1

Meta-Transfer Learning for Few-Shot Learning Qianru Sun 1,3* Yaoyao Liu 1,2* Tat-Seng Chua 1 Bernt Schiele 3 1 National University of Singapore, 2 Tianjin University, 3 Max Planck Institute for Informatics Motivation & Contributions Meta-Transfer Learning Hard Task Meta-Batch ● Few-shot learning is challenging due to the lack of training data. ● Re-thinking promising methods: ○ deep neural networks (DNN) ○ transfer learning (pre-train, fine-tune) ○ meta-learning (meta gradient descent) ● Problem: “catastrophic forgetting” ● Our solution: Scaling & Shifting (SS) ● Problem: slow meta-training convergence ● Our solution: hard negative task sampling Detail Top performance is achieved! Faster convergence is achieved! Detail ● miniImageNet dataset Few-shot CIFAR-100 (FC100) dataset method backbone 1-shot 5-shot ● Few-shot CIFAR-100 (FC100) dataset method backbone 1-shot 5-shot ★ Code is available at: https://github.com/y2l/ meta-transfer-learning -tensorflow

Meta-Transfer Learning for Few-Shot Learningyaliu/files/meta-transfer... · 2020-07-14 · Meta-Transfer Learning for Few-Shot Learning Qianru Sun1,3* Yaoyao Liu1,2* Tat-Seng Chua1

Download PDF Report

Upload
others
View
3
Download
0

Embed Size (px)

Citation preview

Page 1: Meta-Transfer Learning for Few-Shot Learningyaliu/files/meta-transfer... · 2020-07-14 · Meta-Transfer Learning for Few-Shot Learning Qianru Sun1,3* Yaoyao Liu1,2* Tat-Seng Chua1

Meta-Transfer Learning for Few-Shot LearningQianru Sun1,3* Yaoyao Liu1,2* Tat-Seng Chua1 Bernt Schiele3

1National University of Singapore, 2Tianjin University, 3Max Planck Institute for Informatics

Motivation & Contributions Meta-Transfer Learning Hard Task Meta-Batch● Few-shot learning is challenging

due to the lack of training data.● Re-thinking promising methods:○ deep neural networks (DNN)○ transfer learning (pre-train, fine-tune)○ meta-learning (meta gradient descent)

● Problem: “catastrophic forgetting”

● Our solution: Scaling & Shifting (SS)

● Problem: slow meta-training convergence

● Our solution: hard negative task sampling

Detail

Top performance is achieved!

Faster convergence is achieved!

Detail

● miniImageNet dataset

Few-shot CIFAR-100 (FC100) dataset

method backbone 1-shot 5-shot

● Few-shot CIFAR-100 (FC100) datasetmethod backbone 1-shot 5-shot

★ Code is available at: https://github.com/y2l/meta-transfer-learning-tensorflow

Learn to Cross-lingual Transfer with Meta Graph Learning ...Meta learning, a.k.a learning to learn, (Finn et al.,2017;Snell et al.,2017), addresses the few-shot problems, by extracting

Learn to Cross-lingual Transfer with Meta Graph Learning ...Meta learning, a.k.a learning to learn, (Finn et al.,2017;Snell et al.,2017), addresses the few-shot problems, by extracting

Documents

I Z -SHOT VOICE STYLE TRANSFER VIA DISENTANGLED

I Z -SHOT VOICE STYLE TRANSFER VIA DISENTANGLED

Documents

Few-Shot Learning With Embedded Class Models …openaccess.thecvf.com/content_ICCV_2019/papers/...Few-Shot Learning with Embedded Class Models and Shot-Free Meta Training Avinash Ravichandran

Few-Shot Learning With Embedded Class Models …openaccess.thecvf.com/content_ICCV_2019/papers/...Few-Shot Learning with Embedded Class Models and Shot-Free Meta Training Avinash Ravichandran

Documents

UNDER REVIEW 1 Meta-Transfer Learning through Hard Tasks · novel approach called meta-transfer learning (MTL) which learns to transfer the weights of a deep NN for few-shot learning

UNDER REVIEW 1 Meta-Transfer Learning through Hard Tasks · novel approach called meta-transfer learning (MTL) which learns to transfer the weights of a deep NN for few-shot learning

Documents

One-shot Transfer Learning for Population Mapping

One-shot Transfer Learning for Population Mapping

Documents

One-Shot Visual Imitation Learning via Meta-Learningproceedings.mlr.press/v78/finn17a/finn17a.pdf · Keywords: imitation learning, deep learning, one-shot learning, meta-learning

