Weight Uncertainty in Neural Networks
This work addresses the need for principled uncertainty estimation in neural networks, which is incremental as it builds on existing variational methods but applies them to weight distributions for broader applications.
The paper tackled the problem of learning weight uncertainty in neural networks by introducing Bayes by Backprop, an algorithm that regularizes weights through variational free energy minimization, achieving performance comparable to dropout on MNIST classification and improving generalization in regression and reinforcement learning tasks.
We introduce a new, efficient, principled and backpropagation-compatible algorithm for learning a probability distribution on the weights of a neural network, called Bayes by Backprop. It regularises the weights by minimising a compression cost, known as the variational free energy or the expected lower bound on the marginal likelihood. We show that this principled kind of regularisation yields comparable performance to dropout on MNIST classification. We then demonstrate how the learnt uncertainty in the weights can be used to improve generalisation in non-linear regression problems, and how this weight uncertainty can be used to drive the exploration-exploitation trade-off in reinforcement learning.