Create a Boltzmann Machine Object

[soniajoseph]

Boltzmann machines are similar to Hopfield networks, except they are stochastic. Create a Boltzmann machine object with the same functionality as the Hopfield network object, mimicking its structure with the appropriate, Boltzmann-specific changes.

[andreanne-lemay]

I'd love to contribute to this first issue, is there a mattermost channel to discuss the project?

[soniajoseph]

My mattermost account is currently pending (there are issues with sending out confirmation emails). Let’s discuss it here or DM on Twitter @soniamjoseph until their bug is fixed.

…

On Tuesday, June 16, 2020, andreanne-lemay ***@***.***> wrote: I'd love to contribute to this first issue, is there a mattermost channel to discuss the project? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHBIYMBGWL6ZYXJPQZF75BTRW7BSHANCNFSM4N74QN4Q> .

-- Sonia Joseph Computer Science Computational Neuroscience Princeton University soniajoseph.github.io

[andreanne-lemay]

Great so I'm currently working on implementing this paper: Salakhutdinov, Ruslan, and Geoffrey Hinton. "Deep boltzmann machines." Artificial intelligence and statistics. 2009. If you have any other references that might help with this issue, I'm glad to look into it :)

[soniajoseph]

Sounds fantastic, let me know how it goes! Hopefully Mattermost will clear up by tomorrow and we can chat about it then.

[andreanne-lemay]

I don't have permission to push the code from my branch, but I started to work on first version of a RBM. However, my network has a weird behavior. I reused the code you did for testing the hopfield network, and when running on FMNIST dataset for denoising I get this output:

If I train on only one image, I get this, which is better, but not perfect like the Hopfield network:

I am not quite sure what is the issue, I could definitely use a second opinion on my code. In the mean time, I can start to work on the other issue :)

[soniajoseph]

I've added you as a contributor. Please make the changes in a separate branch so we can keep track of progress :) Interesting results, I'll look at the code once you push it. ᐧ

…

On Tue, Jun 16, 2020 at 10:19 PM andreanne-lemay ***@***.***> wrote: I don't have permission to push the code from my branch, but I started to work on first version of a RBM. However, my network has a weird behavior. I reused the code you did for testing the hopfield network, and when running on FMNIST dataset for denoising I get this output: [image: image] <https://user-images.githubusercontent.com/49137243/84857384-f5654a00-b036-11ea-8c83-e8dcebee3da3.png> If I train on only one image, I get this, which is better, but not perfect like the Hopfield network: [image: image] <https://user-images.githubusercontent.com/49137243/84857578-5e4cc200-b037-11ea-8202-7680e7cf8668.png> I am not quite sure what is the issue, I could definitely use a second opinion on my code. In the mean time, I can start to work on the other issue :) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHBIYMGDFBDC7ANMKHGBFB3RXBG4NANCNFSM4N74QN4Q> .

-- Sonia Joseph Computer Science Computational Neuroscience Princeton University soniajoseph.github.io