Realistic Evaluation of Deep Semi-Supervised Learning Algorithms
This work addresses the gap between benchmark performance and real-world applicability for SSL algorithms, which is crucial for researchers and practitioners aiming to deploy these methods effectively.
The paper tackled the problem that standard benchmarks for deep semi-supervised learning (SSL) algorithms do not reflect real-world challenges, finding that simple baselines often outperform reported results, SSL methods vary in sensitivity to data amounts, and performance degrades with out-of-class examples in unlabeled data.
Semi-supervised learning (SSL) provides a powerful framework for leveraging unlabeled data when labels are limited or expensive to obtain. SSL algorithms based on deep neural networks have recently proven successful on standard benchmark tasks. However, we argue that these benchmarks fail to address many issues that these algorithms would face in real-world applications. After creating a unified reimplementation of various widely-used SSL techniques, we test them in a suite of experiments designed to address these issues. We find that the performance of simple baselines which do not use unlabeled data is often underreported, that SSL methods differ in sensitivity to the amount of labeled and unlabeled data, and that performance can degrade substantially when the unlabeled dataset contains out-of-class examples. To help guide SSL research towards real-world applicability, we make our unified reimplemention and evaluation platform publicly available.