[mlpack] GSOC2020 ProPosal: Implementing Essential Deep Learning Modules(Marcus)

[Marcus Edel]

Hello Himansu,

thanks for reaching out and welcome,

> 1. First one Is trying to RBF as a layer in mlpack I am right now I am working on this one. So, in this Implementation, I am trying to add a layer for Radial Basis Function which can be used with a feed-forward network and help us to construct a Radial Basis Function Networks.
> To test this network I will try to use the same tests which are being used to test vanilla network with dropout layer.

Sounds good, I see you already opened a WIP PR.

> 2. Second thing is to add RBF kernel right now we have only linear kernel SVM in mlpack I want to add support for RBF kernel as well. So, to do this I have two ideas in my mind 
> a) first one to add RBF kernel inside linear_svm directory and change its name to SVM  just like we are doing with GANS
> b) the second one will be to implement it in a separate place and adding everything separately
> Please suggest What do you think? for this one.

I wouldn't put it into the linear_svm folder, so personally I would go with option b.

> 3 Last but not least is to implement a DBN (Deep Belief Nets). We already have an implementation for RBM(Restricted Boltzmann Machine) in mlpack. So, my task to implement this is quite simple I will try to change that implementation So that I can use it as a layer. I need to implement some visitor for this and create an implementation for DBN to train these (RBM) layers.

Sounds good to me as well, always nice to reuse existing code.

> It is a very short explanation of these problems I will try to elaborate on them in my proposal.

The general plan sounds good to me, if you write the application make sure to
discuss the testing part as well, you already mentioned you like to reuse tests,
which I think makes sense but maybe you can be more specific on that point.

Thanks,
Marcus

> On 16. Mar 2020, at 20:11, Geeky Pathak <hpathak336 at gmail.com> wrote:
> 
> Hello everyone,
> I am Himansu Pathak final year undergraduate student. So, this summer I want to work on  3 things to add  mlpack
> 
> 1. First one Is trying to RBF as a layer in mlpack I am right now I am working on this one. So, in this Implementation, I am trying to add a layer for Radial Basis Function which can be used with a feed-forward network and help us to construct a Radial Basis Function Networks.
> To test this network I will try to use the same tests which are being used to test vanilla network with dropout layer.
> 
> 2. Second thing is to add RBF kernel right now we have only linear kernel SVM in mlpack I want to add support for RBF kernel as well. So, to do this I have two ideas in my mind 
> a) first one to add RBF kernel inside linear_svm directory and change its name to SVM  just like we are doing with GANS
> b) the second one will be to implement it in a separate place and adding everything separately
> Please suggest What do you think? for this one.
> 
> 3 Last but not least is to implement a DBN (Deep Belief Nets). We already have an implementation for RBM(Restricted Boltzmann Machine) in mlpack. So, my task to implement this is quite simple I will try to change that implementation So that I can use it as a layer. I need to implement some visitor for this and create an implementation for DBN to train these (RBM) layers.
> 
> It is a very short explanation of these problems I will try to elaborate on them in my proposal.
> Please suggest if you want any changes in any of these Ideas.
> Regards, 
> Himanshu Pathak
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack