Odd-One-Out Representation Learning
This work provides a method for model selection and evaluation of representation learning for practitioners and researchers in scenarios where ground-truth factors of variation are expensive or difficult to obtain, particularly in perception.
This paper addresses the challenge of evaluating representation learning models without ground-truth labels by proposing a weakly-supervised odd-one-out task. They demonstrate that models performing well on this task also show high correlation with performance on a difficult abstract visual reasoning task, and their bespoke VAE model outperforms other unsupervised and weakly-supervised models across several metrics.
The effective application of representation learning to real-world problems requires both techniques for learning useful representations, and also robust ways to evaluate properties of representations. Recent work in disentangled representation learning has shown that unsupervised representation learning approaches rely on fully supervised disentanglement metrics, which assume access to labels for ground-truth factors of variation. In many real-world cases ground-truth factors are expensive to collect, or difficult to model, such as for perception. Here we empirically show that a weakly-supervised downstream task based on odd-one-out observations is suitable for model selection by observing high correlation on a difficult downstream abstract visual reasoning task. We also show that a bespoke metric-learning VAE model which performs highly on this task also out-performs other standard unsupervised and a weakly-supervised disentanglement model across several metrics.