On First-Order Meta-Learning Algorithms

Abstract

This paper considers meta-learning problems, where there is a distribution of tasks, and we would like to obtain an agent that performs well (i.e., learns quickly) when presented with a previously unseen task sampled from this distribution. We analyze a family of algorithms for learning a parameter initialization that can be fine-tuned quickly on a new task, using only first-order derivatives for the meta-learning updates. This family includes and generalizes first-order MAML, an approximation to MAML obtained by ignoring second-order derivatives. It also includes Reptile, a new algorithm that we introduce here, which works by repeatedly sampling a task, training on it, and moving the initialization towards the trained weights on that task. We expand on the results from Finn et al. showing that first-order meta-learning algorithms perform well on some well-established benchmarks for few-shot classification, and we provide theoretical analysis aimed at understanding why these algorithms work.

References in corpus (2)

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm

Cited by in corpus (268)

• Fast Adaptive Task Offloading in Edge Computing based on Meta Reinforcement Learning
• Improving Federated Learning Personalization via Model Agnostic Meta Learning
• Personalized Federated Learning: A Meta-Learning Approach
• Adaptive Personalized Federated Learning
• FedPAQ: A Communication-Efficient Federated Learning Method with Periodic Averaging and Quantization
• Frustratingly Simple Few-Shot Object Detection
• Massively Multilingual Neural Machine Translation in the Wild: Findings and Challenges
• LEAF: A Benchmark for Federated Settings
• Federated Meta-Learning with Fast Convergence and Efficient Communication
• Personalized Federated Learning with Moreau Envelopes
• A Survey of Machine Learning for Computer Architecture and Systems
• A Field Guide to Federated Optimization
• Learning and Evaluating General Linguistic Intelligence
• A Two-Stage Approach to Few-Shot Learning for Image Recognition
• Meta-learning PINN loss functions
• Learning to Demodulate from Few Pilots via Offline and Online Meta-Learning
• Heterogeneous Domain Generalization via Domain Mixup
• A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms
• Finding Task-Relevant Features for Few-Shot Learning by Category Traversal
• How to train your MAML
• Empirical Bayes Transductive Meta-Learning with Synthetic Gradients
• A novel meta-learning initialization method for physics-informed neural networks
• Few-Shot Unsupervised Image-to-Image Translation
• Neuron Linear Transformation: Modeling the Domain Shift for Crowd Counting
• Subgraph Federated Learning with Missing Neighbor Generation
• From Learning to Meta-Learning: Reduced Training Overhead and Complexity for Communication Systems
• Meta-Learning with Warped Gradient Descent
• Meta-learning for Few-shot Natural Language Processing: A Survey
• Discovering Reinforcement Learning Algorithms
• BOIL: Towards Representation Change for Few-shot Learning
• A Game Theoretic Framework for Model Based Reinforcement Learning
• Normalization Techniques in Training DNNs: Methodology, Analysis and Application
• Solving Stochastic Compositional Optimization is Nearly as Easy as Solving Stochastic Optimization
• End-to-End Fast Training of Communication Links Without a Channel Model via Online Meta-Learning
• Transductive Information Maximization For Few-Shot Learning
