Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2609.23310.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Analyzes when communication improves or degrades collective decisions in multi-agent systems. Shows that spectral descriptions of network dynamics can hide direction of accuracy gain, and proposes a task-projected local-response approximation that predicts multi-round communication gains in trained agents with 0.45 percentage point RMSE. Also examines how community-shared bias distributes benefits unevenly.

For those designing multi-agent pipelines: identical-looking networks can flip accuracy from 91% to 66% depending on message orientation, and standard spectral metrics miss the sign. A concrete cautionary finding with a predictive method attached.

Cited by

No citations on record.