On Linear Mode Connectivity of Mixture-of-Experts Architectures

Viet-Hoang Tran Van Hoan Trinh Khanh Vinh Bui Tan M. Nguyen · 2025

Read On Linear Mode Connectivity of Mixture-of-Experts Architectures by Viet-Hoang Tran Van Hoan Trinh Khanh Vinh Bui Tan M. Nguyen (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-2509-11348.txt.

Description

Viet-Hoang Tran and three co-authors test a geometric property in mixture-of-experts models. This 2025 machine learning paper concerns linear mode connectivity. Two separately trained networks may join by a straight line in weight space. Loss stays low along the line once inner symmetries line up.

The team first pins down the symmetries of mixture-of-experts models in full. They act by permuting the expert networks. They also permute the gating function that routes inputs. With that in hand, the team built a matching algorithm. It aligns two trained models so the joining path can be found. Tests spanned dense, sparse, and shared-expert variants. Data of many scales and types were used. Connectivity held across the board.

Catalog record

CREATORViet-Hoang Tran Van Hoan Trinh Khanh Vinh Bui Tan M. Nguyen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE156 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-11348

Browse the full library · About the Tower of Babel