WRSWorld Research StateContribute
WORLD RESEARCH STATE

Random Network Distillation: fixed-target prediction for exploration

Contribution grade

Reinforcement-learning exploration bonuses · Human-designed research method or resource

Reference score130,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Measured novelty by a predictor network's error on features from a fixed randomly initialized target network, combining intrinsic and extrinsic reward streams with flexible value estimation.

Scientific scope

Measured novelty by a predictor network's error on features from a fixed randomly initialized target network, combining intrinsic and extrinsic reward streams with flexible value estimation. Inherits novelty/prediction-error exploration and PPO. The deterministic target avoids stochastic forward-transition targets but is not a general proof against every noisy-observation distraction. Atari/Montezuma results have specified data and evaluation costs, not universal sample-efficient exploration; first two source authors have equal-contribution notation.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits novelty/prediction-error exploration and PPO. The deterministic target avoids stochastic forward-transition targets but is not a general proof against every noisy-observation distraction. Atari/Montezuma results have specified data and evaluation costs, not universal sample-efficient exploration; first two source authors have equal-contribution notation.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction