Meta Dropout: Learning to Perturb Features for Generalization

Abstract

A machine learning model that generalizes well should obtain low errors on unseen test examples. Thus, if we know how to optimally perturb training examples to account for test examples, we may achieve better generalization performance. However, obtaining such perturbation is not possible in standard machine learning frameworks as the distribution of the test data is unknown. To tackle this challenge, we propose a novel regularization method, meta-dropout, which learns to perturb the latent features of training examples for generalization in a meta-learning framework. Specifically, we meta-learn a noise generator which outputs a multiplicative noise distribution for latent features, to obtain low errors on the test instances in an input-dependent manner. Then, the learned noise generator can perturb the training examples of unseen tasks at the meta-test time for improved generalization. We validate our method on few-shot classification datasets, whose results show that it significantly improves the generalization performance of the base model, and largely outperforms existing regularization methods such as information bottleneck, manifold mixup, and information dropout.

References in corpus (17)

• Auto-Encoding Variational Bayes
• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Prototypical Networks for Few-shot Learning
• The information bottleneck method
• Towards Deep Learning Models Resistant to Adversarial Attacks
• Theoretical Models of Learning to Learn
• A Simple Neural Attentive Meta-Learner
• Meta-SGD: Learning to Learn Quickly for Few-Shot Learning
• On First-Order Meta-Learning Algorithms
• DropBlock: A regularization method for convolutional networks
• Variational Dropout and the Local Reparameterization Trick
• Regularizing Neural Networks by Penalizing Confident Output Distributions
• Fast Context Adaptation via Meta-Learning
• Adversarial Training and Robustness for Multiple Perturbations
• Uncertainty-Aware Attention for Reliable Interpretation and Prediction

Cited by in corpus (3)

• Adaptive Adversarial Training for Meta Reinforcement Learning
• Meta-Regularization by Enforcing Mutual-Exclusiveness
• Meta-Learning with Task-Adaptive Loss Function for Few-Shot Learning