A Closer Look at Memorization in Deep Networks

Abstract

We examine the role of memorization in deep learning, drawing connections to capacity, generalization, and adversarial robustness. While deep networks are capable of memorizing noise data, our results suggest that they tend to prioritize learning simple patterns first. In our experiments, we expose qualitative differences in gradient-based optimization of deep neural networks (DNNs) on noise vs. real data. We also demonstrate that for appropriately tuned explicit regularization (e.g., dropout) we can degrade DNN training performance on noise datasets without compromising generalization on real data. Our analysis suggests that the notions of effective capacity which are dataset independent are unlikely to explain the generalization performance of deep networks when trained with gradient based methods because training data itself plays an important role in determining the degree of memorization.

Appears in Proceedings of the 34th International Conference on Machine Learning (ICML 2017), Devansh Arpit, Stanisław Jastrzębski, Nicolas Ballas, and David Krueger contributed equally to this work

References in corpus (3)

• Explaining and Harnessing Adversarial Examples
• On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima
• Gradient-based Hyperparameter Optimization through Reversible Learning

Cited by in corpus (15)

• Deep Learning Scaling is Predictable, Empirically
• How does Disagreement Help Generalization against Label Corruption?
• Overfitting Mechanism and Avoidance in Deep Neural Networks
• Generalization in Deep Networks: The Role of Distance from Initialization
• Explicitizing an Implicit Bias of the Frequency Principle in Two-layer Neural Networks
• Meta-learners' learning dynamics are unlike learners'
• A Closer Look at Data Bias in Neural Extractive Summarization Models
• Rethinking Generalization of Neural Models: A Named Entity Recognition Case Study
• Providing theoretical learning guarantees to Deep Learning Networks
• A Meta Approach to Defend Noisy Labels by the Manifold Regularizer PSDR
• Investigating performance of neural networks and gradient boosting models approximating microscopic traffic simulations in traffic optimization tasks
• A Simple yet Effective Baseline for Robust Deep Learning with Noisy Labels
• Network Cooperation with Progressive Disambiguation for Partial Label Learning
• Impact of Training Dataset Size on Neural Answer Selection Models
• An investigation of model-free planning