Discrete Latent

Discrete latent variable models aim to learn underlying, unobserved categorical structures within data, improving model interpretability and performance. Current research focuses on developing efficient algorithms, such as GFlowNets and joint stochastic approximation, to handle the computational challenges posed by the combinatorial nature of discrete variables, and integrating these variables into various architectures including diffusion models, GANs, and variational autoencoders. This work has significant implications across diverse fields, enhancing model robustness in applications like medical image analysis and improving the accuracy and interpretability of predictions in natural language processing and causal inference.

Papers