DOAJ Open Access 2025

On non-approximability of zero loss global L2 minimizers by gradient descent in deep learning

Chen Thomas Muñoz Ewald Patricia

Lihat Sumber DOI

Abstrak

We analyze geometric aspects of the gradient descent algorithm in Deep Learning (DL), and give a detailed discussion of the circumstance that, in underparametrized DL networks, zero loss minimization cannot generically be attained. As a consequence, we conclude that the distribution of training inputs must necessarily be non-generic in order to produce zero loss minimizers, both for the method constructed in [2, 3], or for gradient descent [1] (which assume clustering of training data).

Topik & Kata Kunci

Mechanics of engineering. Applied mechanics

Penulis (2)

Chen Thomas

Muñoz Ewald Patricia

Format Sitasi

APA MLA BibTeX

Thomas, C., Patricia, M.E. (2025). On non-approximability of zero loss global L2 minimizers by gradient descent in deep learning. https://doi.org/10.2298/TAM250121008C

Akses Cepat

Lihat di Sumber doi.org/10.2298/TAM250121008C

Informasi Jurnal

Tahun Terbit: 2025
Sumber Database: DOAJ
DOI: 10.2298/TAM250121008C
Akses: Open Access ✓