Research map / Multi-agent systems and games / Human-AI and goal-conditioned
On the Critical Role of Conventions in Adaptive Human-AI Collaboration
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 |
|---|---|---|---|
| add marginal-regularization (lambda=0.5) to modular policy | Modular policy agent · Collaborative contextual bandit, zero-shot new task Arms 1 | zero-shot score: 0.30 -> 0.99 | Helped |
| add marginal-regularization (lambda=0.5) to modular policy | Modular policy agent · Collaborative contextual bandit, zero-shot new task Arms 2 | zero-shot score: 0.55 -> 0.89 | Helped |
| add marginal-regularization (lambda=0.5) to modular policy | Modular policy agent · Collaborative contextual bandit, zero-shot new task Arms 3 | zero-shot score: 0.56 -> 0.91 | Helped |
| add marginal-regularization (lambda=0.5) to modular policy | Modular policy agent · Block placing, zero-shot new task | zero-shot score: 13.17 -> 17.68 | Helped |
| set lambda=0 (no marginal regularization) (ablation) | Modular policy agent · Contextual bandit marginals | Wasserstein distance to oracle marginals | No clear effect |
| set lambda=0.5 versus other lambda values (ablation) | Modular policy agent · Contextual bandit and block placing | Wasserstein distance to oracle marginals | Helped |
| use nonzero lambda versus lambda=0 (ablation) | Modular policy agent · Block placing adaptation to new partners | adaptation performance | Helped |
| modular task+partner representation versus BaselineAgg, BaselineModular and MAML | Adaptation to a new partner · Contextual bandit, block placing, Hanabi | adaptation speed | Helped |
| shared optimal action set between train and test contexts (Test-C vs Test-B) | Human pairs · Human-human bandit user study | first-try score: .50 -> 0.9545 | Helped |
| two good actions (Train-B, Train-C) instead of one (Train-A) | Human pairs · Human-human bandit user study | first-try score: 0.9545 -> .4091 | Hurt |
Closest papers by meaning
- Adaptively Coordinating with Novel Partners via Learned Latent Strategies (NeurIPS 2025 · similarity 0.87)
- Partner Modelling Emerges in Recurrent Agents (But Only When It Matters) (NeurIPS 2025 · similarity 0.87)
- Designs for Enabling Collaboration in Human-Machine Teaming via Interactive and Explainable Systems (NeurIPS 2024 · similarity 0.87)
- NestRL: A Nested Training Regime for Mutual Adaptation in Human–AI Teaming (NeurIPS 2026 · similarity 0.86)
- Adaptive Agents for Mixed-Initiative Human-AI Collaborations (AAAI 2025 · similarity 0.86)
- Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators (NeurIPS 2024 · similarity 0.86)
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.