• Weight Poisoning Attacks on Pre-trained Models
• Query-efficient Meta Attack to Deep Neural Networks
• Optimizing Millions of Hyperparameters by Implicit Differentiation
• Meta-Gradient Reinforcement Learning with an Objective Discovered Online
• La-MAML: Look-ahead Meta Learning for Continual Learning
• Negative Margin Matters: Understanding Margin in Few-shot Classification
• Learning to Extrapolate Knowledge: Transductive Few-shot Out-of-Graph Link Prediction
• Theoretical Convergence of Multi-Step Model-Agnostic Meta-Learning
• TaskNorm: Rethinking Batch Normalization for Meta-Learning
• Modular meta-learning
• Dif-MAML: Decentralized Multi-Agent Meta-Learning
• Efficiently Identifying Task Groupings for Multi-Task Learning
• Efficient Continual Learning with Modular Networks and Task-Driven Priors
• Data Augmentation for Meta-Learning
• Rethinking "Batch" in BatchNorm
• Cross-Domain Few-Shot Learning by Representation Fusion
• Federated Reconstruction: Partially Local Federated Learning
• DPGN: Distribution Propagation Graph Network for Few-shot Learning
• Robust MAML: Prioritization task buffer with adaptive learning process for model-agnostic meta-learning
• MetaPix: Few-Shot Video Retargeting
• Differentially Private Meta-Learning
• Meta-Learning for Few-Shot Land Cover Classification
• Towards Practical Multi-Object Manipulation using Relational Reinforcement Learning
• On the Outsized Importance of Learning Rates in Local Update Methods
• Accelerating Quadratic Optimization with Reinforcement Learning
• Backdoor Attacks on Federated Meta-Learning
• Cross-Domain Few-Shot Learning with Meta Fine-Tuning
• RelationNet2: Deep Comparison Columns for Few-Shot Learning
• On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-Learning
• DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection
• Few-Shot Learning on Graphs via Super-Classes based on Graph Spectral Measures
• Flexible Dataset Distillation: Learn Labels Instead of Images
• A Broader Study of Cross-Domain Few-Shot Learning
• When MAML Can Adapt Fast and How to Assist When It Cannot
• Learning Dynamic Alignment via Meta-filter for Few-shot Learning
• How to Train Your MAML to Excel in Few-Shot Classification
• Prototype Rectification for Few-Shot Learning
• Learned Initializations for Optimizing Coordinate-Based Neural Representations
• Meta-Learning with Sparse Experience Replay for Lifelong Language Learning
• Few-shot Image Generation via Cross-domain Correspondence
• MT3: Meta Test-Time Training for Self-Supervised Test-Time Adaption
• Tasks, stability, architecture, and compute: Training more effective learned optimizers, and using them to train themselves
• Gradient Matching for Domain Generalization
• A Comparative Review of Recent Few-Shot Object Detection Algorithms
• Dynamic backdoor attacks against federated learning
• Learning Adaptive Classifiers Synthesis for Generalized Few-Shot Learning
• Low-Resource Task-Oriented Semantic Parsing via Intrinsic Modeling
• Sequential Scenario-Specific Meta Learner for Online Recommendation
• Deep Transfer Learning Based Downlink Channel Prediction for FDD Massive MIMO Systems
• MetaCI: Meta-Learning for Causal Inference in a Heterogeneous Population
• What Do We Mean by Generalization in Federated Learning?
• Learning to Learn with Variational Information Bottleneck for Domain Generalization
• Learning to learn via Self-Critique
• How Important is the Train-Validation Split in Meta-Learning?
