Model-Agnostic Meta-Learning

Introduced by Finn et al. in Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks

MAML, or Model-Agnostic Meta-Learning, is a model and task-agnostic algorithm for meta-learning that trains a model’s parameters such that a small number of gradient updates will lead to fast learning on a new task.

Consider a model represented by a parametrized function $f_{\theta}$ with parameters $\theta$. When adapting to a new task $\mathcal{T}_{i}$, the model’s parameters $\theta$ become $\theta'_{i}$. With MAML, the updated parameter vector $\theta'_{i}$ is computed using one or more gradient descent updates on task $\mathcal{T}_{i}$. For example, when using one gradient update,

$$ \theta'_{i} = \theta - \alpha\nabla_{\theta}\mathcal{L}_{\mathcal{T}_{i}}\left(f_{\theta}\right) $$

The step size $\alpha$ may be fixed as a hyperparameter or metalearned. The model parameters are trained by optimizing for the performance of $f_{\theta'_{i}}$ with respect to $\theta$ across tasks sampled from $p\left(\mathcal{T}_{i}\right)$. More concretely the meta-objective is as follows:

$$ \min_{\theta} \sum_{\mathcal{T}_{i} \sim p\left(\mathcal{T}\right)} \mathcal{L}_{\mathcal{T_{i}}}\left(f_{\theta'_{i}}\right) = \sum_{\mathcal{T}_{i} \sim p\left(\mathcal{T}\right)} \mathcal{L}_{\mathcal{T_{i}}}\left(f_{\theta - \alpha\nabla_{\theta}\mathcal{L}_{\mathcal{T}_{i}}\left(f_{\theta}\right)}\right) $$

Note that the meta-optimization is performed over the model parameters $\theta$, whereas the objective is computed using the updated model parameters $\theta'$. In effect MAML aims to optimize the model parameters such that one or a small number of gradient steps on a new task will produce maximally effective behavior on that task. The meta-optimization across tasks is performed via stochastic gradient descent (SGD), such that the model parameters $\theta$ are updated as follows:

$$ \theta \leftarrow \theta - \beta\nabla_{\theta} \sum_{\mathcal{T}_{i} \sim p\left(\mathcal{T}\right)} \mathcal{L}_{\mathcal{T_{i}}}\left(f_{\theta'_{i}}\right)$$

where $\beta$ is the meta step size.

