featured

We show that empirical weight distributions of SGD-trained neural networks are heavy-tailed and correlated and that incorporating these insights into Bayesian neural network priors can improve their performance and reduce the cold-posterior effect.

We provide a comprehensive review of the recent advances regarding the choice of priors for Bayesian neural networks, variational autoencoders, and (deep) Gaussian processes.

We derive a novel PAC-Bayes bound for meta-learning with Bayesian models, which gives rise to a computationally efficient meta-learning method that outperforms existing approaches on a range of tasks, especially when the number of meta-tasks is small.

We show that introducing a repulsive force between the members of a deep ensemble can improve the ensemble’s diversity and performance, especially when this force is applied in the function space, and that it can also guarantee asymptotic convergence to the true Bayes posterior.

We show that a Laplace-Generalized-Gauss-Newton approximation to the marginal likelihood of Bayesian neural networks can effectively be used for model selection and can often discover better hyperparameter settings than cross-validation.

We show that using a Gaussian process prior in the latent space of a variational autoencoder can improve time series imputation performance, while still allowing for computationally efficient inference through a variational Gauss-Markov process.

We propose a novel version of the classical self-organizing map, which can be used as a structural prior in the latent space of a variational autoencoder, enabling interpretable representations of time series.