• Generalizable Person Re-identification with Relevance-aware Mixture of Experts
• Interference and Generalization in Temporal Difference Learning
• Meta-Learning Acquisition Functions for Transfer Learning in Bayesian Optimization
• Weighted Meta-Learning
• Modeling and Optimization Trade-off in Meta-learning
• Federated Mixture of Experts
• Few-Shot Viewpoint Estimation
• MetaFun: Meta-Learning with Iterative Functional Updates
• Instance Credibility Inference for Few-Shot Learning
• Dual-Awareness Attention for Few-Shot Object Detection
• Memory-Based Optimization Methods for Model-Agnostic Meta-Learning and Personalized Federated Learning
• Multi-Objective Meta Learning
• Contrastive Prototype Learning with Augmented Embeddings for Few-Shot Learning
• Meta Reasoning over Knowledge Graphs
• Learning Fast Sample Re-weighting Without Reward Data
• Modular Meta-Learning with Shrinkage
• Meta-learnt priors slow down catastrophic forgetting in neural networks
• Self-Supervised Fast Adaptation for Denoising via Meta-Learning
• Convergence and Accuracy Trade-Offs in Federated Learning and Meta-Learning
• Revisiting Few-Shot Learning for Facial Expression Recognition
• Prototype-based Incremental Few-Shot Semantic Segmentation
• Structured Prediction for Conditional Meta-Learning
• Measuring and Harnessing Transference in Multi-Task Learning
• Rapid Neural Architecture Search by Learning to Generate Graphs from Datasets
• Adaptive-Step Graph Meta-Learner for Few-Shot Graph Classification
• Finding online neural update rules by learning to remember
• Online Meta-Learning for Multi-Source and Semi-Supervised Domain Adaptation
• Learning Meta Face Recognition in Unseen Domains
• Improved End-to-End Dysarthric Speech Recognition via Meta-learning Based Model Re-initialization
• Attentive Graph Neural Networks for Few-Shot Learning
• Unsupervised Reinforcement Learning of Transferable Meta-Skills for Embodied Navigation
• Few-Shot Anomaly Detection for Polyp Frames from Colonoscopy
• Multiple Meta-model Quantifying for Medical Visual Question Answering
• Low-Resource Domain Adaptation for Compositional Task-Oriented Semantic Parsing
• Information Theoretic Meta Learning with Gaussian Processes
• Meta Learning Black-Box Population-Based Optimizers
• Regularized Fine-grained Meta Face Anti-spoofing
• Continual Learning via Online Leverage Score Sampling
• How Fine-Tuning Allows for Effective Meta-Learning
• $k$Folden: $k$-Fold Ensemble for Out-Of-Distribution Detection
• Adaptive Task Sampling for Meta-Learning
• EvoGrad: Efficient Gradient-Based Meta-Learning and Hyperparameter Optimization
• MxML: Mixture of Meta-Learners for Few-Shot Classification
• Task-Adaptive Neural Network Search with Meta-Contrastive Learning
• Generalization of Model-Agnostic Meta-Learning Algorithms: Recurring and Unseen Tasks
• Sign-MAML: Efficient Model-Agnostic Meta-Learning by SignSGD
• Continual Learning Using World Models for Pseudo-Rehearsal
• A Channel Coding Benchmark for Meta-Learning
• Meta-Learning with Neural Tangent Kernels
• Generating Personalized Dialogue via Multi-Task Meta-Learning
• A User's Guide to Calibrating Robotics Simulators
• Meta Dropout: Learning to Perturb Features for Generalization
• SB-MTL: Score-based Meta Transfer-Learning for Cross-Domain Few-Shot Learning
• DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta
• AdarGCN: Adaptive Aggregation GCN for Few-Shot Learning
• Dynamic population-based meta-learning for multi-agent communication with natural language
• Learning Prototype-oriented Set Representations for Meta-Learning
• MatchingGAN: Matching-based Few-shot Image Generation
• Learning to Recommend via Meta Parameter Partition
• Fast Training of Neural Lumigraph Representations using Meta Learning
• Meta-Learned Invariant Risk Minimization
• What Can Knowledge Bring to Machine Learning? -- A Survey of Low-shot Learning for Structured Data
• Bilevel Optimization for Machine Learning: Algorithm Design and Convergence Analysis
• MetaHTR: Towards Writer-Adaptive Handwritten Text Recognition
• Meta Adversarial Perturbations
• Investigating Meta-Learning Algorithms for Low-Resource Natural Language Understanding Tasks
• Online Hyperparameter Meta-Learning with Hypergradient Distillation
• Real-Time Visual Object Tracking via Few-Shot Learning
• Gradient Inversion with Generative Image Prior
• Optimizing Reusable Knowledge for Continual Learning via Metalearning
• A contrastive rule for meta-learning
• Pandemic Drugs at Pandemic Speed: Infrastructure for Accelerating COVID-19 Drug Discovery with Hybrid Machine Learning- and Physics-based Simulations on High Performance Computers
• Meta-Learned Attribute Self-Gating for Continual Generalized Zero-Shot Learning
• Meta Reinforcement Learning with Autonomous Inference of Subtask Dependencies
• Global Convergence and Generalization Bound of Gradient-Based Meta-Learning with Deep Neural Nets
• NOVAS: Non-convex Optimization via Adaptive Stochastic Search for End-to-End Learning and Control
• How Does the Task Landscape Affect MAML Performance?
