Feature Partitioning for Efficient Multi-Task Architectures
This work addresses the challenge of resource-efficient multi-task learning for AI practitioners, though it appears incremental as it builds on existing search and distillation techniques.
The paper tackles the problem of efficiently designing multi-task architectures under resource constraints by proposing an automated search method with a compact representation of parameter sharing strategies and feature distillation for quick evaluation, resulting in architectures that effectively trade off task resource requirements while achieving high performance on the Visual Decathlon benchmark.
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 while taking resource constraints into consideration. We propose a search space that compactly represents different parameter sharing strategies. This provides more effective coverage and sampling of the space of multi-task architectures. We also present a method for quick evaluation of different architectures by using feature distillation. Together these contributions allow us to quickly optimize for efficient multi-task models. We benchmark on Visual Decathlon, demonstrating that we can automatically search for and identify multi-task architectures that effectively make trade-offs between task resource requirements while achieving a high level of final performance.