Deep learning weekly short piece: Neural networks made easy
For this week’s deep learning piece, I wanted to draw your attention to this excellent piece, written by Ophir Tanz, CEO of GumGum.
It gives a high-level overview of some important technical concepts in the space. I like the below graphic from the article, as it represents the “layer-wise” way that both deep neural nets, and our own brains, process images.
Note that the often used analogy to our brains is highly flawed, largely because our neurons are interconnected, allowing bidirectional flow of signals, whereas artificial neural networks (ANNs) are predominantly unidirectional in their information flow. Moreover, there isn’t any reason to believe our brains are “trained”, nor respond to inputs, in the same way as ANNs. The similarity between them is largely based on the conceptual idea that there are neurons that “process input signals and then pass the result on to other neurons”, but the analogy largely ends there.
