I was decided to change my halted advances about this subject. So I started with a full course on a MOOC (deeplearning.ai, but could also have been the fast.ai one). It’s not hard to confess that I were totally wrong from the beginning. For the newbes (as I’m), I have only this recommendation, you MUST go through one of this courses just to have a barely idea of what’s is going on. There is a lot of algebra implied (to get the thing in a nice shape), and a lot of tricks to get the shape fit (convoluting, pooling, padding, striding,…).

If you’re like me, I was so eager to learn that fully completed 3 of 5 Andrew Ng courses under a week.

So, it’s now that I’m starting to understand the graphics you post in this blog, or why are those noisy charts there.

But even I had listened to all the interviews in that course, and Geoffrey Hinton was the first one, I did not get noticed of the capsule networks until now. I imagined him as a respetable man doing his things in his lab/university and sendind papers that were not accepted… (I though, for myself. what can he share if the community is not hearing him?). Thank you for posting this.

If anyone else interested I would recommend to take a look at Max Pechyonkin posts (Hinton’s Capsule Networks Series), as I think they are a bit less academic than the papers.

I made myself a purpose to VBA/Excel a ConvNet add-in this year. It’s eleven months from now to see if I reach the target.

See ya!. And keep posting

Thank you, Juergen,
I’m really excited by this beast of an algorithm and am already finding uses for it in the real (Python/Tensorflow) world.
Best wishes,
Richard