One Model To Learn Them All

TL;DR

This paper introduces a versatile single deep learning model capable of handling multiple diverse tasks across domains, reducing the need for task-specific architectures and tuning.

Contribution

The authors propose a unified model architecture combining various computational blocks, trained concurrently on multiple tasks, demonstrating broad applicability and improved performance through joint training.

Findings

Multi-task training benefits low-data tasks significantly.

Adding diverse blocks never harms and often improves performance.

The model performs well across vision, language, and speech tasks.

Abstract

Deep learning yields great results across many fields, from speech recognition, image classification, to translation. But for each problem, getting a deep model to work well involves research into the architecture and a long period of tuning. We present a single model that yields good results on a number of problems spanning multiple domains. In particular, this single model is trained concurrently on ImageNet, multiple translation tasks, image captioning (COCO dataset), a speech recognition corpus, and an English parsing task. Our model architecture incorporates building blocks from multiple domains. It contains convolutional layers, an attention mechanism, and sparsely-gated layers. Each of these computational blocks is crucial for a subset of the tasks we train on. Interestingly, even if a block is not crucial for a task, we observe that adding it never hurts performance and in most cases improves it on all tasks. We also show that tasks with less data benefit largely from joint training with other tasks, while performance on large tasks degrades only slightly if at all.