[mlpack] Example Zoo

[Marcus Edel]

Hello Kaushal,

thanks for getting in touch, I like all three ideas, especially audio clustering sound
interesting. My main recommendation is to swap the second example with something
else. At this point, there is no Transformer implementation in mlpack and it sounded
like your main focus is on implementing examples rather than implementing a new
layer/model; correct me if I'm wrong.

Let me know if I should clarify anything further.

Thanks,
Marcus

> On Apr 9, 2021, at 6:03 PM, Kaushal Choudhary <kaushalc64 at gmail.com> wrote:
> 
> Hello Sir,
> 
> I am Kaushal Kumar Choudhary from Sage University, Indore, India pursuing B.Tech in Computer Science .
> 
> I have been working on the Idea of Example Zoo for a month now and had got some Interesting and useful Examples for this project.
> 
> As the original Idea of Sir Marcus Edel of creating an Example Zoo where Students/Developers can learn and visualize working machine/deep learning models in order to engage on Machine Learning with C++ and use mlpack extensively for their projects.
> 
> There are three Working Zoo Examples I want to present in GSoc 2021 ,only three because of the reduced time this year.
> 
> 1. DCGAN(Deep Convolutional Generative Adversarial Networks) :- In this project we will show an example of generating MNIST digits from a random seed after training it with MNIST image Dataset . It consists of a Generator(The Artist) and a Discriminator(The Art Critic) ,where the generator tries to generate fake images of real ones and the discriminator identifies the fake images .The discriminator is a CNN based image classifier and generator uses conv2DTranspose for generating images. making a GIF of this generator discriminator process will be a cool and effective way to learn this model  .
> 
> 2. Transformer Model for language understanding :- In this example we will show how to translate portuguese to english language .This uses scaled dot product attention and multi head attention layers instead of RNNs and CNNs.
> 
> 3. Audio Clustering :- Audio dataset is yet to be used in mlpack ,well this being a new dimension to mlpack library is quite interesting as well . Students/Developers will be more intrigued and eager to understand . The .WAV dataset is not supported in mlpack so for this I have to write a custom WAV file reader and after applying STFT(short time fourier transform) ,it will become an array and will be able to load in mlpack using data::load and then convert it to a spectrogram for visualizing the distribution . Then we can use clustering algorithms like spectral clustering and ICA(independent component analysis) and other algorithms as well . 
> This particular example is first of its kind and I would really like all the mentors to comment on this and understand the scope of this project.  
> 
> @marcusedel @rcurtin @aakashkaushik and all other mentors , it would be really helpful if you can give some time on this and comment and  raise any questions on this.
>  
> Thankyou
> 
> Kaushal Kumar Choudhary
> _______________________________________________
> mlpack mailing list
> mlpack at lists.mlpack.org
> http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack