Research map / Reinforcement learning / RL from feedback
Mega-Reward: Achieving Human-Level Play without Extrinsic Rewards
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 mega-reward instead of the strongest listed intrinsic-reward baseline | PPO · Atari intrinsically-motivated play (18 games) | score | Mixed |
| add latent control to direct-control reward (ablation) | Mega-reward · Atari games (overall) | relative ablation score: 34.82% overall drop without latent control | Helped |
| use the proposed mega-reward computation instead of count-based novelty on the control map (ablation) | Mega-reward · Atari games (overall) | relative ablation score: 55.48% overall drop with count-based novelty | Helped |
| use a 4 × 4 grid (ablation) | Mega-reward · Breakout | efficiency | Helped |
| mask RND observations with the accumulated latent-control map | RND · 6 hard-exploration Atari games, STD 0.3 | score | Mixed |
| pretrain with mega-reward rather than world-model, MOREL, or OOMDP pretraining | PPO · Atari games (18 games) | relative improvement: 14, 15, and 17 out of 18 games | Helped |
Closest papers by meaning
- Self-Rewarding Language Models (ICML 2024 · similarity 0.85)
- Motif: Intrinsic Motivation from Artificial Intelligence Feedback (ICLR 2024 · similarity 0.84)
- Potential-Based Reward Shaping for Intrinsic Motivation (Student Abstract) (AAAI 2024 · similarity 0.84)
- Eureka: Human-Level Reward Design via Coding Large Language Models (ICLR 2024 · similarity 0.84)
- Auto MC-Reward: Automated Dense Reward Design with Large Language Models for Minecraft (CVPR 2024 · similarity 0.84)
- Human-Aligned Chess With a Bit of Search (ICLR 2025 · similarity 0.83)
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.