Items or Relations -- what do Artificial Neural Networks learn?
This addresses a fundamental question in understanding neural network behavior, with implications for improving generalization in AI systems, though it is incremental as it uses simplified models.
The study investigates whether artificial neural networks learn individual training items or the relations between them, finding that linear networks generalize by capturing symmetries, while non-linear networks tend to memorize items with limited generalization.
What has an Artificial Neural Network (ANN) learned after being successfully trained to solve a task - the set of training items or the relations between them? This question is difficult to answer for modern applied ANNs because of their enormous size and complexity. Therefore, here we consider a low-dimensional network and a simple task, i.e., the network has to reproduce a set of training items identically. We construct the family of solutions analytically and use standard learning algorithms to obtain numerical solutions. These numerical solutions differ depending on the optimization algorithm and the weight initialization and are shown to be particular members of the family of analytical solutions. In this simple setting, we observe that the general structure of the network weights represents the training set's symmetry group, i.e., the relations between training items. As a consequence, linear networks generalize, i.e., reproduce items that were not part of the training set but are consistent with the symmetry of the training set. In contrast, non-linear networks tend to learn individual training items and show associative memory. At the same time, their ability to generalize is limited. A higher degree of generalization is obtained for networks whose activation function contains a linear regime, such as tanh. Our results suggest ANN's ability to generalize - instead of learning items - could be improved by generating a sufficiently big set of elementary operations to represent relations and strongly depends on the applied non-linearity.