Good Initializations of Variational Bayes for Deep Models
This addresses a specific bottleneck in Bayesian deep learning for researchers and practitioners, but it is incremental as it builds on established variational inference techniques.
The paper tackles the problem of poor initialization for stochastic variational inference in deep models by proposing a novel layer-wise initialization strategy based on Bayesian linear models. The result is faster and better convergence on regression and classification tasks, including Bayesian DeepNets and ConvNets, compared to existing methods.
Stochastic variational inference is an established way to carry out approximate Bayesian inference for deep models. While there have been effective proposals for good initializations for loss minimization in deep learning, far less attention has been devoted to the issue of initialization of stochastic variational inference. We address this by proposing a novel layer-wise initialization strategy based on Bayesian linear models. The proposed method is extensively validated on regression and classification tasks, including Bayesian DeepNets and ConvNets, showing faster and better convergence compared to alternatives inspired by the literature on initializations for loss minimization.