The volume transmission of small molecules implements a context-factorized hypernetwork capable of online credit assignment and compositional learning reminiscent of neurobiology.

With an intention to better exploit the task relationships for continual learning, we propose a transferability-based task embedding named H-embedding and present a hypernet under its guidance.

Hypernetwork-based framework for dynamic adaption of precomputed database features.

We propose a post-training technique, that uses a hypernetwork to efficiently steer diffusion model initial noise towards distributions favored by reward models, enhancing generation quality at low computational cost.