Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces Universal Hypernetworks (UHNs), a fixed-architecture generator that predicts weights for arbitrary models based on parameter, architecture, and task descriptors. By decoupling the generator from the target network parameterization, a single UHN can instantiate diverse models across vision, graph, text, and formula-regression tasks while remaining competitive with directly trained models. The authors demonstrate multi-model generalization, multi-task learning, and stable recursive generation using UHNs.
Forget retraining hypernetworks for every new architecture - a single Universal Hypernetwork can now generate weights for diverse models across vision, graphs, text, and regression tasks.
Conventional hypernetworks are typically engineered around a specific base-model parameterization, so changing the target architecture often entails redesigning the hypernetwork and retraining it from scratch. We introduce the \emph{Universal Hypernetwork} (UHN), a fixed-architecture generator that predicts weights from deterministic parameter, architecture, and task descriptors. This descriptor-based formulation decouples the generator architecture from target-network parameterization, so one generator can instantiate heterogeneous models across the tested architecture and task families. Our empirical claims are threefold: (1) one fixed UHN remains competitive with direct training across vision, graph, text, and formula-regression benchmarks; (2) the same UHN supports both multi-model generalization within a family and multi-task learning across heterogeneous models; and (3) UHN enables stable recursive generation with up to three intermediate generated UHNs before the final base model. Our code is available at https://github.com/Xuanfeng-Zhou/UHN.