Special Bioengineering Seminar

The coding theorem from algorithmic information theory (AIT) suggests that many processes in nature are highly biased towards simple outputs.   Here simple means highly compressible, or more formally, outputs with relatively lower Kolmogorov complexity.    We have recently derived a form of the coding theorem that can be applied to many different practical systems [1].  For RNA folding and protein quaternary structures, we observe a strong bias towards simplicity both model calculations and in nature.   Simpler (and where relevant more symmetric) phenotypes emerge not because they are favoured by natural selection, but rather  because they are easier to discover in evolutionary search.   Finally, this bias is also present in deep neural networks, and helps explain why they generalise so well.
[1] K. Dingle, C. Camargo and AAL, Nat Comm 9, 761 (2018)