The presentation should not be longer than 20-25 minutes and there will be a 5-10 minute discussion session after the presentation. You are free to choose whether you want to give a blackboard presentation or use slides. Timetable My presentation is on Friday 27th May in L5, at 12:30. Practice it
See slides at slides for a nice and ordered presentation of the ideas.
Evolution
Contingency, convergence and hyper-astronomical numbers in biological evolution Notes on Ard Louis' paper on contingency, convergence and hyper-astronomical numbers in biological evolution
Hunting Darwin's Snark: which maps shall we use?
Bias in GP maps
The effect found in many Genotype-phenotype maps by which some phenotypes have many more corresponding genotypes, than other phenotypes. This effect is important in Evolution
See MMathPhys oral presentation
Simplicity bias
Effects of bias in GP maps
Arrival of the frequent
- The Arrival of the Frequent: How Bias in Genotype-Phenotype Maps Can Steer Populations to Local Optima pdf
–The structure of the genotype–phenotype map strongly constrains the evolution of non-coding RNA pdf. Notes on the RNA GP map bias paper
Common features of GP maps
Origin of bias in GP maps
- Constrained and unconstrained parts. See The organization of biological sequences into constrained and unconstrained parts determines fundamental properties of genotype–phenotype maps pdf
- Algorithmic information theory: Shannon-Fano code, universal probability. See below. See draft paper.
- Random finite transducer. See Simplicity bias in finite-state transducers
Genotype-phenotype map (GP map)
Evolutionary Robotics and computing, uses GPMs. See Evolutionary computing and Optimization .. See References from Complex Behavior in Evolutionary Robotics book
Survival of the flattest
An effect, where effectively large neutral spaces are also favoured, but in equilibrium, not out of equilibrium as in the Arrival of the frequent
More
Convergent evolution as natural experiment: the tape of life reconsidered
Applications to Deep learning and ANNs? Chico's application to networks. His slides
Relation b/w bias for simplicity in GP maps, and regularization in Machine learning.
Genotype is the weights of the NN, phenotype is the function the NN approximates. NNs are expected to find "simple" functions much easier then I suppose. In other words, they are able to recognize patterns much more easily if there is actually a pattern (in the sense of a simple pattern..)