Effect of the initial configuration of weights on the training and function of artificial neural networks
This research provides insights into the training dynamics of neural networks, specifically for researchers and practitioners using SGD, by suggesting its efficiency is limited to the vicinity of initial weight configurations.
This paper statistically characterizes the deviation of weights in two-hidden-layer ReLU networks trained with SGD from their initial random configurations. It found that successful training keeps the network weights in close proximity to their initial state, with an abrupt increase in deviation observed within the overfitting region, coinciding with a similar jump in the loss function.
The function and performance of neural networks is largely determined by the evolution of their weights and biases in the process of training, starting from the initial configuration of these parameters to one of the local minima of the loss function. We perform the quantitative statistical characterization of the deviation of the weights of two-hidden-layer ReLU networks of various sizes trained via Stochastic Gradient Descent (SGD) from their initial random configuration. We compare the evolution of the distribution function of this deviation with the evolution of the loss during training. We observed that successful training via SGD leaves the network in the close neighborhood of the initial configuration of its weights. For each initial weight of a link we measured the distribution function of the deviation from this value after training and found how the moments of this distribution and its peak depend on the initial weight. We explored the evolution of these deviations during training and observed an abrupt increase within the overfitting region. This jump occurs simultaneously with a similarly abrupt increase recorded in the evolution of the loss function. Our results suggest that SGD's ability to efficiently find local minima is restricted to the vicinity of the random initial configuration of weights.