One-Shot Visual Imitation Learning via Meta-Learningproceedings.mlr.press/v78/finn17a/finn17a.pdf · Keywords: imitation learning, deep learning, one-shot learning, meta-learning

Documents

Meta-Learning: from Few-Shot Learning to Rapid

Meta-Learning: from Few-Shot Learning to Rapid

Documents

Meta Learning - cs.ubc.calsigal/532S_2018W2/6a.pdf · Meta Learning Shared Hierarchies (Frans et. al, 2017) One-shot visual imitation learning via meta-learning (Finn et. al, 2017)

Meta Learning - cs.ubc.calsigal/532S_2018W2/6a.pdf · Meta Learning Shared Hierarchies (Frans et. al, 2017) One-shot visual imitation learning via meta-learning (Finn et. al, 2017)

Documents

Valuation 11: Benefit Transfer and Meta-Analysis Why benefit transfer? Different types of benefit transfer –Value transfer –Function transfer Validity

Valuation 11: Benefit Transfer and Meta-Analysis Why benefit transfer? Different types of benefit transfer –Value transfer –Function transfer Validity

Documents

Transfer Learning using Meta-learning · MINFaculty DepartmentofInformatics Transfer Learning using Meta-learning NileshVijayrania UniversityofHamburg FacultyofMathematics,InformaticsandNaturalSciences

Transfer Learning using Meta-learning · MINFaculty DepartmentofInformatics Transfer Learning using Meta-learning NileshVijayrania UniversityofHamburg FacultyofMathematics,InformaticsandNaturalSciences

Documents

Unraveling Meta-Learning: Understanding FeatureRepresentations … · 2020. 7. 2. · Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks Micah Goldblum1

Unraveling Meta-Learning: Understanding FeatureRepresentations … · 2020. 7. 2. · Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Tasks Micah Goldblum1

Documents

Knowledge Distillation for Model-Agnostic Meta-Learningecai2020.eu/papers/1264_paper.pdf · 3.1 Model-Agnostic Meta-learning (MAML) MAML is a meta-learning framework for one-shot

Knowledge Distillation for Model-Agnostic Meta-Learningecai2020.eu/papers/1264_paper.pdf · 3.1 Model-Agnostic Meta-learning (MAML) MAML is a meta-learning framework for one-shot

Documents

Weak-shot Fine-grained Classiﬁcation via Similarity Transfer

Weak-shot Fine-grained Classiﬁcation via Similarity Transfer

Documents

Few-Shot One-Class Classiﬁcation via Meta-Learning

Few-Shot One-Class Classiﬁcation via Meta-Learning

Documents

OOD-MAML: Meta-Learning for Few-Shot Out-of-Distribution … · In meta-testing phase, given K-shot N-way samples, we construct Nsub-tasks of K-shot OOD detection with respect to

OOD-MAML: Meta-Learning for Few-Shot Out-of-Distribution … · In meta-testing phase, given K-shot N-way samples, we construct Nsub-tasks of K-shot OOD detection with respect to

Documents

Emergent Complexity and Zero-shot Transfer via

Emergent Complexity and Zero-shot Transfer via

Documents

Meta-Transfer Learning for Zero-Shot Super-Resolution...We found that simply employing transfer learning or ﬁne-tuning from a pre-trained network does not yield plausible results

Meta-Transfer Learning for Zero-Shot Super-Resolution...We found that simply employing transfer learning or ﬁne-tuning from a pre-trained network does not yield plausible results

Documents

Zero-Shot Language Transfer for Cross-Lingual Sentence

Zero-Shot Language Transfer for Cross-Lingual Sentence

Documents

Zero shot learning through cross-modal transfer

Zero shot learning through cross-modal transfer

Presentations & Public Speaking

ONE-SHOT PARAMETRIC AUDIO PRODUCTION STYLE TRANSFER …njb/research/icassp2020_style_transfer.… · ONE-SHOT PARAMETRIC AUDIO PRODUCTION STYLE TRANSFER WITH APPLICATION TO FREQUENCY

ONE-SHOT PARAMETRIC AUDIO PRODUCTION STYLE TRANSFER …njb/research/icassp2020_style_transfer.… · ONE-SHOT PARAMETRIC AUDIO PRODUCTION STYLE TRANSFER WITH APPLICATION TO FREQUENCY

Documents

Policy Transfer Algorithms for Meta Inverse Reinforcement ...Policy Transfer Algorithms for Meta Inverse Reinforcement Learning Benjamin Kha Abstract Inverse reinforcement learning

