Evolutionary Approach to Machine Learning and Deep Neural N. N-E...

Around 1990, I learned about GP (Genetic Programming) and developed interest in this area shortly after learning about GAs (Genetic Algorithms). At that time, the term GP was not yet established, and the concept was denoted by terms such as structural GA and tree representation GA [1–3]. Since I had been researching AI so far, I presumed that the only goal of GA was optimization (my opinion has changed since), and I considered GA to be somewhat unsatisfactory. Consequently, as I assumed that GA could not be used for handling knowledge representation, programs, concept trees, and similar notions, I attempted to extend it. At exactly the same time, when I presented my research to Dr. Philip D. Laird from NASA, who was a Visiting Researcher at the Electrotechnical Laboratory. Dr. Laird is a researcher in machine learning and is renowned for his book Learning from Good and Bad Data [4]. He introduced me to the research of Prof. John Koza of Stanford University. I still remember the excitement I felt while reading the technical reports [5] written by Koza over the new year holidays. These reports were later compiled in a massive volume that exceeded 800 pages [6]. Afterward, I stayed at Stanford University in his laboratory and had a wonderful research life there. Recently, an increasing number of researchers and students are specializing in only GA and GP (and fields centered on them). Also, in comparison with the time when AI and GA/GP were created, there are fewer and fewer researchers with unique personalities. Although this is not in anyway disappointing, it cannot be understood unless one has experienced the turmoil and the emotions associated with the establishment of a new field. I expect such pleasant nervousness and excitement to be maintained in this academic society as well, setting the path for the development of a healthy community. This is one of my motivations in writing this book.