Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Agent Modelling under Partial Observability for Deep Reinforcement Learning
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 |
|---|---|---|---|
| Use LIAM representations learned from the controlled agent's local trajectory, with modelled-agent trajectories as training targets | A2C policy · Double speaker-listener, level-based foraging, and predator-prey | episodic return | Helped |
| Replace LIAM's trajectory reconstruction representation with VariBad task representations | A2C policy · Double speaker-listener, level-based foraging, and predator-prey | episodic return | Hurt |
| Replace LIAM's reconstruction objective with contrastive agent representation learning | A2C policy · Double speaker-listener, level-based foraging, and predator-prey | episodic return | Hurt |
| Represent the other agent by its policy identity instead of reconstructing its trajectory | A2C policy · Double speaker-listener | episodic return | Mixed |
| Represent the other agent by its policy identity instead of reconstructing its trajectory | A2C policy · Level-based foraging and predator-prey | episodic return | Hurt |
| Remove reconstruction of modelled-agent actions (ablation) | LIAM · Double speaker-listener | episodic return | Hurt |
| Remove reconstruction of modelled-agent observations (ablation) | LIAM · Double speaker-listener | episodic return | Hurt |
| Remove controlled-agent actions from the encoder input (ablation) | LIAM · Double speaker-listener | episodic return | Hurt |
| Remove controlled-agent observations from the encoder input (ablation) | LIAM · Double speaker-listener | episodic return | Hurt |
| Train the decoder to reconstruct the controlled agent's next observation and action instead of the modelled agent's trajectory (ablation) | LIAM-Local · Double speaker-listener | episodic return | Hurt |
| Use a variational encoder-decoder with an HMM latent prior (ablation) | LIAM-VAE · Double speaker-listener | episodic return | Hurt |
| Add a beta coefficient or MMD-squared distance to adjust VAE regularisation (ablation) | LIAM-VAE · Double speaker-listener | episodic return | No clear effect |
Closest papers by meaning
- Learning Policy Representations in Multiagent Systems (ICML 2018 · similarity 0.90)
- MA$^2$E: Addressing Partial Observability in Multi-Agent Reinforcement Learning with Masked Auto-Encoder (ICLR 2025 · similarity 0.89)
- Enhancing Cooperative Multi-Agent Reinforcement Learning with State Modelling and Adversarial Exploration (ICML 2025 · similarity 0.88)
- Studying the Interplay Between the Actor and Critic Representations in Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Mimicking To Dominate: Imitation Learning Strategies for Success in Multiagent Games (NeurIPS 2024 · similarity 0.85)
- Provable Partially Observable Reinforcement Learning with Privileged Information (NeurIPS 2024 · similarity 0.85)
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.