The unreasonable effectiveness of additive combinatorics in computer science

While I did take upper level math courses in Uni, this is still beyond me. It does remind me of my PhD work and the approach we took.

The problem, Symbolic Regression, search all math formulas for good solutions.

The space, all math formulas, a graph of trees

The search, like djikstra, but explore from multiple points in parallel

The relation, decomposing / reducing the space by combinatorial factors. For example, a+b=b+a (assoc & community props). Eliminate bad choices over trying to make best choice. Using combinatorics to reduce wasted effort.

[2007]

A provocative title, but I'm not mathematically sophisticated enough to make heads or tails of the content, nor to grasp its connection to computer science. Would anyone care to elucidate?

(2007)