Mathematical Evolution of Human Behaviors - Osama Salah Eldin

572 views
Download
  • Share
#Osama Salah Eldin #2015 #IEEE CIS Webinar competition 2015 #Human behavior modeling #Intelligent agents #Genetic algorithms

Human behavior modeling has been the focus of many researchers over the last several years. It is spreading enormously over a wide variety of applications such as video games, social studies, marketing, and more. Different techniques have been used to dress intelligent agents in a human‐like behavior.

Most previous trials to model human behaviors targeted predicting the future actions that a person or a group of persons may take. This includes things like predicting the next steps of a video game player, estimating the movements of a pedestrian on a street, or anticipating the future choices of a user based on his past choices. This webinar talks about the artificial imitation of actual human behaviors such as wisdom, rashness, carelessness, etc. Once an intelligent agent (IA) imitates a certain human behavior, it can autonomously take the same actions that a human user, with the same behavior, is expected to take. That is: a wise IA is expected to behave like a wise person when encountering the same situation.

Behavior is defined as “The action or reaction of something under specified circumstances.” This means that for an intelligent agent to imitate a human behavior, it must take the same actions a human may take when encountering a certain situation. Thus, a human exemplar must exist. Imitating that exemplar follows one of two techniques: direct and indirect behavior imitation.

In direct behavior imitation, a controller is trained to output the same actions a human took when he faced the same situation. This means that the performance of the controller depends on the performance of the human exemplar. On the other hand, an indirect behavior imitation controller is allowed to generate actions freely, that is, unguided by a human exemplar. Afterward, the controller is evaluated by comparing its actions with those of the human exemplar. This freedom in generating actions gives the indirect techniques more generalization over the direct ones. The novelty of the method introduced in this webinar is that it gives the IA more freedom and more generalization even over the indirect techniques. It extends the freedom of the indirect techniques by allowing the IA to evaluate itself without any human exemplar. That is, the IA is just provided with a fitness function and left in the environment. It examines the environment, evolves different human‐like behaviors, and selects the fittest one of them for that environment.

The novel IA introduced in this webinar uses GA along with two optimization methods (CMA‐ES and Nelder‐Mead) to evolve human behaviors. This is accomplished by formulating the given problem into a mathematical objective function and using the aforementioned methods to suggest different solutions of that objective function. Every solution of the objective function can be interpreted as a human behavior.

The presented technique is tested on Robocode game. It is noticed that a Robocode agent that uses that technique is able to not only evolve human‐like behaviors but also select the most appropriate one for every opponent.

In this webinar, Osama Salah Eldin talks about the artificial imitation of actual human behaviors, such as wisdom, rashness, or carelessness. Once an intelligent agent (IA) imitates a certain human behavior, it can autonomously take the same actions that a human user, with the same behavior, is expected to take. That is, a wise IA is expected to behave like a wise person when encountering the same situation.

 

Advertisment

Advertisment