Hierarchically Structured Meta-learning

Abstract

In order to learn quickly with few samples, meta-learning utilizes prior knowledge learned from previous tasks. However, a critical challenge in meta-learning is task uncertainty and heterogeneity, which can not be handled via globally sharing knowledge among tasks. In this paper, based on gradient-based meta-learning, we propose a hierarchically structured meta-learning (HSML) algorithm that explicitly tailors the transferable knowledge to different clusters of tasks. Inspired by the way human beings organize knowledge, we resort to a hierarchical task clustering structure to cluster tasks. As a result, the proposed approach not only addresses the challenge via the knowledge customization to different clusters of tasks, but also preserves knowledge generalization among a cluster of similar tasks. To tackle the changing of task relationship, in addition, we extend the hierarchical structure to a continual learning environment. The experimental results show that our approach can achieve state-of-the-art performance in both toy-regression and few-shot image classification problems.

arXiv commentICML 2019; Errata: this version fix the results of A1 in Table 10

Cited by in corpus (32)

• Transferring Robustness for Graph Neural Network Against Poisoning Attacks
• GeneraLight: Improving Environment Generalization of Traffic Signal Control via Meta Reinforcement Learning
• Many-Class Few-Shot Learning on Multi-Granularity Class Hierarchy
• Automated Relational Meta-learning
• Doing More with Less: Overcoming Data Scarcity for POI Recommendation via Cross-Region Transfer
• How to Train Your MAML to Excel in Few-Shot Classification
• Few-shot Learning for Time-series Forecasting
• Improving Generalization in Meta-learning via Task Augmentation
• Weighted Meta-Learning
• Online Structured Meta-learning
• Structured Prediction for Conditional Meta-Learning
• Meta-Learning Dynamics Forecasting Using Task Inference
• Graph Few-shot Learning via Knowledge Transfer
• Graceful Degradation and Related Fields
• Meta-Learning with Neural Tangent Kernels
• Few-Shot Unsupervised Continual Learning through Meta-Examples
• Hierarchical Expert Networks for Meta-Learning
• Task-adaptive Neural Process for User Cold-Start Recommendation
• An Optimization-Based Meta-Learning Model for MRI Reconstruction with Diverse Dataset
• Dialogue Generation on Infrequent Sentence Functions via Structured Meta-Learning
• ST-MAML: A Stochastic-Task based Method for Task-Heterogeneous Meta-Learning
• Few-shot Learning for Spatial Regression
• Model-Agnostic Learning to Meta-Learn
• Meta-learning One-class Classifiers with Eigenvalue Solvers for Supervised Anomaly Detection
• Zero-shot task adaptation by homoiconic meta-mapping
• Affinity-Based Hierarchical Learning of Dependent Concepts for Human Activity Recognition
• Meta-learning for Matrix Factorization without Shared Rows or Columns
• Next-item Recommendations in Short Sessions
• Meta-Active Learning for Node Response Prediction in Graphs
• Dataset Bias in Few-shot Image Recognition
• Relation-aware Meta-learning for Market Segment Demand Prediction with Limited Records
• Few-shot Learning for Topic Modeling