Syllabus
1) AI, Generalization, and Unsupervised Learning - history: Computing Machinery and Intelligence, A. Turing - present: deep learning, neural networks - possible future: more generalization from less training examples - evolution as an inspiration for AI research instead of neuroscience (the brain)
2) Wolfram's elementary cellular automata (ECA): simple evolving models where structures emerge Barbora Hudcova: more formal classification of ECAs using their transition lengths
3) Towards metrics of complexity: Occam's razor, Minimum description length, Kolmogorov complexity, Algorithmic probability The Quark and The Jaguar, Murray Gell-Mann: measures of complexity proposed by Gell-Mann
4) Hugo Cisneros: Evolving Structures in Complex Systems - metric of structured complexity based on compression algorithms
5) Von Neumann's Self-Reproducing Automata, A. W. Burks - maybe the first attempt to design non-trivial self-reproducing systems capable of evolution
6) Studying Artificial Life with Cellular Automata, C. G. Langton - mathematical structures that can have similar properties to how we define life: self-reproduction, evolution
7) Other related topics: Genetic Algorithms, J. Holland - We will discuss the basic ideas behind evolutionary and genetic algorithms and genetic programming, and compare these algorithms with the previously discussed attempts to design objects that can evolve. Neuroevolution - Evolving neural networks through augmenting topologies, K. O. Stanley and R. Miikkulainen - Another attempt to simulate evolution that uses neural networks. In this talk, we will briefly discuss the basics of artificial neural networks, and extend these to models that can grow in complexity.
Annotation
Seminar on evolving structures in mathematics.