Recurrent neural network

[wangjiawen2013]

Hi,
Does burn provide recurrent neural network now ?

[laggui]

We've have RNNs for a while now. Gru and Lstm/BiLstm more precisely.

Are you looking for something else specifically?

[wangjiawen2013]

In fact, I'm trying to adopt burn in my daily work. There are lots of artifical neural networks now (https://www.geeksforgeeks.org/types-of-neural-networks/) and I am interested in Generative Adversarial Networks (GAN), so I wonder if burn is well-established.

The types in the red rectangle have already been included in burn. What about others, do you have a roadmap ?

[nathanielsimard]

Burn is a dynamic neural network framework. You can build pretty much any neural network modules with it, but we may not have premade implementations for all kinds of neural networks.

[laggui]

We have transformer blocks. Autoencoders, as with a lot of the other networks in the list, are just combinations of layers that we already support (e.g., convolutions) with a specific structure and loss.

Spiking neural networks I am not very familiar with but I think that's an entirely different type of application which might not be suited to what we offer.

[wangjiawen2013]

Hi, Could you tell where to find autoencoders and generative adversarial network examples written in burn ?

[laggui]

We don't have any official examples, but I remember a community member posted this VAE implementation on discord a while back: https://github.com/code-cp/vae-burn