Neural Operator Variational Inference Based on Regularized Stein Discrepancy for Deep Gaussian Processes
Published in IEEE Transactions on Neural Networks and Learning Systems (TNNLS), 36(4), 2024
NOVI replaces the conventional mean-field Gaussian posterior over Deep GP inducing variables with a neural-network generator optimised by minimising a regularized Stein discrepancy rather than the KL term in the ELBO. The generator behaves as a neural operator mapping noise to inducing-variable samples, giving a substantially more expressive posterior while keeping inference tractable. Demonstrated on regression and image classification (CIFAR-10 conv-DGP).
Recommended citation: Xu, J., Du, S., Yang, J., Ma, Q. and Zeng, D. (2025). Neural Operator Variational Inference Based on Regularized Stein Discrepancy for Deep Gaussian Processes. IEEE Transactions on Neural Networks and Learning Systems, 36(4), pp.6723-6737.
Download Paper