AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

Emergent Communication Under Misinformation

Heeyoung Lee, Kyungwoo Song

ICML 2026 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Emergent communication in referential games

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace random noise with a malign intermediary (misinformation)emergent communication sender/receiver · synthetic attribute-value game, k=1..4TopSimImproved
replace random noise with misinformationemergent communication sender/receiver · synthetic attribute-value game, k=1..4generalizationImproved
reinitialize the intermediary at six evenly spaced intervals (new generations) (ablation)misinformation game · synthetic attribute-value gameTopSimImproved
intermediary reset (ablation)misinformation game · synthetic attribute-value gameBosDisNo clear effect
add intermediary resets, in (6,7) configuration (ablation)misinformation game · synthetic attribute-value game (N,|X|)=(6,7)compositional generalization: 87.1 -> 96.7Improved
lower intermediary entropy coefficient (less adaptive intermediary) (ablation)misinformation without reset · synthetic attribute-value game, k=3TopSimHurt
train new receivers on languages from misinformation with resetnew receiver learning · synthetic attribute-value gamelearning speedImproved
partial misinformation (attack subset of attributes)misinformation game · synthetic attribute-value game, k=3TopSimImproved
resample target attribute each step instead of prolonged exposure (ablation)partial misinformation · synthetic attribute-value gamecompositionalityHurt
replace partial misinformation with positionally biased random noise (ablation)random noise · synthetic attribute-value game, k=3PosDisHurt
use misinformation with reset versus random noise across message lengths 8-12misinformation game · synthetic attribute-value gameTopSimImproved
misinformation (no reset) vs random noise in (4,18) configuration (ablation)misinformation game · synthetic attribute-value game (N,|X|)=(4,18)TopSim: 46.6 vs 46.6No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy