I begun my AI research in third semester by means of independent tutoring by then M.C. Juan Ramón Castro. Time has past and last semester I taught a course on Artificial Intelligence with a strong focus on game theory at TBC and its relevance to communications engineering.
Artificial Intelligences attempts to emulate the human thought process by means of computers. Computers offer a much greater data storing capacity and offer superior response rates to a human brain, however computers are unable to cope with noisy, incomplete and often ambiguous data as well as we can. Many techniques are focused on the problem solving techniques used by humans, which deal with this kind of data.
Widely adopted AI techniques include:
The first three are considered softcomputing techniques. These techniques make use of incomplete or imperfect sets information with degrees of uncertainty in computer reasoning, as opposed to the "hard" delimitation of Boolean logic in which things are or aren't.
Neural networks emulate the "hardware" of the brain by means of dividing the task in hand between many limited processing entities (neurons). These are all interconnected, and store their knowledge in these interconnections (network). A neural networks capacity for problem solving depends on the number of neurons, type of neurons and network architecture.
A good place to start may be this perceptron learning applet(Laboratory of Computational Neuroscience, Swiss Federal Institute of Technology Lausanne).
Lots of information may also be found at the
Neural network warehouse
Fuzzy logic attempts to emulate the "software" of the brain. This is to utilize the same reasoning process.
In fuzzy logic uncertainty is its essential component. It permits the use of constructs such as "its a little hot" or "I'm going very fast". A fuzzy variable may have a value of truth in the range between 0 and 1. These variables are subject to a set of rules from which information is inferred.
A good reference is the first part of Lofti A. Zadeth's paper "Toward a perception-based theory of probabilistic reasoning with imprecise probabilities"(BISC The Berkeley Initiative in Soft Computing Institute of Technology Lausanne).
I developed A Fuzzy Logic kit, an abstract of this poster may be found here.
Genetic algorithms provide a way for optimum search alternative to gradient descent methods. In the first iteration random values are given, these are evaluated discarding unsuitable ones. On the next iteration new random values are introduced and altered (mutated) nearer optimum values are obtained.
While obviously not an Artificial Intelligence technique itself, it allows agent oriented problem solving as well as the creation and modelling of intelligent behaviour in itself (such as p2p networks). That said it, it is an interesting field that will be briefly discussed here.
