Bayesian Neural Networks with Weight Sharing Using Dirichlet Processes

• Published in: IEEE transactions on pattern analysis and machine intelligence Vol. 42; no. 1; pp. 246 - 252
• Main Authors: Roth, Wolfgang, Pernkopf, Franz
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Bayes methods; Bayesian neural networks; Computational modeling; Dirichlet problem; Dirichlet processes; Gibbs sampling; hybrid Monte-Carlo; Memory management; Monte Carlo methods; Neural networks; non-conjugate models; Task analysis; Weight; weight sharing
• ISSN: 0162-8828; 1939-3539; 2160-9292
• DOI: 10.1109/TPAMI.2018.2884905

We extend feed-forward neural networks with a Dirichlet process prior over the weight distribution. This enforces a sharing on the network weights, which can reduce the overall number of parameters drastically. We alternately sample from the posterior of the weights and the posterior of assignments of network connections to the weights. This results in a weight sharing that is adopted to the given data. In order to make the procedure feasible, we present several techniques to reduce the computational burden. Experiments show that our approach mostly outperforms models with random weight sharing. Our model is capable of reducing the memory footprint substantially while maintaining a good performance compared to neural networks without weight sharing.