On the Optimum Geometry and Training Strategy for Chemical Classifiers that Recognize the Shape of a Sphere

摘要: In this paper, we continue the discussion on database classifiers constructed with networks of interacting chemical oscillators. our previous papers demonstrated that a small, regular network oscillators can predict if three random numbers in range $[0,1]$ describe point located inside sphere inscribed within unit cube $[0,1] \times [0,1] [0,1]$ accuracy exceeding $80 \%$. The parameters were determined using evolutionary optimization. Here apply same technique to investigate classifier for problem be improved by selecting specific geometry We also address questions optimum size training optimization and minimum testing dataset objective evaluation accuracy.