CARLS: Cross-platform Asynchronous Representation Learning System

TL;DR

CARLS is a new asynchronous framework that enhances deep learning by enabling components like trainers and knowledge sources to collaborate across hardware platforms, improving learning paradigms such as semi-supervised and multimodal learning.

Contribution

It introduces a novel asynchronous system architecture for deep learning that integrates knowledge components across hardware platforms, supporting scalable learning paradigms.

Findings

Supports semi-supervised, curriculum, and multimodal learning.

Open-sourced implementation available.

Enhances deep learning scalability and flexibility.

Abstract

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