• The Advantage of Conditional Meta-Learning for Biased Regularization and Fine-Tuning
• Unsupervised Meta-Learning through Latent-Space Interpolation in Generative Models
• Meta-Learning for Domain Generalization in Semantic Parsing
• Meta-Curriculum Learning for Domain Adaptation in Neural Machine Translation
• Bottom-Up Meta-Policy Search
• Bilevel Continual Learning
• Multi-step Estimation for Gradient-based Meta-learning
• A Novel Privacy-Preserved Recommender System Framework based on Federated Learning
• Testing the Genomic Bottleneck Hypothesis in Hebbian Meta-Learning
• Learn Faster and Forget Slower via Fast and Stable Task Adaptation
• Meta-Learning with Network Pruning
• Model-Agnostic Meta-Learning using Runge-Kutta Methods
• Safe Exploration by Solving Early Terminated MDP
• Off-Policy Meta-Reinforcement Learning Based on Feature Embedding Spaces
• CrossFit: A Few-shot Learning Challenge for Cross-task Generalization in NLP
• Self-Supervised Video Representation Learning with Meta-Contrastive Network
• Model-Agnostic Meta-Learning for EEG Motor Imagery Decoding in Brain-Computer-Interfacing
• Sequential Learning for Domain Generalization
• AdaCrowd: Unlabeled Scene Adaptation for Crowd Counting
• Collision Avoidance Robotics Via Meta-Learning (CARML)
• An Optimization-Based Meta-Learning Model for MRI Reconstruction with Diverse Dataset
• Federated Reconnaissance: Efficient, Distributed, Class-Incremental Learning
• MetaInv-Net: Meta Inversion Network for Sparse View CT Image Reconstruction
• Flying Through a Narrow Gap Using End-to-end Deep Reinforcement Learning Augmented with Curriculum Learning and Sim2Real
• Model-Agnostic Learning to Meta-Learn
• Uncertainty-Aware Few-Shot Image Classification
• Inductive Unsupervised Domain Adaptation for Few-Shot Classification via Clustering
• Modular Adaptation for Cross-Domain Few-Shot Learning
• Meta-Adversarial Inverse Reinforcement Learning for Decision-making Tasks
• Identifying Physical Law of Hamiltonian Systems via Meta-Learning
• Meta-Reinforcement Learning for Adaptive Motor Control in Changing Robot Dynamics and Environments
• System Identification via Meta-Learning in Linear Time-Varying Environments
• Liquid Warping GAN with Attention: A Unified Framework for Human Image Synthesis
• Meta Variational Monte Carlo
• GCCN: Global Context Convolutional Network
• Inconsistent Few-Shot Relation Classification via Cross-Attentional Prototype Networks with Contrastive Learning
• Meta Gradient Adversarial Attack
• Personalizing Pre-trained Models
• X-METRA-ADA: Cross-lingual Meta-Transfer Learning Adaptation to Natural Language Understanding and Question Answering
• Adaptive Knowledge-Enhanced Bayesian Meta-Learning for Few-shot Event Detection
• An Expectation-Maximization Perspective on Federated Learning
• FLIX: A Simple and Communication-Efficient Alternative to Local Methods in Federated Learning
• Sequential Reptile: Inter-Task Gradient Alignment for Multilingual Learning
• Task Affinity with Maximum Bipartite Matching in Few-Shot Learning
• Should We Be Pre-training? An Argument for End-task Aware Training as an Alternative
• Learning Meta Representations for Agents in Multi-Agent Reinforcement Learning
• Memory Efficient Meta-Learning with Large Images
• Large-Scale Meta-Learning with Continual Trajectory Shifting
• Extended Few-Shot Learning: Exploiting Existing Resources for Novel Tasks
