Hacker News Reader

That's fun, but the whole point of neural networks is to train them against data, not tweak the weights by hand. Hand-tuning a neural network is not done much - it's like setting raw filter parameters by hand.

View on HN

True, but it's useful to get an intuitive understanding quickly. This doesn't go into the various loss functions or neuron types or matrices, just play-with-it level intuition. I bookmarked it under "Research/MachineLearning/Intro"

Well said. That's exactly the point: to get "play-with-it level intuition". Even though its not rigorous, I still find it helps.

I am planning on building a simulation of the back-prop algorithm though!

Reply on news.ycombinator.com