Abstract
The analytical method to capture the dropout induced distribution of forwarding output in a neural network as Gaussian mixture model (GMM) was proposed. In dropout Bayesian DNN, if the network is dropout-trained and a test data is dropout-forwarded for inference, then its output, usually approximated as a single mode Gaussian, becomes a posterior whose variance tells uncertainty of its inference [1]. Here, the proposed method can capture the arbitrary distribution analytically with high accuracy without Monte Carlo (MC) method for any network equipped with dropout and fully connected (FC) layers. Therefore, it is applicable to the general non-Gaussian posterior case for a better uncertainty estimate. The proposed method also has the advantage to provide a multimodal analysis in distribution by factoring which can be tuned with a user defined expressibility parameter while a MC estimate provides only a “flat” image. This helps to understand how the FC layer tries to code a dropout injected highly multimodal data into a single mode Gaussian while the unknown data becomes a complicated distribution.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Gal, Y.: Uncertainty in Deep Learning. PhD thesis, University of Cambridge (2016)
Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.: Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res. 15(1), 1929–1958 (2014)
Gal, Y., Ghahramani, Z.: Dropout as a Bayesian approximation: Representing model uncertainty in deep learning. In: Balcan, M.F., Kilian Q.W. (eds), Proceedings of The 33rd International Conference on Machine Learning. Proceedings of Machine Learning Research, vol. 48, pp. 1050–1059, New York, USA, 20–22 Jun 2016. PMLR
Leibig, C., Allken, V., Berens, P., Wahl, S.: Leveraging uncertainty information from deep neural networks for disease detection. bioRxiv (2016)
Louizos, C., Welling, M.: Multiplicative normalizing flows for variational Bayesian neural networks. In: Proceedings of the 34th International Conference on Machine Learning, ICML 2017, vol. 70, pp. 2218–2227. JMLR.org (2017)
Wang, S.I., Manning, C.D.: Fast dropout training. In: Proceedings of the 30th International Conference on International Conference on Machine Learning, ICML 2013, vol. 28, pp. II-118-II-126. JMLR.org (2013)
Tahir, M.H., Ghazali, S.S.A., Gilani, G.M.: On the variance of the sample mean from finite population, approach iii (2005)
Wikipedia. Rectified Gaussian distribution – Wikipedia, the free encyclopedia. https://en.wikipedia.org/wiki/Rectified_Gaussian_distribution. Accessed 01 Jul 2019
Manjunath, B.G., Wilhelm, S.: Moments calculation for the double truncated multivariate normal density. SSRN Electron. J. (2009)
Krizhevsky, A., Sutskever, I., Hinton, G.E.: Imagenet classification with deep convolutional neural networks. In Proceedings of the 25th International Conference on Neural Information Processing Systems, NIPS 2012, vol. 1, pp. 1097–1105, USA. Curran Associates Inc. (2012)
ImageNet. http://www.image-net.org/
BVLC caffe AlexNet. https://github.com/BVLC/caffe/tree/master/models/bvlc_reference_caffenet
THE MNIST DATABASE. http://yann.lecun.com/exdb/mnist/
NotMNIST Dataset. https://www.kaggle.com/lubaroli/notmnist/
Hinton, G.E., Salakhutdinov, R.R.: Reducing the dimensionality of data with neural networks. Science 313, 504–507 (2006)
Hershey, J.R., Olsen, P.A.: Approximating the kullback leibler divergence between Gaussian mixture models. In: 2007 IEEE International Conference on Acoustics, Speech and Signal Processing, April 2007, ICASSP 2007. IEEE (2007)
Daunizeau, J.: Semi-analytical approximations to statistical moments of sigmoid and softmax mappings of normal variables (2017)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Switzerland AG
About this paper
Cite this paper
Adachi, J. (2019). Estimating and Factoring the Dropout Induced Distribution with Gaussian Mixture Model. In: Tetko, I., Kůrková, V., Karpov, P., Theis, F. (eds) Artificial Neural Networks and Machine Learning – ICANN 2019: Theoretical Neural Computation. ICANN 2019. Lecture Notes in Computer Science(), vol 11727. Springer, Cham. https://doi.org/10.1007/978-3-030-30487-4_60
Download citation
DOI: https://doi.org/10.1007/978-3-030-30487-4_60
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-30486-7
Online ISBN: 978-3-030-30487-4
eBook Packages: Computer ScienceComputer Science (R0)