5 ms·
Hey, author here. Any/all feedback is welcome, and I'm happy to answer questions. Previous discussion on HN of the "introduction to Neural Networks" referenced
by vzhou842 7y ago
Hey, author here. Any/all feedback is welcome, and I'm happy to answer questions.
Previous discussion on HN of the "introduction to Neural Networks" referenced in this article: https://news.ycombinator.com/item?id=19320217 https://news.ycombinator.com/item?id=19320217
Runnable code from the article: https://repl.it/@vzhou842/A-CNN-from-scratch-Part-1 https://repl.it/@vzhou842/A-CNN-from-scratch-Part-1
Github: https://github.com/vzhou842/cnn-from-scratch https://github.com/vzhou842/cnn-from-scratch
- parkaboy 7y agoHonestly: the "convolution" (cross-correlation) part in your article is the clearest step-by-step explanation I've ever encountered. Well done!
- Ranlot 7y agoFor those interested in a clear explanation of the backward pass (especially for convolution layers), here's a good resource: https://arxiv.org/abs/1811.11987 https://arxiv.org/abs/1811.11987 https://github.com/Ranlot/backpropagation-CNNs https://github.com/Ranlot/backpropagation-CNNs
- savant_penguin 7y agocame here exactly looking for the backward pass
- vzhou842 7y agoI'll have a sequel explaining the backward pass with full code up by next week!
- p1esk 7y agoI've done it as an exercise a while ago: https://github.com/michaelklachko/NN/blob/master/Python/convnet_in_python_single_layer.py https://github.com/michaelklachko/NN/blob/master/Python/conv...
- alteria 7y agoI like your writing style, clear and concise. Your NN posts would have been super helpful when I was taking an Intro AI course.
- SomeHacker44 7y ago“Each of the 4 filters in the conv layer produces a 26x26 output, so stacked together they make up a 26x26x8 volume. All of this happens because of 3 × 3 (filter size) × 8 (number of filters) = only 72 weights!” There seems to be a mismatch here with 4 and 8. Probably 4 is wrong.
- master_yoda_1 7y agoFirs thing first stop stealing. You copy code and figures from karpathy lecture notes and did not cite him. Its call plagiarism.