Theoretical Models of Learning to Learn

Abstract

A Machine can only learn if it is biased in some way. Typically the bias is supplied by hand, for example through the choice of an appropriate set of features. However, if the learning machine is embedded within an {\em environment} of related tasks, then it can {\em learn} its own bias by learning sufficiently many tasks from the environment. In this paper two models of bias learning (or equivalently, learning to learn) are introduced and the main theoretical results presented. The first model is a PAC-type model based on empirical process theory, while the second is a hierarchical Bayes model.

arXiv admin note: text overlap with arXiv:1106.0245

Cited by in corpus (42)

• Neural Architecture Search with Reinforcement Learning
• Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
• Searching for Activation Functions
• Massively Multilingual Neural Machine Translation in the Wild: Findings and Challenges
• One-Shot Visual Imitation Learning via Meta-Learning
• Learning Task Grouping and Overlap in Multi-task Learning
• Recasting Gradient-Based Meta-Learning as Hierarchical Bayes
• Learning to Generalize: Meta-Learning for Domain Generalization
• Deep Online Learning via Meta-Learning: Continual Adaptation for Model-Based RL
• Provable Guarantees for Gradient-Based Meta-Learning
• Rewriting History with Inverse RL: Hindsight Inference for Policy Improvement
• Excess risk bounds for multitask learning with trace norm regularization
• Effective Cross-lingual Transfer of Neural Machine Translation Models without Shared Vocabularies
• Revisiting Meta-Learning as Supervised Learning
• Learning Adaptive Loss for Robust Learning with Noisy Labels
• Quantitative toxicity prediction using topology based multi-task deep neural networks
• Sentence Embedding Alignment for Lifelong Relation Extraction
• Gradient-free Policy Architecture Search and Adaptation
• Learning to Transfer
• Generalizing from a few environments in safety-critical reinforcement learning
• Fully-adaptive Feature Sharing in Multi-Task Networks with Applications in Person Attribute Classification
• Structure Learning in Motor Control:A Deep Reinforcement Learning Model
• A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning
• Task Selection Policies for Multitask Learning
• A Sample Complexity Separation between Non-Convex and Convex Meta-Learning
• Incremental Few-Shot Learning for Pedestrian Attribute Recognition
• MxML: Mixture of Meta-Learners for Few-Shot Classification
• Text Generation with Exemplar-based Adaptive Decoding
• Intelligence, physics and information -- the tradeoff between accuracy and simplicity in machine learning
• Meta-Unsupervised-Learning: A supervised approach to unsupervised learning
• Learning State Abstractions for Transfer in Continuous Control
• Discriminative Few-Shot Learning Based on Directional Statistics
• An Optimization Framework for Semi-Supervised and Transfer Learning using Multiple Classifiers and Clusterers
• Hierarchical Meta Learning
• Meta-learning for mixed linear regression
• Visual Transfer Learning: Informal Introduction and Literature Overview
• Local Nonparametric Meta-Learning
• Modular meta-learning in abstract graph networks for combinatorial generalization
• Reasoning about Unforeseen Possibilities During Policy Learning
• Learning Efficient and Effective Exploration Policies with Counterfactual Meta Policy
• Transfer Regression via Pairwise Similarity Regularization
• Personalizing Dialogue Agents via Meta-Learning