Meaning
A reinforcement learning algorithm that normalizes reward baselines across output candidate groups without requiring an independent critic network.
Articles Featuring This Term
Related Terms
- Trapdoor Function One-Way Mathematical Function
- Multi-Head Latent Attention Compressed Latent Attention Architecture
- MuJoCo Physics Engine Multi-Joint Dynamics Simulation
- Mel-Spectrogram
- Process Reward Model Step-by-Step Logic Evaluator
- Dataset Shift Hospital Generalization Drift
- Integer Factorization Prime Number Decomposition
- RingAttention Distributed Sequence Parallelism
- Text-to-Speech (TTS)