Transforming Vision Transformer: Towards Efficient Multi-Task Asynchronous Learning

TL;DR

This paper introduces EMTAL, a novel method for multi-task learning with Vision Transformers that improves efficiency and performance by transforming pre-trained models and optimizing asynchronous learning processes.

Contribution

The paper proposes a new framework called EMTAL that transforms pre-trained Vision Transformers into efficient multi-task learners with reparameterization and asynchronous optimization.

Findings

Outperforms state-of-the-art multi-task learning methods on benchmarks.

Achieves higher inference speed without sacrificing accuracy.

Effectively maintains task performance during asynchronous training.

Abstract

Multi-Task Learning (MTL) for Vision Transformer aims at enhancing the model capability by tackling multiple tasks simultaneously. Most recent works have predominantly focused on designing Mixture-of-Experts (MoE) structures and in tegrating Low-Rank Adaptation (LoRA) to efficiently perform multi-task learning. However, their rigid combination hampers both the optimization of MoE and the ef fectiveness of reparameterization of LoRA, leading to sub-optimal performance and low inference speed. In this work, we propose a novel approach dubbed Efficient Multi-Task Learning (EMTAL) by transforming a pre-trained Vision Transformer into an efficient multi-task learner during training, and reparameterizing the learned structure for efficient inference. Specifically, we firstly develop the MoEfied LoRA structure, which decomposes the pre-trained Transformer into a low-rank MoE structure and employ LoRA to fine-tune the parameters. Subsequently, we take into account the intrinsic asynchronous nature of multi-task learning and devise a learning Quality Retaining (QR) optimization mechanism, by leveraging the historical high-quality class logits to prevent a well-trained task from performance degradation. Finally, we design a router fading strategy to integrate the learned parameters into the original Transformer, archiving efficient inference. Extensive experiments on public benchmarks demonstrate the superiority of our method, compared to the state-of-the-art multi-task learning approaches.