Post

It's alive! Input layer at the top, output layer at the bottom. Connection colour represents positive (green) or negative (orange) weight, thickness and opacity represents how large that weight is. Really neat to see the training process strengthening specific connections in the network.

27/12/24, 10:24 pm2 reposts70 likes

@joshwcomeau.com

Every ML tutorial I’ve ever seen has these graphs and I still have only a vague idea about what they mean. 😂 Looking forward to your post!

They’re surprisingly simple, I’m looking forward to helping gold understand what they’re seeing.

*folks, not gold. Thanks autocorrect.

@lemancha.bsky.social

what is it?

It's a visualised multilayer perceptron. Basically a binary classifier neural network.

@maxime.bsky.social

what's the training process behind this (i.e. what is backing those numbers/what are you training for) 👀?

Behind the scenes I’ve ported github.com/karpathy/mic... to TypeScript and made a few modifications to make it easier to hook up to a visualisation. The actual data I’m running through it is just made up BMI data, and it’s training to recognise people in the “normal” BMI category.

I don’t think I’ll use the BMI example in any post, it’s not a great demonstration of the tech because it’s just a formula. But it was easy to fake and it’s a small number of input parameters, so it’s helping me develop and test my code.

Sebastian Gabriel

@sgabriel.bsky.social

I love working with SVG, never seen Yoga used for that, great work! I would love collaborating on those graphics

Justin Garrison

@justingarrison.com

Did you build this or is it another site/tutorial?

I built it. I’m in the “let’s see what we can do” phase of writing a post about neural networks.

Justin Garrison

@justingarrison.com

I’d like to review it whenever you have pieces ready 👍

I’d love your feedback! I’m fairly sure I’ll be writing the post at this point, I feel like I can pull the pieces I need together. It’s a ways off, though. I don’t yet know what the “story arc” is or what the endpoint would be. I also need to do a bunch of background reading.

Also for anyone who might be thinking I’m selling out and jumping on the AI hype wave, the multilayer perceptron was first described in 1958. Nice try.

SVG is such a cool technology. The web is so cool.

True SVGs are cool especially for these applications so flexible..

What’s a true SVG?

It's a typo haha I meant to say true to your statement

@alexandersix.com

Dang this is a cool visualization!

It’s rough atm. I need to make it more clear what weights contribute to the output, had some trouble coming up with a nice way to normalise it. In reality it’ll go slower, and much simpler networks will be shown before this one. But ultimately I want readers to train something useful in the post.

@alexandersix.com

Yeah, this feels like one of those areas where you have to bring people along the journey ~slowly~ since it can be so complex. The visualizations are a great tool for that, though