Cybernetic diagrams
Here we will be presenting some cybernetic diagrams of possible machine intelligence systems:
Reinforcement learning system
A diagram of a conventional reinforcement learning system looks something like this:
|
Explanation: Here the supervisor contains the systems goals -
in the form of a single scalar reward signal calculated on the basis of
current sensory inputs.
|
Simple compression-based system
A diagram of a compression-based agent looks something like this:
|
Explanation: Here the system's goals are represented in the
evaluator. This plays much the same role as the supervisor did - but it
operates on expected sensory inputs. In a classical reinforcement
learning system, the agent has to figure out for itself that expected utility
maximisation as an effective strategy. By contrast, here, the expected utility
framework is wired in by the programmers.
|
Compression-based system - with exploded compressor
A diagram of a compression-based agent with the details of the compressor is shown below:
|
Explanation: This is the same diagram as above - but showing
the inner workings of the compressor. The compressor is envisaged as a simple
reinforcement learning system - with the system being rewarded for successful
predictions, for building a more compact model - or for some combination of
the two.
|
Tim Tyler |
Contact |
http://matchingpennies.com/
|
|
