In this work, we propose CARLS, a novel framework for augmenting the capacity of existing deep learning frameworks by enabling multiple components -- model trainers, knowledge makers and knowledge banks -- to concertedly work together in an asynchronous fashion across hardware platforms. The proposed CARLS is particularly suitable for learning paradigms where model training benefits from additional knowledge inferred or discovered during training, such as node embeddings for graph neural networks or reliable pseudo labels from model predictions. We also describe three learning paradigms -- semi-supervised learning, curriculum learning and multimodal learning -- as examples that can be scaled up efficiently by CARLS. One version of CARLS has been open-sourced and available for download at: https://github.com/tensorflow/neural-structured-learning/tree/master/research/carls
@article{arxiv.2105.12849,
title = {CARLS: Cross-platform Asynchronous Representation Learning System},
author = {Chun-Ta Lu and Yun Zeng and Da-Cheng Juan and Yicheng Fan and Zhe Li and Jan Dlabal and Yi-Ting Chen and Arjun Gopalan and Allan Heydon and Chun-Sung Ferng and Reah Miyara and Ariel Fuxman and Futang Peng and Zhen Li and Tom Duerig and Andrew Tomkins},
journal= {arXiv preprint arXiv:2105.12849},
year = {2021}
}