Source: Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks

Latest Papers

PAPER DATE
La-MAML: Look-ahead Meta Learning for Continual Learning
Gunshi GuptaKarmesh YadavLiam Paull
2020-07-27
Few-Shot Bearing Anomaly Detection Based on Model-Agnostic Meta-Learning
Shen ZhangFei YeBingnan WangThomas G. Habetler
2020-07-25
Few-Shot One-Class Classification via Meta-Learning
Ahmed FrikhaDenis KrompaßHans-Georg KöpkenVolker Tresp
2020-07-08
Few-shot Relation Extraction via Bayesian Meta-learning on Relation Graphs
| Meng QuTianyu GaoLouis-Pascal A. C. XhonneuxJian Tang
2020-07-05
On the Global Optimality of Model-Agnostic Meta-Learning
Lingxiao WangQi CaiZhuoran YangZhaoran Wang
2020-06-23
Meta Learning in the Continuous Time Limit
Ruitu XuLin ChenAmin Karbasi
2020-06-19
Convergence of Meta-Learning with Task-Specific Adaptation over Partial Parameters
Kaiyi JiJason D. LeeYingbin LiangH. Vincent Poor
2020-06-16
Learning-to-Learn Personalised Human Activity Recognition Models
Anjana WijekoonNirmalie Wiratunga
2020-06-12
Attentive Feature Reuse for Multi Task Meta learning
Kiran LekkalaLaurent Itti
2020-06-12
BI-MAML: Balanced Incremental Approach for Meta Learning
Yang ZhengJinlin XiangKun SuEli Shlizerman
2020-06-12
Global Convergence of MAML for LQR
Igor MolybogJavad Lavaei
2020-05-31
When does MAML Work the Best? An Empirical Study on Model-Agnostic Meta-Learning in NLP Applications
Zequn LiuRuiyi ZhangYiping SongMing Zhang
2020-05-24
Information-Theoretic Generalization Bounds for Meta-Learning and Applications
Sharu Theresa JoseOsvaldo Simeone
2020-05-09
Hierarchical Attention Network for Action Segmentation
Harshala GammulleSimon DenmanSridha SridharanClinton Fookes
2020-05-07
Bayesian Online Meta-Learning with Laplace Approximation
Pau Ching YapHippolyt RitterDavid Barber
2020-04-30
Learning to Classify Intents and Slot Labels Given a Handful of Examples
Jason KroneYi ZhangMona Diab
2020-04-22
Knowledge-graph based Proactive Dialogue Generation with Improved Meta-Learning
Hongcai XuJunpeng BaoJunqing Wang
2020-04-19
MetaSleepLearner: Fast Adaptation of Bio-signals-Based Sleep Stage Classifier to New Individual Subject Using Meta-Learning
Nannapas BanluesombatkulPichayoot OuppaphanPitshaporn LeelaarpornPayongkit LakhanBusarakum ChaitusaneyNattapong JaimchariyatamEkapol ChuangsuwanichNat DilokthanakulTheerawit Wilaiprasitporn
2020-04-08
Tracking by Instance Detection: A Meta-Learning Approach
Guangting WangChong LuoXiaoyan SunZhiwei XiongWenjun Zeng
2020-04-02
On-the-Fly Adaptation of Source Code Models using Meta-Learning
| Disha ShrivastavaHugo LarochelleDaniel Tarlow
2020-03-26
Weighted Meta-Learning
Diana CaiRishit ShethLester MackeyNicolo Fusi
2020-03-20
Online Fast Adaptation and Knowledge Accumulation: a New Approach to Continual Learning
| Massimo CacciaPau RodriguezOleksiy OstapenkoFabrice NormandinMin LinLucas CacciaIssam LaradjiIrina RishAlexandre LacosteDavid VazquezLaurent Charlin
2020-03-12
Fast Online Adaptation in Robotics through Meta-Learning Embeddings of Simulated Priors
| Rituraj KaushikTimothée AnneJean-Baptiste Mouret
2020-03-10
Curriculum in Gradient-Based Meta-Reinforcement Learning
Bhairav MehtaTristan DeleuSharath Chandra RaparthyChris J. PalLiam Paull
2020-02-19
Theoretical Convergence of Multi-Step Model-Agnostic Meta-Learning
Kaiyi JiJunjie YangYingbin Liang
2020-02-18
Provably Convergent Policy Gradient Methods for Model-Agnostic Meta-Reinforcement Learning
Alireza FallahAryan MokhtariAsuman Ozdaglar
2020-02-12
Task-Robust Model-Agnostic Meta-Learning
Liam CollinsAryan MokhtariSanjay Shakkottai
2020-02-12
Few-Shot One-Class Classification via Meta-Learning
Anonymous
2020-01-01
Role of two learning rates in convergence of model-agnostic meta-learning
Anonymous
2020-01-01
Decoupling Adaptation from Modeling with Meta-Optimizers for Meta Learning
Anonymous
2020-01-01
Efficient Meta Learning via Minibatch Proximal Update
Pan ZhouXiaotong YuanHuan XuShuicheng YanJiashi Feng
2019-12-01
Fair Meta-Learning: Learning How to Learn Fairly
Dylan SlackSorelle FriedlerEmile Givental
2019-11-06
When MAML Can Adapt Fast and How to Assist When It Cannot
Sébastien M. R. ArnoldShariq IqbalFei Sha
2019-10-30
Multimodal Model-Agnostic Meta-Learning via Task-Aware Modulation
| Risto VuorioShao-Hua SunHexiang HuJoseph J. Lim
2019-10-30
HIDRA: Head Initialization across Dynamic targets for Robust Architectures
| Rafael Rego DrumondLukas BrinkmeyerJosif GrabockaLars Schmidt-Thieme
2019-10-28
Model-Agnostic Meta-Learning using Runge-Kutta Methods
Daniel Jiwoong ImYibo JiangNakul Verma
2019-10-16
Improving Federated Learning Personalization via Model Agnostic Meta Learning
