Artificial Intelligence and Human Evolution by Ameet Joshi

Author:Ameet Joshi
Language: eng
Format: epub
ISBN: 9781484298077
Publisher: Apress

The camera can observe the motion of the arm. There is a central processing unit on the arm assembly, which we can call the brain of the machine. The brain is generating these actions as electric signals as well as receiving electric signals from the camera in the form of sensory information. This setup takes care of the first two steps outlined in the process of learning. The next steps need to be handled by the brain of the machine. First, we need to define a goal. Let’s say the goal is to move an object that is kept in front of the arm to the left by one foot. The brain of the system is able to sense the information from the camera and compare it with this goal and generate a label in the form of good, bad, or neutral. For example, if the arm moves the block in the right direction, it would generate a positive label, if it tries to move it in the opposite direction, it would generate a negative label, and if it does not move at all or moves in a perpendicular direction, it will generate a neutral label. It can even go further to generate a numeric value based on how good or bad the outcome is. Now, we have all the six steps in the online learning paradigm taken care of and we have a generic machine that is ready to start learning.

The brain can start producing random movements of the arm within its limitation and start capturing the labels based on the camera input. This process is also called exploration, where the system is just exploring the action space. If the same process continues for a long time, say a week, the brain will have a huge set of labeled actions recorded in its memory, but the operation of the arm would not improve at all. Now, in order to improve the operation of the arm and get closer to the desired goal of moving the object one foot to the left, we need to introduce the concept of exploitation. Our brain alternates between exploration and exploitation as it continues to record the labeled actions. Exploration means generating random actions without using any of the recorded memory, while exploitation means creating action that it knows would produce a good or positive label or feedback. As the machine starts for the first time, the memory is pretty much empty, and there is nothing to exploit. However, as it continues to explore more, the memory starts to get some actions associated with positive feedback or label. Then it can start to exploit these prerecorded actions and maybe try to make slight tweaks to them to see if the feedback gets better or worse. After such an operation continues for a sufficient amount of time, the arm would reach pretty close to the desired goal of moving the object by one foot to the left. As the machine gets more and more positive

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.