%0 Journal Article %1 lin2019simple %A Lin, Wu %A Khan, Mohammad Emtiyaz %A Schmidt, Mark %D 2019 %K deep-learning flows optimization readings stats theory variational %T Fast and Simple Natural-Gradient Variational Inference with Mixture of Exponential-family Approximations %U http://arxiv.org/abs/1906.02914 %X Natural-gradient methods enable fast and simple algorithms for variational inference, but due to computational difficulties, their use is mostly limited to minimal exponential-family (EF) approximations. In this paper, we extend their application to estimate structured approximations such as mixtures of EF distributions. Such approximations can fit complex, multimodal posterior distributions and are generally more accurate than unimodal EF approximations. By using a minimal conditional-EF representation of such approximations, we derive simple natural-gradient updates. Our empirical results demonstrate a faster convergence of our natural-gradient method compared to black-box gradient-based methods. Our work expands the scope of natural gradients for Bayesian inference and makes them more widely applicable than before.

@article{lin2019simple, abstract = {Natural-gradient methods enable fast and simple algorithms for variational inference, but due to computational difficulties, their use is mostly limited to \emph{minimal} exponential-family (EF) approximations. In this paper, we extend their application to estimate \emph{structured} approximations such as mixtures of EF distributions. Such approximations can fit complex, multimodal posterior distributions and are generally more accurate than unimodal EF approximations. By using a \emph{minimal conditional-EF} representation of such approximations, we derive simple natural-gradient updates. Our empirical results demonstrate a faster convergence of our natural-gradient method compared to black-box gradient-based methods. Our work expands the scope of natural gradients for Bayesian inference and makes them more widely applicable than before.}, added-at = {2019-06-11T05:05:39.000+0200}, author = {Lin, Wu and Khan, Mohammad Emtiyaz and Schmidt, Mark}, biburl = {https://www.bibsonomy.org/bibtex/20f6b24dd16aef0b2870c9f7755560a5d/kirk86}, description = {[1906.02914] Fast and Simple Natural-Gradient Variational Inference with Mixture of Exponential-family Approximations}, interhash = {c0cf51bdf10e4afbc4e5c61d73d8d368}, intrahash = {0f6b24dd16aef0b2870c9f7755560a5d}, keywords = {deep-learning flows optimization readings stats theory variational}, note = {cite arxiv:1906.02914Comment: Accepted as a conference paper at ICML 2019}, timestamp = {2020-02-24T04:31:20.000+0100}, title = {Fast and Simple Natural-Gradient Variational Inference with Mixture of Exponential-family Approximations}, url = {http://arxiv.org/abs/1906.02914}, year = 2019 }