Fast Context Adaptation via Meta-Learning

Abstract

We propose CAVIA for meta-learning, a simple extension to MAML that is less prone to meta-overfitting, easier to parallelise, and more interpretable. CAVIA partitions the model parameters into two parts: context parameters that serve as additional input to the model and are adapted on individual tasks, and shared parameters that are meta-trained and shared across tasks. At test time, only the context parameters are updated, leading to a low-dimensional task representation. We show empirically that CAVIA outperforms MAML for regression, classification, and reinforcement learning. Our experiments also highlight weaknesses in current benchmarks, in that the amount of adaptation needed in some cases is small.

Published at the International Conference on Machine Learning (ICML) 2019

Cited by in corpus (77)

• Personalized Federated Learning: A Meta-Learning Approach
• Learning to Demodulate from Few Pilots via Offline and Online Meta-Learning
• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• Meta-Learning without Memorization
• Meta-Learning with Warped Gradient Descent
• VariBAD: A Very Good Method for Bayes-Adaptive Deep RL via Meta-Learning
• BOIL: Towards Representation Change for Few-shot Learning
• Multi-level Second-order Few-shot Learning
• End-to-End Fast Training of Communication Links Without a Channel Model via Online Meta-Learning
• Rectifying the Shortcut Learning of Background for Few-Shot Learning
• Self-supervised Auxiliary Learning with Meta-paths for Heterogeneous Graphs
• Theoretical Convergence of Multi-Step Model-Agnostic Meta-Learning
• Meta-Learned Confidence for Few-shot Learning
• Torchmeta: A Meta-Learning library for PyTorch
• Meta-Consolidation for Continual Learning
• Deep Reinforcement Learning amidst Lifelong Non-Stationarity
• Communication-Efficient Robust Federated Learning with Noisy Labels
• Single Episode Policy Transfer in Reinforcement Learning
• Learning Compositional Neural Programs with Recursive Tree Search and Planning
• RelationNet2: Deep Comparison Columns for Few-Shot Learning
• Bayesian Active Meta-Learning for Few Pilot Demodulation and Equalization
• Learning to Learn Variational Semantic Memory
• Few-shot Action Recognition with Permutation-invariant Attention
• Inverse Rational Control with Partially Observable Continuous Nonlinear Dynamics
• Operator Learning with Neural Fields: Tackling PDEs on General Geometries
• Multi-task learning for virtual flow metering
• ReFine: Re-randomization before Fine-tuning for Cross-domain Few-shot Learning
• Dynamic backdoor attacks against federated learning
• Improving Generalization in Meta-learning via Task Augmentation
• Meta-Learning with Fewer Tasks through Task Interpolation
• Offline Meta-Reinforcement Learning with Advantage Weighting
• Context Meta-Reinforcement Learning via Neuromodulation
• Multi-task Batch Reinforcement Learning with Metric Learning
• Learning a Universal Template for Few-shot Dataset Generalization
• Evolving Inborn Knowledge For Fast Adaptation in Dynamic POMDP Problems
• Modular Meta-Learning with Shrinkage
• Meta Feature Modulator for Long-tailed Recognition
• Continuous Coordination As a Realistic Scenario for Lifelong Learning
• The Differentiable Cross-Entropy Method
• AdaRL: What, Where, and How to Adapt in Transfer Reinforcement Learning
• Adaptive-Step Graph Meta-Learner for Few-Shot Graph Classification
• A Survey of Exploration Methods in Reinforcement Learning
• Few-shot Sequence Learning with Transformers
• Multitask Learning with Single Gradient Step Update for Task Balancing
• Self-supervised Auxiliary Learning for Graph Neural Networks via Meta-Learning
• Meta-Learned Invariant Risk Minimization
• Sign-MAML: Efficient Model-Agnostic Meta-Learning by SignSGD
• Few-Shot Unsupervised Continual Learning through Meta-Examples
• Meta Dropout: Learning to Perturb Features for Generalization
• Improving Generalization in Meta-RL with Imaginary Tasks from Latent Dynamics Mixture
• Learning to See Through Obstructions with Layered Decomposition
• Reviewing continual learning from the perspective of human-level intelligence
• Meta Adversarial Perturbations
• Block Contextual MDPs for Continual Learning
• A contrastive rule for meta-learning
• HetMAML: Task-Heterogeneous Model-Agnostic Meta-Learning for Few-Shot Learning Across Modalities
• Meta-learning for mixed linear regression
• Time Series Continuous Modeling for Imputation and Forecasting with Implicit Neural Representations
• Meta-Learning with Network Pruning
• Reinforced Few-Shot Acquisition Function Learning for Bayesian Optimization
• Learning to Customize Model Structures for Few-shot Dialogue Generation Tasks
• VIABLE: Fast Adaptation via Backpropagating Learned Loss
• Memory Efficient Meta-Learning with Large Images
• Learning to Rectify for Robust Learning with Noisy Labels
• Local Nonparametric Meta-Learning
• Model-based Meta Reinforcement Learning using Graph Structured Surrogate Models
• MetricOpt: Learning to Optimize Black-Box Evaluation Metrics
• Towards Understanding Residual and Dilated Dense Neural Networks via Convolutional Sparse Coding
• Learning to Learn to Compress
• Bootstrapped Meta-Learning
• Learning to Transfer: A Foliated Theory
• Domain Conditional Predictors for Domain Adaptation
• Adaptation-Agnostic Meta-Training
• Meta-Learning with Adjoint Methods
• On Data Efficiency of Meta-learning
• CLTA: Contents and Length-based Temporal Attention for Few-shot Action Recognition
• On the Practical Consistency of Meta-Reinforcement Learning Algorithms