Read the original document by opening this link in a new tab.
Table of Contents
1. Introduction
2. Related Work and Notation
3. Method
- Single Polynomial Models
- CCP (Coupled CP decomposition)
- NCP (Nested coupled CP decomposition)
- NCP-Skip (Nested coupled CP decomposition with skip)
4. Comparison between the Models
Summary
Deep Polynomial Neural Networks by G. G. Chrysos et al introduces a new class of function approximators called '-Nets' based on polynomial expansions. These polynomial neural networks demonstrate high expressiveness and achieve state-of-the-art results in tasks such as image generation, face verification, and 3D mesh representation learning. The paper discusses the limitations of existing structures and proposes novel architectures for neural networks based on tensor decompositions. The authors present three models for polynomial approximations: CCP, NCP, and NCP-Skip, each offering a unique approach to reducing parameters and improving performance. The models are compared in terms of their recursive forms and decompositions, showcasing the potential of polynomial networks in learning high-dimensional distributions.