From Complexity to Clarity: Analytical Expressions of Deep Neural Network Weights via Clifford Algebra and Convexity
Abstract: In this paper, we introduce a novel analysis of neural networks based on geometric (Clifford) algebra and convex optimization. We show that optimal weights of deep ReLU neural networks are given by the wedge product of training samples when trained with standard regularized loss. Furthermore, the training problem reduces to convex optimization over wedge product features, which encode the geometric structure of the training dataset. This structure is given in terms of signed volumes of triangles and parallelotopes generated by data vectors. The convex problem finds a small subset of samples via $\ell_1$ regularization to discover only relevant wedge product features. Our analysis provides a novel perspective on the inner workings of deep neural networks and sheds light on the role of the hidden layers.
Submission Length: Long submission (more than 12 pages of main content)
Changes Since Last Submission: Section 3.4 Illustrative Calculations in Geometric Algebra added.
Section 4.1.1 added.
Section 4.1.3 and Section 5 are expanded.
Figures 6 10, 11, 12 are added
Figures 2 and 3 are revised.
Typos are fixed.
Remarks are added after certain theorems.
Supplementary Material: zip
Assigned Action Editor: ~Kangwook_Lee1
Submission Number: 2636
Loading