Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Deep Learning: A Bayesian Perspective
(Submitted on 1 Jun 2017 (v1), revised 28 Aug 2017 (this version, v2), latest version 14 Nov 2017 (v4))
Abstract: Deep learning is a form of machine learning for nonlinear high dimensional pattern matching and prediction. By taking a Bayesian probabilistic perspective, we provide a number of advantages, with more efficient algorithms for optimisation and hyper-parameter tuning, and an explanation of predictive performance. A framework for constructing good Bayesian predictors in high dimensions is provided. Traditional high-dimensional data reduction techniques; principal component analysis (PCA), partial least squares (PLS), reduced rank regression (RRR), projection pursuit regression (PPR) are shown to be shallow learners. Their deep learning counterparts exploit multiple layers of data reduction which leads to performance gains. Stochastic gradient descent (SGD) training, and optimisation and Dropout (DO) provide model and variable selection. Bayesian regularization is central to finding networks and optimizing the bias-variance trade-off, to achieve good out-of sample performance. To illustrate our methodology, we provide an analysis of first time international bookings on Airbnb. Finally, we conclude with directions for future research.
Submission history
From: Vadim Sokolov [view email][v1] Thu, 1 Jun 2017 19:50:37 GMT (5073kb,D)
[v2] Mon, 28 Aug 2017 00:57:42 GMT (5309kb,D)
[v3] Tue, 5 Sep 2017 01:33:09 GMT (5309kb,D)
[v4] Tue, 14 Nov 2017 03:36:51 GMT (5370kb,D)
Link back to: arXiv, form interface, contact.