Feature Partitioning for Efficient Multi-Task Architectures

Alejandro Newell, Lu Jiang, Chong Wang +2

Multi-task learning holds the promise of less data, parameters, and time than training of separate models. We propose a method to automatically search over multi-task architectures…