Using neural nets to recognize handwritten digits

fermatslibrary.com

53 points·mondaine··12 comments

Unless it is the same person who created this PDF and this [1], it's some kind of plagiarism.

[1] http://neuralnetworksanddeeplearning.com

This is fermatslibrary.com a platform to annotate scientific papers. But I don't understand why that paper you cited is there - it's easy to read already.

Agree. OP link should be changed to http://neuralnetworksanddeeplearning.com/chap1.html

FTA: "This is the first chapter of Michael Nielsen's book Neural Networks and Deep Learning."

The book itself is CC BY-NC 3.0.

I would say that writing a "hidden" (by default) comment saying who is the author is stretching the "BY" too much...

I'm guessing that the creator simply printed the webpage (without link on the header/footer), and has honestly referenced the source on the first comment. So probably there was no malice intended.

I did however leave a Disqus comment on how they could make it clearer. The ones I could think of -- either print HTML to PDF with a link on the header, or modify the source of the page to add a note before you printing to PDF.

There's also this tutorial from a while back:

https://www.tensorflow.org/versions/r0.11/tutorials/mnist/be...

"Most people effortlessly recognize those digits as 504192."

I initially recognized that as 504/92 based on the slant. Character recognition can be surprisingly context sensitive.

Did you see the later training data example? http://imgur.com/DRll02M

Neutral nets to recognize digits were the holy Grail of postal mail sorting in the '80s!

What do they do now?

If you only read HN, you'd think that the only application of machine learning is image recognition. Just once I'd like to see a more interesting example...

Reply on news.ycombinator.com