PPL
Taiji Suzuki
Papers (4)
How Neural Reward Models Learn Features for Policy Optimization: A Single-Index AnalysisProvably Learning Diffusion Models under the Manifold Hypothesis: Collapse and RefineIntrinsic Wasserstein Rates for Score-Based Generative Models on Smooth ManifoldsThe Mechanism of Weak-to-Strong Generalization: Feature Elicitation from Latent Knowledge