Theory and Evaluation Metrics for Learning Disentangled Representations
This work addresses a foundational problem in machine learning for researchers by providing theoretical clarity and evaluation tools for disentanglement learning, though it is incremental in refining existing concepts.
The paper tackles the lack of precise definitions and robust evaluation metrics for disentangled representations by defining semantics along informativeness, separability, and interpretability dimensions and proposing corresponding metrics. The result shows that these metrics correctly characterize representations and enable fair comparisons, with empirical insights into VAE-based methods.
We make two theoretical contributions to disentanglement learning by (a) defining precise semantics of disentangled representations, and (b) establishing robust metrics for evaluation. First, we characterize the concept "disentangled representations" used in supervised and unsupervised methods along three dimensions-informativeness, separability and interpretability - which can be expressed and quantified explicitly using information-theoretic constructs. This helps explain the behaviors of several well-known disentanglement learning models. We then propose robust metrics for measuring informativeness, separability and interpretability. Through a comprehensive suite of experiments, we show that our metrics correctly characterize the representations learned by different methods and are consistent with qualitative (visual) results. Thus, the metrics allow disentanglement learning methods to be compared on a fair ground. We also empirically uncovered new interesting properties of VAE-based methods and interpreted them with our formulation. These findings are promising and hopefully will encourage the design of more theoretically driven models for learning disentangled representations.