Deep Belief Network on CIFAR-10
€30-250 EUR
Pagado a la entrega
The database of CIFAR-10 can be found on this page (CIFAR-10 Matlab version) : [login to view URL]~kriz/[login to view URL]
The algorithm theory is based on MSc thesis of Alex Krizhevsky (Learning Multiple Layers of Features from Tiny Images).
The paper can be found in the bottom of the same page ([login to view URL]~kriz/[login to view URL])
1) The developer must first download the database
2) Implement ZCA whitenning transformation to the data
3) Create a 2-layer (2 is the number of stacked Gaussian - Bernoulli Boltzman machines).
4) Train the Deep Belief Network using greedy layer-wise method
5) Train each Boltzman machine with CD (contrastive divergence) method
6) Extract an error on train images
* Extract some images of the filters from 1-layer
* Extract some images of the filters from 2-layer
The algorithm must be as simple and readable as possible and well commented in Matlab, impmlemented for ECU not GPU.
All informations are in the paper of Alex Krizhevsky.
Nº del proyecto: #8920255
Sobre el proyecto
5 freelancers están ofertando un promedio de €636 por este trabajo
Dear Madam or Sir, I have great experience with machine learning Python, C++ and Matlab. Moreover, as a mathematician, I am familiar with both the theoretical and the applied part of machine learning, especially de Más
Greetings! After going through your project description, I am excited to work on this project. I am experienced freelancer and fully capable to produce TOP notch work according to your requirements. Contact me to discu Más