Search papers, labs, and topics across Lattice.
1
0
3
1
Achieving up to 6脳 greater sample efficiency in diffusion RLHF by strategically reweighting timesteps and reusing informative trajectories could revolutionize how we align generative models with human preferences.
