EWGN: Elastic Weight Generation and Context Switching in Deep Learning

Shriraj P. Sawant, Krishna P. Miyapuram · 2025

Read EWGN: Elastic Weight Generation and Context Switching in Deep Learning by Shriraj P. Sawant, Krishna P. Miyapuram (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-2506-02065.txt.

Description

Shriraj Sawant and Krishna Miyapuram introduce, in a 2025 paper, a neural network design meant to reduce catastrophic forgetting when a model learns tasks in sequence. Forgetting is the familiar failure where training on a new task wrecks performance on an older one. Humans switch contexts fluidly; networks usually cannot.

Their design adds a second network that generates the main network's weights on the fly. Because generation depends on the input, the system can switch weight sets by context, keeping tasks from interfering. The generator also consolidates what was learned, rather than overwriting it. The authors tested the idea on two standard image data sets, handwritten digits and fashion items. They analyzed how well earlier task representations survived in fully connected networks. The work frames weight generation as a route to models that keep learning without erasing their past.

Catalog record

CREATORShriraj P. Sawant, Krishna P. Miyapuram
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-02065

Browse the full library · About the Tower of Babel