arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1908.04339cs.LGcs.CVstat.ML
收藏

面向高效多任务体系结构的特征划分

Feature Partitioning for Efficient Multi-Task Architectures

Princeton University(普林斯顿大学);Google(谷歌);ByteDance(字节跳动);Stanford(斯坦福大学)

Alejandro Newell, Lu Jiang, Chong Wang, Li-Jia Li, Jia Deng

摘要:

多任务学习比单独模型的训练具有更少的数据,参数和时间。我们提出了一种方法来自动搜索多任务架构,同时考虑到资源的限制。我们提出了一个搜索空间,compensation表示不同的参数共享策略。这提供了对多任务架构的空间的更有效的覆盖和采样。我们还提出了一种方法,通过使用特征提取快速评估不同的体系结构。这些贡献使我们能够快速优化高效的多任务模型。我们以Visual Decathlon为基准,证明我们可以自动搜索和识别多任务架构,这些架构可以有效地在任务资源需求之间进行权衡,同时实现高水平的最终性能。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