On the geometry of generalization and memorization in deep neural networks

Abstract

Understanding how large neural networks avoid memorizing training data is key to explaining their high generalization performance. To examine the structure of when and where memorization occurs in a deep network, we use a recently developed replica-based mean field theoretic geometric analysis method. We find that all layers preferentially learn from examples which share features, and link this behavior to generalization performance. Memorization predominately occurs in the deeper layers, due to decreasing object manifolds' radius and dimension, whereas early layers are minimally affected. This predicts that generalization can be restored by reverting the final few layer weights to earlier epochs before significant memorization occurred, which is confirmed by the experiments. Additionally, by studying generalization under different model sizes, we reveal the connection between the double descent phenomenon and the underlying model geometry. Finally, analytical analysis shows that networks avoid memorization early in training because close to initialization, the gradient contribution from permuted examples are small. These findings provide quantitative evidence for the structure of memorization across layers of a deep neural network, the drivers for such structure, and its connection to manifold geometric properties.

arXiv commentICLR 2021

References in corpus (12)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
• Deep Learning is Robust to Massive Label Noise
• Learning Overparameterized Neural Networks via Stochastic Gradient Descent on Structured Data
• SVCCA: Singular Vector Canonical Correlation Analysis for Deep Learning Dynamics and Interpretability
• Classification and Geometry of General Perceptual Manifolds
• Are All Layers Created Equal?
• An analytic theory of generalization dynamics and transfer learning in deep linear networks
• Linear Readout of Object Manifolds
• Emergence of Separable Manifolds in Deep Language Representations
• Untangling in Invariant Speech Recognition

Cited by in corpus (2)

• Deep Learning Through the Lens of Example Difficulty
• When and how epochwise double descent happens