Sihan Feng

Yuchen Lin, Yong Zhang, Sihan Feng et al.

Advancements in artificial intelligence call for a deeper understanding of the fundamental mechanisms underlying deep learning. In this work, we propose a theoretical framework to analyze learning dynamics through the lens of dynamical systems theory. We redefine the notions of linearity and nonlinearity in neural networks by introducing two fundamental transformation units at the neuron level: order-preserving transformations and non-order-preserving transformations. Different transformation modes lead to distinct collective behaviors in weight vector organization, different modes of information extraction, and the emergence of qualitatively different learning phases. Transitions between these phases may occur during training, accounting for key phenomena such as grokking. To further characterize generalization and structural stability, we introduce the concept of attraction basins in both sample and weight spaces. The distribution of neurons with different transformation modes across layers, along with the structural characteristics of the two types of attraction basins, forms a set of core metrics for analyzing the performance of learning models. Hyperparameters such as depth, width, learning rate, and batch size act as control variables for fine-tuning these metrics. Our framework not only sheds light on the intrinsic advantages of deep learning, but also provides a novel perspective for optimizing network architectures and training strategies.

Sihan Feng, Yong Zhang, Fuming Wang et al.

Despite their great success, neural networks still remain as black-boxes due to the lack of interpretability. Here we propose a new analyzing method, namely the weight pathway analysis (WPA), to make them transparent. We consider weights in pathways that link neurons longitudinally from input neurons to output neurons, or simply weight pathways, as the basic units for understanding a neural network, and decompose a neural network into a series of subnetworks of such weight pathways. A visualization scheme of the subnetworks is presented that gives longitudinal perspectives of the network like radiographs, making the internal structures of the network visible. Impacts of parameter adjustments or structural changes to the network can be visualized via such radiographs. Characteristic maps are established for subnetworks to characterize the enhancement or suppression of the influence of input samples on each output neuron. Using WPA, we discover that neural network store and utilize information in a holographic way, that is, subnetworks encode all training samples in a coherent structure and thus only by investigating the weight pathways can one explore samples stored in the network. Furthermore, with WPA, we reveal fundamental learning modes of a neural network: the linear learning mode and the nonlinear learning mode. The former extracts linearly separable features while the latter extracts linearly inseparable features. The hidden-layer neurons self-organize into different classes for establishing learning modes and for reaching the training goal. The finding of learning modes provides us the theoretical ground for understanding some of the fundamental problems of machine learning, such as the dynamics of learning process, the role of linear and nonlinear neurons, as well as the role of network width and depth.