Research map / Reinforcement learning / Imitation learning
Curriculum Design for Teaching via Demonstrations: Theory and Applications
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 the combined teacher- and learner-difficulty ranking (Cur) | MaxEnt-IRL and CrossEnt-BC learners in the car-driving environment · Car driving (T0–T7) | total expected reward convergence | Helped |
| Rank demonstrations using teacher difficulty alone (Cur-T), holding learner difficulty constant (ablation) | Cur teacher · Car driving (T0–T7) | total expected reward convergence | Hurt |
| Rank demonstrations using learner difficulty alone (Cur-L), holding teacher difficulty constant (ablation) | Cur teacher · Car driving (T0–T7) | total expected reward convergence | Hurt |
| Use random ordering (Agn) instead of the combined difficulty ranking | MaxEnt-IRL and CrossEnt-BC learners in the car-driving environment · Car driving (T0–T7) | total expected reward convergence | Hurt |
| Use batch set-cover teaching (Scot) instead of interactive curriculum selection | MaxEnt-IRL learner · Car driving (T0–T7) | total expected reward convergence | Hurt |
| Use the omniscient teacher (Omn), which selects demonstrations using learner gradients | MaxEnt-IRL learner · Car driving (T0–T7) | total expected reward convergence | Mixed |
| Use the blackbox teacher (BBox), which requires the true reward function | MaxEnt-IRL learner · Car driving (T0–T7) | total expected reward convergence | No clear effect |
| Estimate the learner policy with limited probing rather than observe it fully | Cur teacher with MaxEnt-IRL learner · Car driving (T0–T7), limited observability | total expected reward convergence | No clear effect |
| Probe every 40 time steps rather than every time step, with unlimited queries per probe | Cur teacher with MaxEnt-IRL learner · Car driving (T0–T7), limited observability | total expected reward convergence | Mixed |
| Use the combined difficulty ranking rather than uniform task selection | Neural CrossEnt-BC learner · Shortest-path navigation | test-set reward convergence | Helped |
| Use the combined difficulty ranking rather than uniform task selection | Neural CrossEnt-BC learner · Travelling-salesman navigation | test-set reward convergence | Helped |
| Use learner difficulty alone (Cur-L), omitting task-specific teacher difficulty (ablation) | Neural CrossEnt-BC learner · Shortest-path and travelling-salesman navigation | test-set reward convergence | No clear effect |
Closest papers by meaning
- Learner-aware Teaching: Inverse Reinforcement Learning with Preferences and Constraints (NeurIPS 2019 · similarity 0.89)
- Teaching Inverse Reinforcement Learners via Features and Demonstrations (NeurIPS 2018 · similarity 0.89)
- Diffusion-based Curriculum Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Unsupervised Training Sequence Design: Efficient and Generalizable Agent Training (AAAI 2024 · similarity 0.86)
- Learning from Demonstrations via Capability-Aware Goal Sampling (NeurIPS 2025 · similarity 0.86)
- Hybrid Inverse Reinforcement Learning (ICML 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.