High-dimensional dynamics of generalization error in neural networks
From MaRDI portal
Publication:2057778
Abstract: We perform an average case analysis of the generalization dynamics of large neural networks trained using gradient descent. We study the practically-relevant "high-dimensional" regime where the number of free parameters in the network is on the order of or even larger than the number of examples in the dataset. Using random matrix theory and exact solutions in linear models, we derive the generalization error and training error dynamics of learning and analyze how they depend on the dimensionality of data and signal to noise ratio of the learning problem. We find that the dynamics of gradient descent learning naturally protect against overtraining and overfitting in large networks. Overtraining is worst at intermediate network sizes, when the effective number of free parameters equals the number of samples, and thus can be reduced by making a network smaller or larger. Additionally, in the high-dimensional regime, low generalization error requires starting with small initial weights. We then turn to non-linear neural networks, and show that making networks very large does not harm their generalization performance. On the contrary, it can in fact reduce overtraining, even without early stopping or regularization of any sort. We identify two novel phenomena underlying this behavior in overcomplete models: first, there is a frozen subspace of the weights in which no learning occurs under gradient descent; and second, the statistical properties of the high-dimensional regime yield better-conditioned input correlations which protect against overtraining. We demonstrate that naive application of worst-case theories such as Rademacher complexity are inaccurate in predicting the generalization performance of deep neural networks, and derive an alternative bound which incorporates the frozen subspace and conditioning effects and qualitatively matches the behavior observed in simulation.
Recommendations
- Hausdorff dimension, heavy tails, and generalization in neural networks*
- scientific article; zbMATH DE number 1728675
- Generalization Error Analysis of Neural Networks with Gradient Based Regularization
- scientific article; zbMATH DE number 1182753
- scientific article; zbMATH DE number 1453028
- Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness
- Generalization and chaos in a layered neural network
- Generalization Error in Deep Learning
- Generalization and learning error for nonlinear perceptron.
Cites work
- 10.1162/153244303321897690
- A mean field view of the landscape of two-layer neural networks
- Benign overfitting in linear regression
- Dynamics of Learning Near Singularities in Layered Networks
- Dynamics of learning near singularities in radial basis function networks
- Foundations of machine learning
- Generalization in a linear perceptron in the presence of noise
- scientific article; zbMATH DE number 3244317 (Why is no real title available?)
- Learning and generalization in a linear perceptron stochastically trained with noisy data
- On early stopping in gradient descent learning
- Reconciling modern machine-learning practice and the classical bias-variance trade-off
- Spectral analysis of large dimensional random matrices
- Statistical mechanics of complex neural systems and high dimensional data
- Statistical mechanics of learning
- The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices
- The singular values and vectors of low rank perturbations of large rectangular random matrices
- Universal approximation bounds for superpositions of a sigmoidal function
- Universality of covariance matrices
Cited in
(56)- Generalization and learning error for nonlinear perceptron.
- Rademacher complexity and the generalization error of residual networks
- Fast generalization error bound of deep learning without scale invariance of activation functions
- Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness
- A phase transition for finding needles in nonlinear haystacks with LASSO artificial neural networks
- The interpolation phase transition in neural networks: memorization and generalization under lazy training
- Surprises in high-dimensional ridgeless least squares interpolation
- Geometry and generalization: eigenvalues as predictors of where a network will fail to generalize
- An analysis of training and generalization errors in shallow and deep networks
- A comparative analysis of optimization and generalization properties of two-layer neural network and random feature models under gradient descent dynamics
- Landscape and training regimes in deep learning
- Misspecified diffusion models with high-frequency observations and an application to neural networks
- A statistician teaches deep learning
- Free dynamics of feature learning processes
- A random matrix analysis of random Fourier features: beyond the Gaussian kernel, a precise phase transition, and the corresponding double descent*
- Generalisation error in learning with random features and the hidden manifold model*
- Hausdorff dimension, heavy tails, and generalization in neural networks*
- Generalization in Overparameterized Models
- Align, then memorise: the dynamics of learning with feedback alignment
- On the Exact Computation of Linear Frequency Principle Dynamics and Its Generalization
- Align, then memorise: the dynamics of learning with feedback alignment*
- Generalization error rates in kernel regression: the crossover from the noiseless to noisy regime*
- An analytical theory of curriculum learning in teacher–student networks*
- The inverse variance-flatness relation in stochastic gradient descent is critical for finding flat minima
- Overparameterization and generalization error: weighted trigonometric interpolation
- The optimal ridge penalty for real-world high-dimensional data can be zero or negative due to the implicit ridge regularization
- What causes the test error? Going beyond bias-variance via ANOVA
- Comparing dynamics: deep neural networks versus glassy systems
- Scaling description of generalization with number of parameters in deep learning
- Disentangling feature and lazy training in deep neural networks
- Dynamics of stochastic gradient descent for two-layer neural networks in the teacher–student setup*
- A Unifying Tutorial on Approximate Message Passing
- A jamming transition from under- to over-parametrization affects generalization in deep learning
- Prediction errors for penalized regressions based on generalized approximate message passing
- Large-dimensional random matrix theory and its applications in deep learning and wireless communications
- An instance-dependent simulation framework for learning with label noise
- Regularization, early-stopping and dreaming: a Hopfield-like setup to address generalization and overfitting
- Precise learning curves and higher-order scaling limits for dot-product kernel regression
- Self-consistent dynamical field theory of kernel evolution in wide neural networks
- Redundant representations help generalization in wide neural networks
- Tuning parameters of deep neural network training algorithms pays off: a computational study
- Functional data analysis using deep neural networks
- Deep networks for system identification: a survey
- Dropout drops double descent
- Neural networks trained with SGD learn distributions of increasing complexity^*
- Understanding the initial condensation of convolutional neural networks
- On the robustness of the minimim _2 interpolator
- Step-by-step diffusion: an elementary tutorial
- Phase transition analysis for shallow neural networks with arbitrary activation functions
- Stochastic collapse: how gradient noise attracts SGD dynamics towards simpler subnetworks
- Dimension free ridge regression
- Cascade of phase transitions in the training of energy-based models
- Learning from higher-order statistics, efficiently: hypothesis tests, random features, and neural networks
- Scaling and renormalization in high-dimensional regression
- Casual attention: adaptive enforcement of causality in physics-informed neural networks
- The statistical physics of learning in under- and over-parameterized layered neural networks
This page was built for publication: High-dimensional dynamics of generalization error in neural networks
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2057778)
