Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Emergent Quantized Communication
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace Gumbel-softmax with quantized communication | Instant channel · Object game, 10000 candidates | accuracy: 0.008 -> 1.000 | Helped |
| replace Gumbel-softmax with quantized communication | Instant channel · Image game, 100 candidates | accuracy: 0.132 -> 0.962 | Helped |
| replace Gumbel-softmax with quantized communication | Instant channel · Sent-Ref, 77 candidates | accuracy: 0.035 -> 0.945 | Helped |
| replace Gumbel-softmax with quantized communication | Instant channel · Sent-Cls, 77 candidates | accuracy: 0.258 -> 0.925 | Helped |
| apply quantization during training as well as inference (ablation) | Quantized Instant channel · Object game, 2 and maximum candidates | accuracy: 1.00 -> 1.00; 1.00 -> 1.00 | No clear effect |
| apply quantization during training as well as inference (ablation) | Quantized Instant channel · Image game, 2 and maximum candidates | accuracy: 0.99 -> 1.00; 0.96 -> 0.96 | No clear effect |
| apply quantization during training as well as inference (ablation) | Quantized Instant channel · Sent-Ref, 2 and maximum candidates | accuracy: 0.99 -> 1.00; 0.95 -> 0.95 | No clear effect |
| apply quantization during training as well as inference (ablation) | Quantized Instant channel · Sent-Cls, 2 and maximum candidates | accuracy: 0.99 -> 0.99; 0.92 -> 0.92 | No clear effect |
| increase word length (ablation) | Quantized communication · Object game, Instant channel | accuracy | Helped |
| increase alphabet size (ablation) | Quantized communication · Object game, Instant channel | accuracy | Helped |
| use a binary alphabet with word length 10 (ablation) | Quantized communication · Sent-Cls | accuracy: 91% | Helped |
Closest papers by meaning
- RGMComm: Return Gap Minimization via Discrete Communications in Multi-Agent Reinforcement Learning (AAAI 2024 · similarity 0.86)
- Learning Multi-Agent Communication with Contrastive Learning (ICLR 2024 · similarity 0.85)
- Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning (ICLR 2025 · similarity 0.85)
- Expressive Multi-Agent Communication via Identity-Aware Learning (AAAI 2024 · similarity 0.85)
- Multi-Agent Reinforcement Learning with Communication-Constrained Priors (NeurIPS 2025 · similarity 0.85)
- A compressive-expressive communication framework for compositional representations (NeurIPS 2025 · similarity 0.84)
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.