• Meta-Learning for One-Class Classification with Few Examples using Order-Equivariant Network
• Fuzzy Simplicial Networks: A Topology-Inspired Model to Improve Task Generalization in Few-shot Learning
• On Hyper-parameter Tuning for Stochastic Optimization Algorithms
• No-regret Non-convex Online Meta-Learning
• Meta Label Correction for Noisy Label Learning
• Domain-Agnostic Few-Shot Classification by Learning Disparate Modulators
• MAME : Model-Agnostic Meta-Exploration
• Knowledge Transferring via Model Aggregation for Online Social Care
• MetaStyle: Three-Way Trade-Off Among Speed, Flexibility, and Quality in Neural Style Transfer
• Cooperative Bi-path Metric for Few-shot Learning
• Is the Meta-Learning Idea Able to Improve the Generalization of Deep Neural Networks on the Standard Supervised Learning?
• Learning to Learn to Compress
• Augmented Bi-path Network for Few-shot Learning
• One-Shot Learning for Language Modelling
• Characterizing Policy Divergence for Personalized Meta-Reinforcement Learning
• L2E: Learning to Exploit Your Opponent
• MetaMix: Improved Meta-Learning with Interpolation-based Consistency Regularization
• Generalization Bounds for Meta-Learning via PAC-Bayes and Uniform Stability
• Masked Training of Neural Networks with Partial Gradients
• Task Attended Meta-Learning for Few-Shot Learning
• A Few-Shot Sequential Approach for Object Counting
• Channel Relationship Prediction with Forget-Update Module for Few-shot Classification
• Uniform Sampling over Episode Difficulty
• Bootstrapped Meta-Learning
• Meta-Learning with Adjoint Methods
• Meta-Forecasting by combining Global Deep Representations with Local Adaptation
• MetalGAN: a Cluster-based Adaptive Training for Few-Shot Adversarial Colorization
• MDFM: Multi-Decision Fusing Model for Few-Shot Learning
• CLTA: Contents and Length-based Temporal Attention for Few-shot Action Recognition
• Bilevel Online Deep Learning in Non-stationary Environment
• Training few-shot classification via the perspective of minibatch and pretraining
• Meta-Learning with Variational Semantic Memory for Word Sense Disambiguation
• Multi-Pair Text Style Transfer on Unbalanced Data
• Meta Self-Learning for Multi-Source Domain Adaptation: A Benchmark
• Eden: A Unified Environment Framework for Booming Reinforcement Learning Algorithms
• MapRE: An Effective Semantic Mapping Approach for Low-resource Relation Extraction
• Learning by Examples Based on Multi-level Optimization
• Wav-BERT: Cooperative Acoustic and Linguistic Representation Learning for Low-Resource Speech Recognition
• On sensitivity of meta-learning to support data
• Accelerating Gradient-based Meta Learner
• Multilayer Lookahead: a Nested Version of Lookahead
• BiOpt: Bi-Level Optimization for Few-Shot Segmentation
• Performance-Weighed Policy Sampling for Meta-Reinforcement Learning
• Meta Learning as Bayes Risk Minimization
• Meta-Regularization by Enforcing Mutual-Exclusiveness
• Edge-Labeling based Directed Gated Graph Network for Few-shot Learning
• Few-shot Learning with LSSVM Base Learner and Transductive Modules
• On Data Efficiency of Meta-learning
• Proxy Network for Few Shot Learning
• Class Interference Regularization
• Covariate Distribution Aware Meta-learning
• Tackling Long-Tailed Relations and Uncommon Entities in Knowledge Graph Completion
• Submodular Meta-Learning