Yihan JiangJakub KonečnýKeith RushSreeram Kannan
2019-09-27
MERL: Multi-Head Reinforcement Learning
Yannis Flet-BerliacPhilippe Preux
2019-09-26
ES-MAML: Simple Hessian-Free Meta Learning
Xingyou SongWenbo GaoYuxiang YangKrzysztof ChoromanskiAldo PacchianoYunhao Tang
2019-09-25
Coupled Generative Adversarial Network for Continuous Fine-grained Action Segmentation
Harshala GammulleTharindu FernandoSimon DenmanSridha SridharanClinton Fookes
2019-09-20
Fine-grained Action Segmentation using the Semi-Supervised Action GAN
Harshala GammulleSimon DenmanSridha SridharanClinton Fookes
2019-09-20
Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
Aniruddh RaghuMaithra RaghuSamy BengioOriol Vinyals
2019-09-19
Modular Meta-Learning with Shrinkage
Yutian ChenAbram L. FriesenFeryal BehbahaniArnaud DoucetDavid BuddenMatthew W. HoffmanNando de Freitas
2019-09-12
Meta-Learning with Implicit Gradients
| Aravind RajeswaranChelsea FinnSham KakadeSergey Levine
2019-09-10
On the Convergence Theory of Gradient-Based Model-Agnostic Meta-Learning Algorithms
Alireza FallahAryan MokhtariAsuman Ozdaglar
2019-08-27
Meta Reasoning over Knowledge Graphs
Hong WangWenhan XiongMo YuXiaoxiao GuoShiyu ChangWilliam Yang Wang
2019-08-13
Towards Understanding Generalization in Gradient-Based Meta-Learning
Simon GuiroyVikas VermaChristopher Pal
2019-07-16
Evolvability ES: Scalable and Direct Optimization of Evolvability
Alexander GajewskiJeff CluneKenneth O. StanleyJoel Lehman
2019-07-13
A Model-based Approach for Sample-efficient Multi-task Reinforcement Learning
Nicholas C. LandolfiGarrett ThomasTengyu Ma
2019-07-11
Evolutionary Reinforcement Learning for Sample-Efficient Multiagent Coordination
Shauharda KhadkaSomdeb MajumdarSantiago MiretStephen McAleerKagan Tumer
2019-06-18
Alpha MAML: Adaptive Model-Agnostic Meta-Learning
Harkirat Singh BehlAtılım Güneş BaydinPhilip H. S. Torr
2019-05-17
Follow the Attention: Combining Partial Pose and Object Motion for Fine-Grained Action Detection
Mohammad Mahdi Kazemi MoghaddamEhsan AbbasnejadJaven Shi
2019-05-11
Attentive Task-Agnostic Meta-Learning for Few-Shot Text Classification
Xiang JiangMohammad HavaeiGabriel ChartrandHassan ChouaibThomas VincentAndrew JessonNicolas ChapadosStan Matwin
2019-05-01
NoRML: No-Reward Meta Learning
| Yuxiang YangKen CaluwaertsAtil IscenJie TanChelsea Finn
2019-03-04
Online Meta-Learning
Chelsea FinnAravind RajeswaranSham KakadeSergey Levine
2019-02-22
Learning to Generalize from Sparse and Underspecified Rewards
| Rishabh AgarwalChen LiangDale SchuurmansMohammad Norouzi
2019-02-19
Meta-Curvature
| Eunbyung ParkJunier B. Oliva
2019-02-09
An Investigation of Few-Shot Learning in Spoken Term Classification
Yangbin ChenTom KoLifeng ShangXiao ChenXin JiangQing Li
2018-12-26
Toward Multimodal Model-Agnostic Meta-Learning
Risto VuorioShao-Hua SunHexiang HuJoseph J. Lim
2018-12-18
The effects of negative adaptation in Model-Agnostic Meta-Learning
Tristan DeleuYoshua Bengio
2018-12-05
How to train your MAML
| Antreas AntoniouHarrison EdwardsAmos Storkey
2018-10-22
Gradient Agreement as an Optimization Objective for Meta-Learning
Amir Erfan EshratifarDavid EigenMassoud Pedram
2018-10-18
Fast Context Adaptation via Meta-Learning
| Luisa M ZintgrafKyriacos ShiarlisVitaly KurinKatja HofmannShimon Whiteson
2018-10-08
Meta-Learning by the Baldwin Effect
Chrisantha Thomas FernandoJakub SygnowskiSimon OsinderoJane WangTom SchaulDenis TeplyashinPablo SprechmannAlexander PritzelAndrei A. Rusu
2018-06-06
On the Importance of Attention in Meta-Learning for Few-Shot Text Classification
Xiang JiangMohammad HavaeiGabriel ChartrandHassan ChouaibThomas VincentAndrew JessonNicolas ChapadosStan Matwin
2018-06-03
On First-Order Meta-Learning Algorithms
| Alex NicholJoshua AchiamJohn Schulman
2018-03-08
Deep Meta-Learning: Learning to Learn in the Concept Space
Fengwei ZhouBin WuZhenguo Li
2018-02-10
Recasting Gradient-Based Meta-Learning as Hierarchical Bayes
Erin GrantChelsea FinnSergey LevineTrevor DarrellThomas Griffiths
2018-01-26
Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
| Zhenguo LiFengwei ZhouFei ChenHang Li
2017-07-31
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
| Chelsea FinnPieter AbbeelSergey Levine
2017-03-09

Tasks

TASK PAPERS SHARE
Meta-Learning 62 43.06%
Few-Shot Learning 18 12.50%
Image Classification 9 6.25%
Few-Shot Image Classification 8 5.56%
Text Classification 3 2.08%
Action Segmentation 3 2.08%
Omniglot 3 2.08%
Continual Learning 2 1.39%
Time Series 2 1.39%

Components

COMPONENT TYPE
🤖 No Components Found You can add them if they exist; e.g. Mask R-CNN uses RoIAlign

Categories