This is a method by which an algorithm like the tolman eichenbaum machine could be used in a real environment. In it the output of the voting layer gets converted to an image/frame. In order to train the algorithm human experts will wear recording headsets for some time to create a data set. in order to establish processing I believe that the frontal lobes are responsible for setting up new graphs from old.they do this after considering the relationship between the reward function and the generated internal environment. so what we need to do is generate a separate environment and feed it’s output together with the state to the algorithm. Since the explicit rewards would have been extracted, through the use of the equivalent of lie detector body signal monitoring equipment worn by the human experts, they will be used to evaluate the fitness of the generated fake environment which will be fed in batches together with the state in order to train the algorithm. We won’t have access to what the human experts were imagining so this part will have to be trained or optimised afterwards with reinforcement learning.so we alternate between generating the imagination and generating the state. We hypothesise that in the act of generating fake environments internal dialogue will emerge as a plagiarism from reality as well as interesting dynamic will occur as a result of the relationship between the extracted reward function and the found states in the training set when considering what imaginings the algorithm will produce.
1 Like