JIVE: Jacobian-Informed Volume Expansion for Diverse Generative Sampling
Generative models often suffer from mode collapse and limited sample diversity. While prior works attempt to mitigate this by jointly generating a batch of samples and repelling their trajectories, these heuristics do not explicitly maximize the diversity of the resulting endpoints. We introduce JIVE, a training-free f...
Guang-Xun Zhang, Brian Cai, Boxuan Zhang et al.
· 0 citations