6 ms·
Learn about Neural Networks and Backpropagation
- lettergram 8y agoI wrote a similar style guide introducing neural networks from embeddings, data splitting, to MLPs, RNNs, and CNNs and event basic hyperparameter tuning: https://austingwalters.com/neural-networks-to-production-from-an-engineer/ https://austingwalters.com/neural-networks-to-production-fro... I think what generally makes articles unapproachable is their insistence on jumping right into the math. Even this on one section in... I understand the math, but it makes it far less approachable. In reality, at least today, utilizing neural networks doesn’t even require that much understanding of the math. It’s mostly intuition about how they work. I do like the article and think it’s good (upvoted). I just wish they focused more on the intuition than math of it.
- shock 8y agoThe `Canvas Defender` addon warns me that you are doing canvas fingerprinting on austinwalters.com. If that's the case, I'm curious why you need that on what seems to be a personal blog.
- lettergram 8y agoHuh, I’m assuming that’s one of the Wordpress plugins. Not intentional and will track it down. I do use google analytics, so I wouldn’t be supprised if that did it.
- shock 8y agoI block google analytics, so I don't think that's the source. Thanks for looking into it.
- mediumpost 8y agoThanks for your comment and suggestion. I understand that math makes it far less approachable, but I did it like that based on my experience. I understand the concept easy but I was missing the explanation of backprop and all those formulas. However, I would reconsider putting more about the intuition of NN (for broader audience) and maybe making the math part more like "optional" or "bonus" part of the post.
- lettergram 8y agoI mean, to each their own. It’s probably better we (as a community) present it both ways. I just typically find people interested in neural networks (data scientists or the like) use the mathematical representations, but engineers typically just have their eyes glaze over. I suppose it depends on the target audience
- anjc 8y agoThis is great, thanks
- mediumpost 8y agoGlad that you like it, thanks
- thisismyaccoun7 8y agoI highly suggest the Stanford course https://cs231n.github.io https://cs231n.github.io. All the notes are in the link. I found it a very good introduction to backpropagation