Work

Functional Correspondence through Dictionary Reuse

Year: 2026

Type: preprint

Abstract: We compared the extent to which functional correspondencebetween corresponding Transformer blocks emerges or is preservedacross different training runs when atoms designatedas active during Source training are reused together with thedictionary scales 𝐿 in subsequent training, relative to a standardDense model. In this study, functional correspondencerefers to the degree to which corresponding Transformerblocks from different training runs exhibit similar functionalresponses. Across the experiments, functional correspondence wasconsistently higher when Source-active dictionary atoms andscales 𝐿 were preserved.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.22037223

Author: Lee, Changhoon

Language: English

Topic: Neural Networks and Applications

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7203913709

Cites:

Cited by:

Related to: