What is the role of sexual reproduction in evolution? This has been called "the Queen of Problems" in evolutionary biology. Mixability theory, proposed by Dr. Livnat and collaborators, examines the dynamics of allele frequencies in the presence of sexual recombination and natural selection. It holds that, in the presence of sex, natural selection favors not the best particular combinations of alleles, but rather alleles that perform well across a wide variety of different genetic combinations. Thus, the interaction of sex and natural selection promotes genetic modularity and robustness. This picture is further compounded by mutation, also explored in our lab. Intriguingly, mixability theory served as a motivational source for the development of an algorithmic technique called "dropout," which generates robust units that work well across different combinations of units in deep artificial neural networks. Dropout has become one of the key breakthroughs leading to the success of deep learning and thus the AI revolution of our times.
