Hello Perception is a "Hello World" of the Perception AI logic currently written in vanilla Javascript. It is the 1st iteration of a dream to build autonomous robots.
There are libraries that do what I’m doing from scratch but that is boring. I want to understand it deeply. I want… no I NEED to create. Plus, I don’t enjoy looking up library documentation I have to do that in my job every day.
Eventually, I will be writing this as an embedded system in all of my robots, but hosting a website cost money, and my host charges more if I use any other language than Javascript. I didn’t want to pay more.
This is going to be a very long process, including a lot of learning and exciting development.
But the first step has been taken.
One of the “Gotcha” moments I had was my calculus was good, but I kept getting the wrong result until I made a larger net work. Which was confusing cause I couldn’t find any errors in my calculus and in the end there was none.
This first success has 30 layers each with 30 nodes
It uses rectified linear units as an activation method activation method for all my layers , except for my outfit layer, which uses a soft max activation . And I use categorical cross entropy for my loss method.
I am calling this a confidence curve, because it is a linear description of the confidence in the guess as it grows over epochs
I am teaching myself everything so as I grow, I will learn what I get wrong and all the steps I took that were wrong will show them selves, but right now that is what I have learned
Let the progress continue