Local Expectation Gradients for Black Box Variational Inference

By Michalis K. Titsias, Miguel L'azaro-Gredilla. 2015

90 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Variational Inference: A Review for Statisticians (2016, neutral): “There has also been a flurry of recent work on methods that optimize the variational objective with Monte Carlo estimates of the gradient (Paisley et al., 2012; Kingma and Welling, 2013; Rezende et al., 2014; Ranganath et al., 2014; Salimans and Knowles, 2014; Titsias and Lázaro-Gredilla, 2014; Tits”
The Generalized Reparameterization Gradient (2016, background): “, through local expectations (Titsias and Lázaro-Gredilla, 2015) and importance sampling (Ruiz et al.”
Gradient Estimation with Stochastic Softmax Tricks (2020, background): “[9] Michalis Titsias RC AUEB and Miguel Lázaro-Gredilla.”
Gaussian variational approximation with sparse precision matrices (2016, methodology): “…gradient methods (Robbins and Monro, 1951) able to optimize L(λ) efficiently in this situation (Ji et al., 2010; Nott et al., 2012; Paisley et al., 2012; Kingma and Welling, 2014; Hoffman et al., 2013; Ranganath et al., 2014; Rezende et al., 2014; Titsias and Lázaro-Gredilla, 2014, 2015).”
Deep Unsupervised Clustering with Gaussian Mixture Variational Autoencoders (2016, methodology): “Several possibilities for estimating the gradient of discrete variables have been proposed (Glynn, 1990; Titsias & L´azaro-Gredilla, 2015).”
Advances in Variational Inference (2017, methodology): “While the original BBVI paper introduces both Rao-Blackwellization and control variates, [194] points out that good choices for control variates might be modeldependent.”
DVAE#: Discrete Variational Autoencoders with Relaxed Boltzmann Priors (2018, background): “ grouped into ﬁve main categories: i)Exhaustive approaches marginalize all discrete variables [25,26] and which are not scalable to more than a few discrete variables. ii) Local expectation gradients [27] and reparameterization and marginalization [28] estimators compute low-variance estimates at th”
Monte Carlo Gradient Estimation in Machine Learning (2019, methodology): “…is essential for effective use of this estimator, and has been explored in several areas, by Greensmith et al. (2004) in reinforcement learning, Titsias and Lázaro-Gredilla (2015) in variational inference, Capriotti (2008) in computational finance, or more recently by Walder et al. (2019).”
The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables (2016, methodology): “For models with discrete latent variables, the discontinuous nature of which makes reparameterization not useful, a number of score function estimators have been developed [31, 17, 33, 28, 42, 18], which differ primarily in the variance reduction techniques used.”