Tree-based routing balances expert use without auxiliary loss in MoE models

The BRANCH-MoE architecture uses exponential moving average anchors at binary tree nodes to guarantee utilization and reduce communication.

Big Tech

Gang Fu · Adel Javanmard · MohammadHossein Bateni · Vahab Mirrokni

Google Research · University of Southern California

Research Digest·6 October 2026·3 min read

The authors propose BRANCH-MoE, a routing architecture that places experts at leaves of a binary decision tree.

The authors designed a hierarchical router with a binary tree of depth log₂ E, where E=16 experts are leaf nodes.

Why this paper

From Google Research and University of Southern California

BRANCH-MoE uses a binary decision tree with EMA anchors to balance expert utilization without auxiliary losses.

Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.