论文标题
自动搜索资源有效的分支多任务网络
Automated Search for Resource-Efficient Branched Multi-Task Networks
论文作者
论文摘要
许多视力问题的多模式性质要求神经网络体系结构可以同时执行多个任务。通常，这种体系结构是在文献中手工制作的。但是，鉴于问题的大小和复杂性，这种手动体系结构探索可能超出了人类的设计能力。在本文中，我们提出了一种植根于可区分的神经体系结构搜索的原则方法，以在多任务神经网络的编码阶段自动定义分支（树状）结构。为了允许在资源受限的环境中灵活性，我们引入了一个无靠近的资源感知损失，该损失动态控制模型大小。各种密集预测任务的评估表明，我们的方法始终在有限的资源预算中找到高性能的分支结构。
The multi-modal nature of many vision problems calls for neural network architectures that can perform multiple tasks concurrently. Typically, such architectures have been handcrafted in the literature. However, given the size and complexity of the problem, this manual architecture exploration likely exceeds human design abilities. In this paper, we propose a principled approach, rooted in differentiable neural architecture search, to automatically define branching (tree-like) structures in the encoding stage of a multi-task neural network. To allow flexibility within resource-constrained environments, we introduce a proxyless, resource-aware loss that dynamically controls the model size. Evaluations across a variety of dense prediction tasks show that our approach consistently finds high-performing branching structures within limited resource budgets.
