Causal Modeling of DeepMind D-Sprites Data with a Variational Autoencoder (Ansari et al)
Latent variable models are a broad class of machine learning algorithms that map observed variables back to a latent class class of variables, often of lower dimension.
This work features use of a VAE architecture to build a causal model of the d-sprites dataset from Deepmind. This was a class project by Farhanur Rahim Ansari, Gourang Patel, Sarang Pande, and Vidhey Oza, students in the Spring 2020 cohort of the causal machine learning course.
See the project code in https://www.github.com/robertness/cau...
Interested in learning how to build these kinds of models? Take the free course on causal generative machine learning at https://altdeep.ai