Amritpal Nijjar is my former master's student who dedicated a year trying to use a Genetic Algorithm to get a good analytical approximation for the damping tail of the Cosmic Microwave Background temperature lensed spectrum.

The presence of a damping tail in the CMB power spectra leads to additional difficulties in training machine-learning models. By pre-processing the TT power spectrum by simply rescaling the spectra as e^(-2\tau), our group was able to reduce the span of the data vectors by orders of magnitude, which assists in training the emulator (work led by Ph.D. student Yijie Zhu).

Genetic Algorithms (GAs) are a class of machine learning techniques that learn to fit data by mimicking evolutionary biology, such as random gene mutation and cross-breeding. GAs have been widely used in the field of cosmology, for example, to model the linear matter power spectrum, nonlinear matter power spectrum[51], and matter transfer function. Symbolic Regression, in particular, is a kind of GA that attempts to learn a symbolic expression to fit a given dataset.

Amritpal Nijjar designed a GA model that takes a fixed functional form for the lensed damping tail. The functional form for Nijjar's fit has two components, the damping tail and the lensing effect, with a total of 15 free parameters.

Amritpal Nijjar transitioned to industry, and we wish him the best of luck. Thanks for your hard work!