Search papers, labs, and topics across Lattice.
Tsinghua University, Carnegie Mellon University, UC Berkeley, Harvard University
Tsinghua AI2
0
4
Online imitation learning can outperform offline methods, but only when the student can effectively represent the expert鈥攔ealizability is key.
Sampling from diffusion models with quadratic rewards can be surprisingly hard: negative-definite tilts are intractable even in simple cases, while a new algorithm makes low-rank positive-definite tilts tractable.
