Why is Bayesian regularization backpropagation (Neural Network Toolbox) so very very slow?

Question

jkr 2016-8-28

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/301122-why-is-bayesian-regularization-backpropagation-neural-network-toolbox-so-very-very-slow

回答： Mustafa Sobhy 2019-8-22

Empirically I've found with a challenging pattern recognition problem I'm working on, that Bayesian regularization backpropagation (trainbr) outperforms more standard tools such as trainlm, trainscg and trainrp by quite a bit. But, it takes an extraordinarily longer time to compute. In its original formulation (MacKay 1992), Bayesian regularization required calculation of the Hessian matrix, which is very computationally demanding, and would account for the long time. However, in Foresee 1997 (both works cited in Matlab doc for trainbr), an alternative was developed that claims to reduce the computational challenge to be similar to e.g. trainlm. This latter work is cited in the documentation, but is it implemented? Can I find a library somewhere that implements it? I'm pretty confident that trainbr as implemented in the Neural Network Toolbox requires calculation of the Hessian, because it refuses to run on a GPU, identifying lack of support for inversion of the (related) Jacobian as the reason. But, I'd be happy to be educated on that.

2 个评论
显示无隐藏无

Andrew Diamond 2018-1-24

did you ever ping matlab support on this? As they say, "Inquiring minds want to know."

jkr 2018-1-25

No, I've moved on other issues. If you inquire, please somehow link to hear - I'd be happy to learn more in case I return to this problem.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Greg Heath 2016-8-28

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/301122-why-is-bayesian-regularization-backpropagation-neural-network-toolbox-so-very-very-slow#answer_232977

在 MATLAB Online 中打开

Use the command

type trainbr

Thank you for formally accepting my answer

Greg

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

jkr 2016-8-28

Interesting. I wasn't familiar with the 'type' command. But, the results don't address the question of whether the Gauss-Newton approximation has been implemented, and if not whether I can find an implementation of it.

请先登录，再进行评论。