Improving Performance in Neural Networks by Dendrite-Activated Connection
Paper in proceeding, 2025
We introduce a novel computational unit for neural networks featuring multiple biases, challenging the conventional perceptron structure. Designed to emphasize preserving uncorrupted information as it transfers from one unit to the next, this unit applies activation functions later in the process, incorporating specialized biases for each unit. We posit this unit as an improved design for neural networks and support this with (1) empirical evidence across diverse datasets; (2) a class of functions where this unit utilizes parameters more efficiently; and (3) biological analogies suggesting closer mimicry to natural neural processing. Source code is available at https://github.com/CuriosAI/dac-dev.
Author
Carlo Metta
Istituto di Scienza e Tecnologie dell'Informazione A. Faedo
Marco Fantozzi
University of Parma
Andrea Papini
Chalmers, Mathematical Sciences, Applied Mathematics and Statistics
University of Gothenburg
Gianluca Amato
G. d'Annunzio University of Chieti-Pescara
Matteo Bergamaschi
University of Padua
Andrea Fois
University of Parma
Silvia Giulia Galfré
University of Pisa
Alessandro Marchetti
G. d'Annunzio University of Chieti-Pescara
Michelangelo Vegliò
G. d'Annunzio University of Chieti-Pescara
Maurizio Parton
G. d'Annunzio University of Chieti-Pescara
Francesco Morandin
University of Parma
Studies in Classification Data Analysis and Knowledge Organization
14318814 (ISSN) 21983321 (eISSN)
133-1419783031847011 (ISBN)
14th Scientific Meeting of the Classification and Data Analysis Group of the Italian Statistical Society, CLADAG 2023
Salerno, Italy,
Salerno, Italy,
Subject Categories (SSIF 2025)
Formal Methods
Computer Engineering
Computational Mathematics
DOI
10.1007/978-3-031-84702-8_15