The ultimate objective is to build / train modules that can be plugged together into other neural nets, thus a big design goal of the project is that the output of each "layer" of the neural network is meaningful (Judea Pearl might even call it "casual"). What this means is that I'd like my AI to build "eigenimages" (aka archetypes) that I, as the human expert, would agree is meaningful
Plan Specificaiton
To get to that goal, I'd like to put together the following:
Gather a bunch of screenshots from league
Pipe each image through a neural network consisting of:
First Phase Training
The ultimate objective is to build / train modules that can be plugged together into other neural nets, thus a big design goal of the project is that the output of each "layer" of the neural network is meaningful (Judea Pearl might even call it "casual"). What this means is that I'd like my AI to build "eigenimages" (aka archetypes) that I, as the human expert, would agree is meaningful
Plan Specificaiton
To get to that goal, I'd like to put together the following:
Requirements
In order to satisfy the above specs, engineering will need to put together the following:
TODOs
References