AI Paper Atlas

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

Learning to Communicate and Solve Visual Blocks-World Tasks

Qi Zhang, Richard Lewis, Satinder Singh, Edmund Durfee

AAAI 2019 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: Emergent communication in referential games · 1 citations (OpenAlex)

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-string utterances with utterances of frequent N-grams (ablation)RL speaker-listener · grammatical actions, B=15mean grammatical actions: 3.15 -> 7.60Improved
replace random-string utterances with utterances of frequent N-grams (ablation)RL speaker-listener · grammatical actions, B=25mean grammatical actions: 2.75 -> 7.14Improved
replace random-string utterances with utterances of frequent N-grams (ablation)Bandit speaker-listener · grammatical actions, B=15mean grammatical actions: 4.33 -> 10.11Improved
replace random-string utterances with utterances of frequent N-grams (ablation)Bandit speaker-listener · grammatical actions, B=25mean grammatical actions: 5.47 -> 7.24Improved
replace random-string utterances with utterances of frequent N-grams (ablation)SL speaker-listener · grammatical actions, B=15mean grammatical actions: 6.29 -> 8.06Improved
replace random-string utterances with utterances of frequent N-grams (ablation)SL speaker-listener · grammatical actions, B=25mean grammatical actions: 6.34 -> 6.04No clear effect
add entropy regularization to SL training (ablation)SL speaker-listener · extrapolation, partial completionpartial completionImproved
train with Bandit instead of SLspeaker-listener · extrapolationpartial completionImproved
train with Bandit or RL instead of SLsingle-agent baseline · interpolation/extrapolationfull and partial completionImproved
use speaker-listener instead of single agentblocks-world agents · extrapolationfull and partial completionHurt
sample listener actions from softmax instead of greedyBandit listener · B=25 partial completionpartial completionNo clear effect
sample listener actions from softmax instead of greedySL listener · B=25 partial completionpartial completionNo 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