Research map / Reinforcement learning / Visual and representation RL
Influencing Long-Term Behavior in Multiagent Reinforcement Learning
Dong-Ki Kim, Matthew Riemer, Miao Liu, Jakob Nicolaus Foerster, Michael Everett, Chuangchuang Sun, Gerald Tesauro, JONATHAN P HOW
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace discounted-return objective (LILI) with the average-reward objective (FURTHER) | learning-aware MARL agent · Iterated Bach or Stravinsky vs q-learner | convergence to optimal equilibrium (B,B) | Improved |
| model other agents' learning (LILI/FURTHER) versus ignoring it (MASAC) | MARL agent · Iterated Bach or Stravinsky vs q-learner | average reward | Improved |
| increase discount factor gamma toward 1 in LILI (ablation) | LILI · Iterated Bach or Stravinsky vs q-learner | convergence performance and TD error | Hurt |
| FURTHER versus LILI and MASAC in self-play | FURTHER · Iterated cooperative game (self-play) | average reward / equilibrium reached | Improved |
| LILI versus MASAC in self-play (modeling others' learning) | LILI · Iterated cooperative game (self-play) | average reward | Improved |
| FURTHER versus LILI opponent | FURTHER · Iterated matching pennies | relative accumulated reward | Improved |
| FURTHER versus MASAC opponent compared with versus LILI opponent | FURTHER · Iterated matching pennies | relative accumulated reward | Improved |
| FURTHER versus LILI opponent | FURTHER · MuJoCo RoboSumo | averaged winning rate: 60.6 | Improved |
| FURTHER versus MASAC opponent | FURTHER · MuJoCo RoboSumo | averaged winning rate: 63.9 | Improved |
| mean-field FURTHER (FURTHER-MF) versus mean-field LILI (LILI-MF) | FURTHER-MF · MAgent Battle 25 vs 25 | reward over time | Improved |
| greedy versus GLIE learning of opponent j | agent i in IPD · Iterated prisoner's dilemma | dependence of average reward on initial q-values | Improved |
Closest papers by meaning
- Multi-agent cooperation through learning-aware policy gradients (ICLR 2025 · similarity 0.88)
- A Black-box Approach for Non-stationary Multi-agent Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Measuring Mutual Policy Divergence for Multi-Agent Sequential Exploration (NeurIPS 2024 · similarity 0.88)
- Learning Joint Behaviors with Large Variations (AAAI 2025 · similarity 0.88)
- Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration (NeurIPS 2025 · similarity 0.88)
- Revisiting Multi-Agent World Modeling from a Diffusion-Inspired Perspective (NeurIPS 2025 · similarity 0.87)
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.
