", Viktor Turskyi
"Impact of front-end architecture on development cost", Viktor Turskyi
 
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
De-mystifying Zero to One: Design Informed Techniques for Greenfield Innovati...
 
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...
GenAISummit 2024 May 28 Sri Ambati Keynote: AGI Belongs to The Community in O...
 
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered QualitySoftware Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
Software Delivery At the Speed of AI: Inflectra Invests In AI-Powered Quality
 
Salesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone KomSalesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
Salesforce Adoption – Metrics, Methods, and Motivation, Antone Kom
 
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
Measures in SQL (a talk at SF Distributed Systems meetup, 2024-05-22)
 
IESVE for Early Stage Design and Planning
IESVE for Early Stage Design and PlanningIESVE for Early Stage Design and Planning
IESVE for Early Stage Design and Planning
 
Mission to Decommission: Importance of Decommissioning Products to Increase E...
Mission to Decommission: Importance of Decommissioning Products to Increase E...Mission to Decommission: Importance of Decommissioning Products to Increase E...
Mission to Decommission: Importance of Decommissioning Products to Increase E...
 
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...
SOQL 201 for Admins & Developers: Slice & Dice Your Org’s Data With Aggregate...
 
Powerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara LaskowskaPowerful Start- the Key to Project Success, Barbara Laskowska
Powerful Start- the Key to Project Success, Barbara Laskowska
 
UiPath Test Automation using UiPath Test Suite series, part 2
UiPath Test Automation using UiPath Test Suite series, part 2UiPath Test Automation using UiPath Test Suite series, part 2
UiPath Test Automation using UiPath Test Suite series, part 2
 
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
Integrating Telephony Systems with Salesforce: Insights and Considerations, B...
 
Knowledge engineering: from people to machines and back
Knowledge engineering: from people to machines and backKnowledge engineering: from people to machines and back
Knowledge engineering: from people to machines and back
 
IoT Analytics Company Presentation May 2024
IoT Analytics Company Presentation May 2024IoT Analytics Company Presentation May 2024
IoT Analytics Company Presentation May 2024
 
Assuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyesAssuring Contact Center Experiences for Your Customers With ThousandEyes
Assuring Contact Center Experiences for Your Customers With ThousandEyes
 
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
Slack (or Teams) Automation for Bonterra Impact Management (fka Social Soluti...
 

