Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization

07/24/2023

∙

This paper studies the problem of training a two-layer ReLU network for binary classification using gradient flow with small initialization. We consider a training dataset with well-separated input vectors: Any pair of input data with the same label are positively correlated, and any pair with different labels are negatively correlated. Our analysis shows that, during the early phase of training, neurons in the first layer try to align with either the positive data or the negative data, depending on its corresponding weight on the second layer. A careful analysis of the neurons' directional dynamics allows us to provide an 𝒪(log n/√(μ)) upper bound on the time it takes for all neurons to achieve good alignment with the input data, where n is the number of data points and μ measures how well the data are separated. After the early alignment phase, the loss converges to zero at a 𝒪(1/t) rate, and the weight matrix on the first layer is approximately low-rank. Numerical experiments on the MNIST dataset illustrate our theoretical findings.

READ FULL TEXT

Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization

Learning a Neuron by a Shallow ReLU Network: Dynamics and Implicit Bias for Correlated Inputs

Magnitude and Angle Dynamics in Training Single ReLU Neurons

Random Feature Amplification: Feature Learning and Generalization in Neural Networks

On the Effective Number of Linear Regions in Shallow Univariate ReLU Networks: Convergence Guarantees and Implicit Bias

Global Convergence of Sobolev Training for Overparametrized Neural Networks

Neural Random Projection: From the Initial Task To the Input Similarity Problem

Training shallow ReLU networks on noisy data using hinge loss: when do we overfit and is it benign?

Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization

Related Research

Learning a Neuron by a Shallow ReLU Network: Dynamics and Implicit Bias for Correlated Inputs

Magnitude and Angle Dynamics in Training Single ReLU Neurons

Random Feature Amplification: Feature Learning and Generalization in Neural Networks

On the Effective Number of Linear Regions in Shallow Univariate ReLU Networks: Convergence Guarantees and Implicit Bias

Global Convergence of Sobolev Training for Overparametrized Neural Networks

Neural Random Projection: From the Initial Task To the Input Similarity Problem

Training shallow ReLU networks on noisy data using hinge loss: when do we overfit and is it benign?