Research Paper#Convolutional Neural Networks (CNNs), Physics-Inspired Models🔬 ResearchAnalyzed: Jan 3, 2026 15:37
CNN Filtering with Rectification: A Physics-Inspired Model
Published:Dec 30, 2025 16:44
•1 min read
•ArXiv
Analysis
This paper introduces a novel perspective on understanding Convolutional Neural Networks (CNNs) by drawing parallels to concepts from physics, specifically special relativity and quantum mechanics. The core idea is to model kernel behavior using even and odd components, linking them to energy and momentum. This approach offers a potentially new way to analyze and interpret the inner workings of CNNs, particularly the information flow within them. The use of Discrete Cosine Transform (DCT) for spectral analysis and the focus on fundamental modes like DC and gradient components are interesting. The paper's significance lies in its attempt to bridge the gap between abstract CNN operations and well-established physical principles, potentially leading to new insights and design principles for CNNs.
Key Takeaways
- •Proposes a new model for understanding CNN filtering based on physical principles.
- •Decomposes kernels into even and odd components, analogous to energy and momentum.
- •Uses Discrete Cosine Transform (DCT) for spectral analysis.
- •Links information processing in CNNs to the energy-momentum relation.
Reference
“The speed of information displacement is linearly related to the ratio of odd vs total kernel energy.”