Policy Transfer Algorithms for Meta Inverse Reinforcement ...Policy Transfer Algorithms for Meta Inverse Reinforcement Learning Benjamin Kha Abstract Inverse reinforcement learning

Documents

Zero-shot Knowledge Transfer via Adversarial Belief Matching · Zero-shot Knowledge Transfer via Adversarial Belief Matching Paul Micaelli University of Edinburgh {paul.micaelli}@ed.ac.uk

Zero-shot Knowledge Transfer via Adversarial Belief Matching · Zero-shot Knowledge Transfer via Adversarial Belief Matching Paul Micaelli University of Edinburgh {paul.micaelli}@ed.ac.uk

Documents

Few-Shot Human Motion Prediction via Meta-learning · Few-Shot Human Motion Prediction via Meta-learning Liang-Yan Gui(B), Yu-Xiong Wang, Deva Ramanan, and Jos´eM.F.Moura Carnegie

Few-Shot Human Motion Prediction via Meta-learning · Few-Shot Human Motion Prediction via Meta-learning Liang-Yan Gui(B), Yu-Xiong Wang, Deva Ramanan, and Jos´eM.F.Moura Carnegie

Documents

Meta-Transfer Learning for Few-Shot Learningopenaccess.thecvf.com/content_CVPR_2019/papers/Sun_Meta...Meta-Transfer Learning for Few-Shot Learning Qianru Sun1,3∗ Yaoyao Liu2∗ Tat-Seng

Meta-Transfer Learning for Few-Shot Learningopenaccess.thecvf.com/content_CVPR_2019/papers/Sun_Meta...Meta-Transfer Learning for Few-Shot Learning Qianru Sun1,3∗ Yaoyao Liu2∗ Tat-Seng

Documents

Multi-source Meta Transfer for Low Resource Multiple ... · On the top of it, the meta transfer learning (MTL) is integrated into ... (Seo et al.,2012). Recently, meta learning gains

Multi-source Meta Transfer for Low Resource Multiple ... · On the top of it, the meta transfer learning (MTL) is integrated into ... (Seo et al.,2012). Recently, meta learning gains

Documents

REVIEW DRAFT Using Meta Analysis for Benefits Transfer: Theory and Practiceexcen.gsu.edu/workingpapers/GSU_EXCEN_WP_2006-12.pdf · Using Meta Analysis for Benefits Transfer: Theory

REVIEW DRAFT Using Meta Analysis for Benefits Transfer: Theory and Practiceexcen.gsu.edu/workingpapers/GSU_EXCEN_WP_2006-12.pdf · Using Meta Analysis for Benefits Transfer: Theory

Documents

Meta-DermDiagnosis: Few-Shot Skin Disease Identification using Meta-Learning · 2020. 12. 18. · Vuorio et al’s work ‘Multimodal model-agnostic meta-learning via task-aware modulation.’

Meta-DermDiagnosis: Few-Shot Skin Disease Identification using Meta-Learning · 2020. 12. 18. · Vuorio et al’s work ‘Multimodal model-agnostic meta-learning via task-aware modulation.’

Documents

Meta-DermDiagnosis: Few-Shot Skin Disease Identification ...openaccess.thecvf.com/content_CVPRW_2020/papers/w42/...Meta-DermDiagnosis: Few-Shot Skin Disease Identiﬁcation using Meta-Learning

Meta-DermDiagnosis: Few-Shot Skin Disease Identification ...openaccess.thecvf.com/content_CVPRW_2020/papers/w42/...Meta-DermDiagnosis: Few-Shot Skin Disease Identiﬁcation using Meta-Learning

Documents

Meta-transfer Learning for Few-shot Learningyaliu/files/meta... · Datasets miniImageNet - Reorganized from ImageNet - Vinyals et al.[1] first devised the dataset, and it is widely

Meta-transfer Learning for Few-shot Learningyaliu/files/meta... · Datasets miniImageNet - Reorganized from ImageNet - Vinyals et al.[1] first devised the dataset, and it is widely

Documents

Neural Style Transfer via Meta Networks - CVF Open Accessopenaccess.thecvf.com/content_cvpr_2018/papers/... · Neural Style Transfer via Meta Networks ... near-optimal solution image

Neural Style Transfer via Meta Networks - CVF Open Accessopenaccess.thecvf.com/content_cvpr_2018/papers/... · Neural Style Transfer via Meta Networks ... near-optimal solution image

Documents