Quality at the Tail of Machine Learning Inference

TL;DR

This paper introduces the concept of 'tail quality' to evaluate fluctuations in deep learning inference quality under time constraints, proposing a framework to predict quality variations in safety-critical applications.

Contribution

It uncovers the phenomenon of inference quality fluctuations due to inference time and proposes a new evaluation framework to analyze and predict these fluctuations.

Findings

Inference quality fluctuates with inference time.

The proposed framework effectively predicts quality distribution.

Experiments validate the framework across multiple models and systems.

Abstract

Machine learning inference should be subject to stringent inference time constraints while ensuring high inference quality, especially in safety-critical (e.g., autonomous driving) and mission-critical (e.g., emotion recognition) contexts. Neglecting either aspect can lead to severe consequences, such as loss of life and property damage. Many studies lack a comprehensive consideration of these metrics, leading to incomplete or misleading evaluations. The study unveils a counterintuitive revelation: deep learning inference quality exhibits fluctuations due to inference time. To depict this phenomenon, the authors coin a new term, "tail quality," providing a more comprehensive evaluation, and overcoming conventional metric limitations. Moreover, the research proposes an initial evaluation framework to analyze factors affecting quality fluctuations, facilitating the prediction of the potential distribution of inference quality. The effectiveness of the evaluation framework is validated through experiments conducted on deep learning models for three different tasks across four systems.