Modeling and Optimization Trade-off in Meta-learning

Abstract

By searching for shared inductive biases across tasks, meta-learning promises to accelerate learning on novel tasks, but with the cost of solving a complex bilevel optimization problem. We introduce and rigorously define the trade-off between accurate modeling and optimization ease in meta-learning. At one end, classic meta-learning algorithms account for the structure of meta-learning but solve a complex optimization problem, while at the other end domain randomized search (otherwise known as joint training) ignores the structure of meta-learning and solves a single level optimization problem. Taking MAML as the representative meta-learning algorithm, we theoretically characterize the trade-off for general non-convex risk functions as well as linear regression, for which we are able to provide explicit bounds on the errors associated with modeling and optimization. We also empirically study this trade-off for meta-reinforcement learning benchmarks.

arXiv commentTo appear at NeurIPS 2020

References in corpus (15)

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Trust Region Policy Optimization
• On First-Order Meta-Learning Algorithms
• RL${}^2$: Fast Reinforcement Learning via Slow Reinforcement Learning
• Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning
• Frustratingly Simple Few-Shot Object Detection
• SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning
• Meta-Learning with Implicit Gradients
• Assessing Generalization in Deep Reinforcement Learning
• Bilevel Programming for Hyperparameter Optimization and Meta-Learning
• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm
• Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need?
• On the Iteration Complexity of Hypergradient Computation
• On the Global Optimality of Model-Agnostic Meta-Learning

Cited by in corpus (4)

• How Important is the Train-Validation Split in Meta-Learning?
• Artificial Inductive Bias for Synthetic Tabular Data Generation in Data-Scarce Scenarios
• How Does the Task Landscape Affect MAML Performance?
• FLIX: A Simple and Communication-Efficient Alternative to Local Methods in Federated Learning