Recasting Gradient-Based Meta-Learning as Hierarchical Bayes

Abstract

Meta-learning allows an intelligent agent to leverage prior learning episodes as a basis for quickly improving performance on a novel task. Bayesian hierarchical modeling provides a theoretical framework for formalizing meta-learning as inference for a set of parameters that are shared across tasks. Here, we reformulate the model-agnostic meta-learning algorithm (MAML) of Finn et al. (2017) as a method for probabilistic inference in a hierarchical Bayesian model. In contrast to prior methods for meta-learning via hierarchical Bayes, MAML is naturally applicable to complex function approximators through its use of a scalable gradient descent procedure for posterior inference. Furthermore, the identification of MAML as hierarchical Bayes provides a way to understand the algorithm's operation as a meta-learning procedure, as well as an opportunity to make use of computational strategies for efficient inference. We use this opportunity to propose an improvement to the MAML algorithm that makes use of techniques from approximate inference and curvature estimation.

References in corpus (7)

• Adam: A Method for Stochastic Optimization
• Prototypical Networks for Few-shot Learning
• Theoretical Models of Learning to Learn
• Matching Networks for One Shot Learning
• Towards a Neural Statistician
• Bayesian Multitask Learning with Latent Hierarchies
• Few-Shot Learning Through an Information Retrieval Lens

Cited by in corpus (67)

• Bayesian Model-Agnostic Meta-Learning
• Learning to Demodulate from Few Pilots via Offline and Online Meta-Learning
• One-Shot Imitation from Observing Humans via Domain-Adaptive Meta-Learning
• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• Empirical Bayes Transductive Meta-Learning with Synthetic Gradients
• Meta-Learning without Memorization
• From Learning to Meta-Learning: Reduced Training Overhead and Complexity for Communication Systems
• Improving Generalization in Meta Reinforcement Learning using Learned Objectives
• Deep Learning Theory Review: An Optimal Control and Dynamical Systems Perspective
• Reward Shaping via Meta-Learning
• A Comprehensive Overview and Survey of Recent Advances in Meta-Learning
• Meta-learning of Sequential Strategies
• Prior-Knowledge and Attention-based Meta-Learning for Few-Shot Learning
• TaskNorm: Rethinking Batch Normalization for Meta-Learning
• Meta-Graph: Few Shot Link Prediction via Meta Learning
• Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddings
• Efficiently Identifying Task Groupings for Multi-Task Learning
• Meta-Baseline: Exploring Simple Meta-Learning for Few-Shot Learning
• Defining Benchmarks for Continual Few-Shot Learning
• Learning How to Demodulate from Few Pilots via Meta-Learning
• Domain Adaptive Dialog Generation via Meta Learning
• Meta-Learning Mean Functions for Gaussian Processes
• When MAML Can Adapt Fast and How to Assist When It Cannot
• Toward Multimodal Model-Agnostic Meta-Learning
• Scalable Bayesian Meta-Learning through Generalized Implicit Gradients
• Learning to learn via Self-Critique
• Meta-Learning surrogate models for sequential decision making
• Uncertainty in Multitask Transfer Learning
• Compositional ADAM: An Adaptive Compositional Solver
• Meta-Amortized Variational Inference and Learning
• Stateless Neural Meta-Learning using Second-Order Gradients
• Offline Meta Learning of Exploration
• Bayesian Few-Shot Classification with One-vs-Each Pólya-Gamma Augmented Gaussian Processes
• Memory-Based Optimization Methods for Model-Agnostic Meta-Learning and Personalized Federated Learning
• Online Structured Meta-learning
• Measuring and Harnessing Transference in Multi-Task Learning
• Meta Dialogue Policy Learning
• Generalizing meanings from partners to populations: Hierarchical inference supports convention formation on networks
• Robust Meta-learning for Mixed Linear Regression with Small Batches
• Compositional Few-Shot Recognition with Primitive Discovery and Enhancing
• PAC-Bayes Bounds for Meta-learning with Data-Dependent Prior
• Non-Gaussian Gaussian Processes for Few-Shot Regression
• Meta-Learning Dynamics Forecasting Using Task Inference
• Information Theoretic Meta Learning with Gaussian Processes
• Cross-Modulation Networks for Few-Shot Learning
• Information-Theoretic Analysis of Epistemic Uncertainty in Bayesian Meta-learning
• Gradient-EM Bayesian Meta-learning
• Physics-aware Spatiotemporal Modules with Auxiliary Tasks for Meta-Learning
• Theoretical bounds on estimation error for meta-learning
• ST-MAML: A Stochastic-Task based Method for Task-Heterogeneous Meta-Learning
• An Optimization-Based Meta-Learning Model for MRI Reconstruction with Diverse Dataset
• Meta-learning for mixed linear regression
• Bayesian Meta-reinforcement Learning for Traffic Signal Control
• Task Affinity with Maximum Bipartite Matching in Few-Shot Learning
• An Expectation-Maximization Perspective on Federated Learning
• Few Shot Learning With No Labels
• Uncertainty-Aware Few-Shot Image Classification
• Quantifying Visual Image Quality: A Bayesian View
• Top-Related Meta-Learning Method for Few-Shot Object Detection
• Meta Learning in the Continuous Time Limit
• Decoupling Features and Coordinates for Few-shot RGB Relocalization
• Uniform Sampling over Episode Difficulty
• Generating meta-learning tasks to evolve parametric loss for classification learning
• Is the Meta-Learning Idea Able to Improve the Generalization of Deep Neural Networks on the Standard Supervised Learning?
• Meta-Learning to Improve Pre-Training
• Factorized Machine Self-Confidence for Decision-Making Agents
• Knowledge as Invariance -- History and Perspectives of Knowledge-augmented Machine Learning