'Neural Networks and Deep Learning' or why I like KISS

Cet article a été traduit en français

I recently graduated from the first deeplearning.ai course on Coursera: “Neural Networks and Deep Learning”.

What I liked about the course was the fact that I was able to design a deep neural network by myself, without using frameworks such as Tensorflow, Theano or Torch.

The idea is that I like to understand how things work. This may seem counterproductive and it is true that in my work I sometimes have to put a lot of effort in, while some of my friends advise me to make my life easier by using pre-packaged solutions.

But my experience has always shown me that it is really difficult to get exactly what you want from a third-party program. That’s normal. I don’t blame the people who developed it, and I’m sure very few people find what I write useful (to be honest, I don’t really know who is interested in what I’m doing).

And in fact, this has always paid off in the end. That’s the main reason why, at [Mediawen] (http://mediawen.com), we are completely technologically independent.

That’s why I like simplicity. I think it is better to give developers bricks that they can use and modify rather than a gas plant that they will be tempted to abandon from one day to the next because they have encountered difficulties.

Keep it Simple, Stupid. That was the philosophy of Unix, before PulseAudio, SystemD and all those nasty things which are impossible to configure or modify simply. That’s why I remain a big fan of FreeBSD and Plan9 operating systems.

Anyway, I still have four more courses to take. I hope to find the time…