Learning with Unpaired Data

  • 1. 2020/12/14 1 Learning with Unpaired Data Jiebo Luo University of Rochester Keynote at ICMLA 2020 December 14, 2020 Introduction ● The recent successes by deep learning have been largely driven by annotated big data
  • 2. 2020/12/14 2 Introduction ● Learning from more or less data Learning from Unpaired Data Introduction ● The mapping from an input to a structured output is desired in ML French→English Une fusillade a eu lieu a` l’aeroport international de Los ´ Angeles. There was a shooting in Los Angeles International Airport. Horse→Zebra Image→Caption a black and white cat sitting in a bathroom sink
  • 3. 2020/12/14 3 Introduction ● The mapping from an input to a structured output is desired in ML & applications Image Enhancement Image Dehazing Artifact Reduction Introduction ● Usually, paired training data are collected to train such a mapping model ● Collecting paired data is expensive Collecting paired data Possible Expensive Machine Translation ✅ 💲 Horse→Zebra ❌ Image Captioning ✅ 💲 Image Enhancement ✅ 💲💲 Image Dehazing ❓ 💲💲💲 Artifact Reduction ❓ 💲💲💲
  • 4. 2020/12/14 4 Introduction ● Unpaired data are much easier to collect ○ No annotation needed, or ○ Light-weight annotation Introduction ● The new challenge: how to align the unpaired data ○ Reconstruction loss ○ GAN loss Domain 1 Domain 2
  • 5. 2020/12/14 5 Introduction ● Going into details ○ Machine Translation ○ Horse→Zebra ○ Image Captioning ○ Image Dehazing ○ Artifact Reduction Unsupervised Neural Machine Translation (NMT) ● Key elements ○ Dual structure (same model for L1-L2 and L2-L1) ○ Shared encoder (same for L1/L2) ○ Fixed embeddings in the encoder (pre-trained cross-lingual embeddings vs. random initialization) Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. "Unsupervised neural machine translation." arXiv preprint arXiv:1710.11041 (2017).
  • 6. 2020/12/14 6 On-the-fly back-translation Guillaume Lample, Alexis Conneau, Ludovic Denoyer, and Marc'Aurelio Ranzato. "Unsupervised machine translation using monolingual corpora only." arXiv preprint arXiv:1711.00043 (2017). Results ● BLEU scores on newstest2014 Mikel Artetxe, Gorka Labaka, Eneko Agirre, and Kyunghyun Cho. "Unsupervised neural machine translation." arXiv preprint arXiv:1710.11041 (2017).
  • 7. 2020/12/14 7 Cycle GAN ● It is (nearly) impossible to collect paired images in some cases Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017. Cycle GAN Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
  • 8. 2020/12/14 8 Cycle GAN Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017. ● Adversarial Loss ● Cycle Consistency Loss ● Full Objective Cycle GAN Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. "Unpaired image-to-image translation using cycle-consistent adversarial networks." In Proceedings of the IEEE international conference on computer vision, pp. 2223-2232. 2017.
  • 9. 2020/12/14 9 Unsupervised Image Captioning a. Supervised b. Novel object c. Domain adaptation d. Pivot e. Semi-supervised f. Unsupervised Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019. Unsupervised Image Captioning Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019.
  • 10. 2020/12/14 10 Unsupervised Image Captioning Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019. Unsupervised Image Captioning ● Images are from MSCOCO dataset ● We crawl a large-scale image description corpus consisting of 2 million natural sentences from Shutterstock to facilitate unsupervised image captioning ● Object detectors trained on the OpenImage dataset are used Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019.
  • 11. 2020/12/14 11 Unsupervised Image Captioning ● Results Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019. Unsupervised Image Captioning ● Results Yang Feng, Lin Ma, Wei Liu, and Jiebo Luo. "Unsupervised image captioning." In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4125-4134. 2019.
  • 12. 2020/12/14 12 Image Dehazing Atmosphere scattering model: I(x) is the observed hazy image, J(x) is the scene radiance (haze-free image), t(x) is the medium transmission map, A is the global atmospheric light. Goal: recover J(x), t(x) and A from I(x). Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In AAAI, pp. 7485-7492. 2018. Image Dehazing ● Uses synthetic hazy image pairs in previous methods ● Generates realistic haze-free images using only unpaired supervision in the proposed method Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In AAAI, pp. 7485-7492. 2018.
  • 13. 2020/12/14 13 Image Dehazing ● Synthetic datasets ○ D-HAZY dataset ○ NYU-Depth dataset ● Real image datasets ○ HazyCity dataset ○ Crawled photos of Beijing ○ Annotating hazy or not ○ 845 natural hazy images ○ 1891 haze-free images Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In AAAI, pp. 7485-7492. 2018. Image Dehazing ● Results Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In AAAI, pp. 7485-7492. 2018.
  • 14. 2020/12/14 14 Image Dehazing ● Results Xitong Yang, Zheng Xu, and Jiebo Luo. "Towards Perceptual Image Dehazing by Physics-Based Disentanglement and Adversarial Training." In AAAI, pp. 7485-7492. 2018. Artifact Reduction Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. Model LossUnpaired clinical data ● Existing models require paired data for training, which are synthesized ● When applied to clinical data, models trained on synthesized data generalize poorly ● We aim to learn metal artifact reduction (MAR) directly from clinical data without using paired data input output artifact-free
  • 15. 2020/12/14 15 Artifact Disentanglement Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. decodingencoding artifact reduced! : artifact space : content space with artifact bones, soft tissues, lesions, implants, etc. dark shadings, streaks, noises, etc. Artifact Disentanglement Network Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
  • 16. 2020/12/14 16 Artifact Disentanglement Network Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. content encoder content decoder Adversarial Loss artifact reduced with artifact without artifact Artifact Disentanglement Network content encoder artifact encoder artifact & content decoder Adversarial Loss with artifact artifact transferred without artifact with artifact Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
  • 17. 2020/12/14 17 Artifact Disentanglement Network artifact & content decoder self- reconstruction Reconstruction Loss (L1) artifact encoder content encoder with artifact with artifact Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. Artifact Disentanglement Network Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. content encoder content decoder self- reconstruction Reconstruction Loss (L1) without artifact without artifact
  • 18. 2020/12/14 18 Artifact Disentanglement Network Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. - - Artifact Consistency Loss (L1) Artifact Disentanglement Network Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. artifact reduced Self-reduction Loss (L1) artifact transferred without artifact with artifact without artifact
  • 19. 2020/12/14 19 Quantitative Results on Synthetic Data ● Results Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643. Qualitative Results on Clinical Data Haofu Liao, Wei-An Lin, S. Kevin Zhou, and Jiebo Luo. "ADN: Artifact disentanglement network for unsupervised metal artifact reduction." IEEE Transactions on Medical Imaging 39, no. 3 (2019): 634-643.
  • 20. 2020/12/14 20 Conclusions ● Learning with unpaired data can greatly reduce the effort in collecting data. ● Some level of explicit disentanglement of compounding components is the key to improved performance beyond using cycle consistency. ● For machine translation and image captioning, the performance of unpaired method is still lower than fully supervised methods. ● Compared with data synthesis, unpaired learning can achieve promising results on many machine learning tasks. Open Issues and Future Directions ● Learning with the combination of unpaired data and small paired data to achieve better performance ● Going beyond text and image modalities, for example, to video and sound modalities
  • 21. 2020/12/14 21 Acknowledgments Yang Feng (Google Cloud) Haofu Liao (AWS) https://www.yangfeng.name http://liaohaofu.com Visual Intelligence and Social Multimedia Analytics