Assessing the Performance of Analog Training for Transfer Learning

Omobayode Fagbohungbe 1 3 , Corey Lammie 2 , Malte J. Rasch 1 , Takashi Ando 1 , Tayfun Gokmen 1 , Vijay Narayanan 1 · 2025

Read Assessing the Performance of Analog Training for Transfer Learning by Omobayode Fagbohungbe 1 3 , Corey Lammie 2 , Malte J. Rasch 1 , Takashi Ando 1 , Tayfun Gokmen 1 , Vijay Narayanan 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-11067.txt.

Description

Omobayode Fagbohungbe and co-authors present a 2025 paper on training neural networks in analog memory hardware, and how well that training transfers to new tasks. Such hardware promises fast, frugal learning by computing where the data sit. Real devices, though, switch unevenly and vary from unit to unit, which breaks standard training recipes.

The team assessed a newer algorithm built to cope with that mess, one that chops updates to tame the devices' lopsided response. Earlier rival methods demanded unrealistically perfect devices or proved fragile. The study weighs how the analog-trained models perform when moved to fresh tasks, the transfer setting where the hardware's speed could pay off most.

Catalog record

CREATOROmobayode Fagbohungbe 1 3 , Corey Lammie 2 , Malte J. Rasch 1 , Takashi Ando 1 , Tayfun Gokmen 1 , Vijay Narayanan 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE20 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-11067

Browse the full library · About the Tower of Babel