Modern deep learning systems are increasingly deployed in situations such as personalization and federated learning where it is necessary to support i) learning on small amounts of data, and ii) communication efficient distributed training protocols. In this work, we develop FiLM Transfer (FiT) which fulfills these requirements in the image classification setting by combining ideas from transfer learning (fixed pretrained backbones and fine-tuned FiLM adapter layers) and meta-learning (automatically configured Naive Bayes classifiers and episodic training) to yield parameter efficient models with superior classification accuracy at low-shot. The resulting parameter efficiency is key for enabling few-shot learning, inexpensive model updates for personalization, and communication efficient federated learning. We experiment with FiT on a wide range of downstream datasets and show that it achieves better classification accuracy than the leading Big Transfer (BiT) algorithm at low-shot and achieves state-of-the art accuracy on the challenging VTAB-1k benchmark, with fewer than 1% of the updateable parameters. Finally, we demonstrate the parameter efficiency and superior accuracy of FiT in distributed low-shot applications including model personalization and federated learning where model update size is an important performance metric.

翻译：在个人化和联合学习等必要情况下,越来越多地部署现代深层学习系统,例如个人化和联合学习,以便支持(一) 学习少量数据,(二) 通信高效分布式培训协议。在这项工作中,我们开发FILM传输系统(FIT),以满足图像分类设置中的这些要求,将转移学习(固定的预先训练的脊柱和微调的FILM适配层)和元学习(自动配置的Naive Bayes分类器和偶发培训)的理念结合起来,产生低发的高级分类精确度的参数高效模型。由此产生的参数效率是使少发的学习、廉价的个人化模型更新和通信高效的联邦化学习成为可能的关键。我们与FIT在广泛的下游数据集上进行实验,并表明它比主要大转移(BIT)低发的低发算法更加精确,并达到具有挑战性的VTAB-1k基准的艺术精确度,更新参数不到1%。最后,我们展示了FIT在分布的低发的低发式应用中的参数效率和高精确度精确度,包括模型的个人化和联合学习尺度。