New Theory Unifies Trainable Sinusoidal Activations for Neural Representations

Researchers have developed a new theoretical framework for trainable sinusoidal activations in Implicit Neural Representations (INRs), which are used for modeling continuous signals. This framework, called Sinusoidal Trainable Activation Functions (STAF), allows for learned amplitudes, frequencies, and phases. The research provides a mathematical construction, analyzes the impact on the Neural Tangent Kernel, and offers a new initialization method. Empirically, STAF shows competitive or superior performance on various INR tasks, including image reconstruction and NeRF, with improved parameter efficiency. AI

IMPACT Introduces a new activation function family that could improve performance and efficiency in various neural representation tasks.

RANK_REASON This is a research paper detailing a new theoretical framework and empirical results for a novel activation function family in neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Amirhossein Kazerouni
• Implicit Neural Representations
• Sinusoidal Trainable Activation Functions

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →