Exploring Loss Landscapes through the Lens of Spin Glass Theory

Abstract

In the past decade, significant strides in deep learning have led to numerous groundbreaking applications. Despite these advancements, the understanding of the high generalizability of deep learning, especially in such an over-parametrized space, remains limited. Successful applications are often considered as empirical rather than scientific achievements. For instance, deep neural networks' (DNNs) internal representations, decision-making mechanism, absence of overfitting in an over-parametrized space, high generalizability, etc., remain less understood. This paper delves into the loss landscape of DNNs through the lens of spin glass in statistical physics, i.e. a system characterized by a complex energy landscape with numerous metastable states, to better understand how DNNs work. We investigated a single hidden layer Rectified Linear Unit (ReLU) neural network model, and introduced several protocols to examine the analogy between DNNs (trained with datasets including MNIST and CIFAR10) and spin glass. Specifically, we used (1) random walk in the parameter space of DNNs to unravel the structures in their loss landscape; (2) a permutation-interpolation protocol to study the connection between copies of identical regions in the loss landscape due to the permutation symmetry in the hidden layers; (3) hierarchical clustering to reveal the hierarchy among trained solutions of DNNs, reminiscent of the so-called Replica Symmetry Breaking (RSB) phenomenon (i.e. the Parisi solution) in analogy to spin glass; (4) finally, we examine the relationship between the degree of the ruggedness of the loss landscape of the DNN and its generalizability, showing an improvement of flattened minima.