Research map / Multi-agent systems and games / Games and equilibria
Algorithmic Improvements for Deep Reinforcement Learning Applied to Interactive Fiction
Vishal Jain, William Fedus, Hugo Larochelle, Doina Precup, Marc G. Bellemare
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 |
|---|---|---|---|
| Add score contextualisation (separate value-function heads keyed by cumulative reward) | LSTM-DRQN · SaladWorld Levels 1–3 | fraction of subtasks completed | Improved |
| Add score contextualisation and masking based on predicted admissibility | LSTM-DRQN · SaladWorld Level 3 | fraction of subtasks completed | Improved |
| Add score contextualisation and masking based on predicted admissibility | LSTM-DRQN · SaladWorld Levels 4–7 | learning speed | Improved |
| Use action gating without score contextualisation (ablation) | LSTM-DRQN · SaladWorld Levels 2–7 | fraction of subtasks completed | No clear effect |
| Use masking rather than dropout or CQLH for action gating (ablation) | Score-contextualised LSTM-DRQN · SaladWorld Level 3 | fraction of subtasks completed | Improved |
| Replace learned admissibility estimates with the oracle admissible-action set (ablation) | Score-contextualised LSTM-DRQN · SaladWorld | learning speed | Improved |
| Use the score-contextualising agent without information-gathering actions | Score-contextualised LSTM-DRQN · Zork I | score | Improved |
| Remove prioritized sampling (ablation) | LSTM-DRQN · Zork I | score | Hurt |
| Remove the infrequent look action (ablation) | LSTM-DRQN · Zork I | score | Hurt |
| Replace CQLH with alternate CQLH, which omits the unchanged-state update term (ablation) | CQLH agent · Zork I | score | Hurt |
| Use one value-function head instead of score-specific heads (ablation) | Score-contextualised LSTM-DRQN · SaladWorld | learning speed | Hurt |
Closest papers by meaning
- Learning to Follow Instructions in Text-Based Games (NeurIPS 2022 · similarity 0.89)
- Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs (NeurIPS 2025 · similarity 0.86)
- TALES: Text-Adventure Learning Environment Suite (NeurIPS 2026 · similarity 0.86)
- Monte Carlo Planning with Large Language Model for Text-Based Game Agents (ICLR 2025 · similarity 0.86)
- Motif: Intrinsic Motivation from Artificial Intelligence Feedback (ICLR 2024 · similarity 0.85)
- AgentOdyssey: Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents (NeurIPS 2026 · 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.
