The Neural Tangent Kernel in High Dimensions: Triple Descent and a Multi-Scale Theory of Generalization

Abstract

Modern deep learning models employ considerably more parameters than required to fit the training data. Whereas conventional statistical wisdom suggests such models should drastically overfit, in practice these models generalize remarkably well. An emerging paradigm for describing this unexpected behavior is in terms of a \emph{double descent} curve, in which increasing a model's capacity causes its test error to first decrease, then increase to a maximum near the interpolation threshold, and then decrease again in the overparameterized regime. Recent efforts to explain this phenomenon theoretically have focused on simple settings, such as linear regression or kernel regression with unstructured random features, which we argue are too coarse to reveal important nuances of actual neural networks. We provide a precise high-dimensional asymptotic analysis of generalization under kernel regression with the Neural Tangent Kernel, which characterizes the behavior of wide neural networks optimized with gradient descent. Our results reveal that the test error has non-monotonic behavior deep in the overparameterized regime and can even exhibit additional peaks and descents when the number of parameters scales quadratically with the dataset size.

arXiv commentPublished as a conference paper in the Proceedings of the 37th International Conference on Machine Learning; 31 pages; 4 figures

Cited by in corpus (19)

• Explaining Neural Scaling Laws
• Learning in High Dimension Always Amounts to Extrapolation
• Dataset Distillation with Infinitely Wide Convolutional Networks
• What causes the test error? Going beyond bias-variance via ANOVA
• Spectra of the Conjugate Kernel and Neural Tangent Kernel for linear-width neural networks
• A Random Matrix Perspective on Mixtures of Nonlinearities for Deep Learning
• Kernel regression in high dimensions: Refined analysis beyond double descent
• Eight challenges in developing theory of intelligence
• Towards an Understanding of Benign Overfitting in Neural Networks
• Taxonomizing local versus global structure in neural network loss landscapes
• Benign Overfitting and Noisy Features
• On the Universality of the Double Descent Peak in Ridgeless Regression
• Minimum ${\ell }_1$-norm interpolators: Precise asymptotics and multiple descent
• Trap of Feature Diversity in the Learning of MLPs
• Model, sample, and epoch-wise descents: exact solution of gradient flow in the random feature model
• Deformed semicircle law and concentration of nonlinear random matrices for ultra-wide neural networks
• Appearance of Random Matrix Theory in Deep Learning
• Covariate Shift in High-Dimensional Random Feature Regression
• Fundamental tradeoffs between memorization and robustness in random features and neural tangent regimes