Local Energy Redistribution Units for Space Dimensionality Reduction in Data Classification
Data
2023
Autorzy
Tytuł czasopisma
ISSN czasopisma
Tytuł tomu
Wydawca
Wydawnictwo Politechniki Łódzkiej
Lodz University of Technology Press
Lodz University of Technology Press
Abstrakt
In this paper, we present locally trained 2-input to 2-output neurons
called Local Energy Redistribution Units (LERUs), which enable to
transfer most of the input data energy to the selected output, and when organized
into properly designed networks, allow for the energy accumulation
in lower-indexed elements of output vectors. This property can be used to
reduce the dimensionality of the input data space, resulting in a reduction in
the number of weights and disk space needed to store neural network models.
We test the effectiveness of the proposed approach experimentally in the
task of data classification using the well-known MNIST dataset.
Opis
Słowa kluczowe
locally trained neurons, compression of neural networks, neurony wytrenowane lokalnie, kompresja sieci neuronowych
Cytowanie
Puchała D., Local Energy Redistribution Units for Space Dimensionality Reduction in Data Classification. W: Progress in Polish Artificial Intelligence Research 4, Wojciechowski A. (Ed.), Lipiński P. (Ed.)., Seria: Monografie Politechniki Łódzkiej Nr. 2437, Wydawnictwo Politechniki Łódzkiej, Łódź 2023, s. 355-360, ISBN 978-83-66741-92-8, doi: 10.34658/9788366741928.56.