PHYDI: Initializing Parameterized Hypercomplex Neural Networks as Identity Functions
This addresses a convergence issue in hypercomplex neural networks for applications like computer vision and NLP, but it is incremental as it builds on existing PHNN architectures.
The paper tackles the problem of controlling convergence in large-scale parameterized hypercomplex neural networks (PHNNs) by proposing PHYDI, a method that improves convergence across scales, leading to more robust performance with increased layers and achieving the same performance in fewer iterations.
Neural models based on hypercomplex algebra systems are growing and prolificating for a plethora of applications, ranging from computer vision to natural language processing. Hand in hand with their adoption, parameterized hypercomplex neural networks (PHNNs) are growing in size and no techniques have been adopted so far to control their convergence at a large scale. In this paper, we study PHNNs convergence and propose parameterized hypercomplex identity initialization (PHYDI), a method to improve their convergence at different scales, leading to more robust performance when the number of layers scales up, while also reaching the same performance with fewer iterations. We show the effectiveness of this approach in different benchmarks and with common PHNNs with ResNets- and Transformer-based architecture. The code is available at https://github.com/ispamm/PHYDI.