Leno

New LeNO method optimizes generative models for reward alignment

Researchers have developed LeNO (Learning Noise Optimization), a novel method for improving reward alignment in few-step generative models. This approach amortizes reward-guided noise optimization into a learned residua…