Are VAEs considered explainable AI? To me, they are because the latent variables are interpretable, e.g, you change one and you might see its effects on the head rotation (for a dataset of faces, for example), unlike GANs or even supervised learning.
$\begingroup$ $\endgroup$
1 
- 3$\begingroup$ you can definitely do it with GANs, and same with supervised learning. you can always observe "what changes in the output if i change a latent representation". the point is, you cannot know why changing that leads the model to change certain characteristics, thus they are not interpretable. $\endgroup$Alberto– Alberto2025-06-06 13:22:26 +00:00Commented Jun 6 at 13:22
Add a comment |
1 Answer
$\begingroup$ $\endgroup$
Yes. The relevant topic is disentangled representation learning (DRL), which is considered as a part of explainable AI because the goal of DRL is to learn a set of explainable latent representations from data.
But there is no reason why GAN-based models cannot be considered as explainable AI. Similar to VAEs, you can inspect the the evolution of generated images by traversing different directions in the latent space e.g. InfoGAN. The method you mentioned is not unique to VAEs.
Reference:
Locatello, F., Bauer, S., Lucic, M., Raetsch, G., Gelly, S., Schölkopf, B. & Bachem, O.. (2019). Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:4114-4124 Available from https://proceedings.mlr.press/v97/locatello19a.html.
Chen, X., Duan, Y., Houthooft, R., Schulman, J., Sutskever, I., & Abbeel, P. (2016). InfoGAN: interpretable representation learning by information maximizing Generative Adversarial Nets. Advances in Neural Information Processing Systems. Presented at the Neural Information Processing Systems (NIPS), Barcelona, Spain.
