Single-Layer Vision Transformers for More Accurate Early Exits with Less Overhead

TL;DR

This paper presents a new single-layer vision transformer architecture with a fine-tuning strategy for early exiting, achieving higher accuracy with less overhead in resource-constrained, time-critical applications across various modalities.

Contribution

It introduces a novel vision transformer-based early exit architecture and a fine-tuning method that improves accuracy and reduces overhead, applicable to multiple data modalities and tasks.

Findings

Enhanced early exit accuracy with less computational overhead

Effective across image, audio, and audiovisual tasks

Supports both classification and regression problems

Abstract

Deploying deep learning models in time-critical applications with limited computational resources, for instance in edge computing systems and IoT networks, is a challenging task that often relies on dynamic inference methods such as early exiting. In this paper, we introduce a novel architecture for early exiting based on the vision transformer architecture, as well as a fine-tuning strategy that significantly increase the accuracy of early exit branches compared to conventional approaches while introducing less overhead. Through extensive experiments on image and audio classification as well as audiovisual crowd counting, we show that our method works for both classification and regression problems, and in both single- and multi-modal settings. Additionally, we introduce a novel method for integrating audio and visual modalities within early exits in audiovisual data analysis, that can lead to a more fine-grained dynamic inference.