The title is a bit misloading; this paper is specifcially about using machine learning to search for patterns in the distribution of prime numbers.
I can see how they got there, though; you can take inspiration the Sieve of Eratosthenes1 and consider the sequence of prime numbers as the simplest aperiodic sequence. In general though, most sequences are aperiodic, and due to their definition, the prime numbers still have various trends to their distribution that could be considered patterns although aren't periodic repetitions of identical parts.
Footnotes
one of the simplest yet most powerful number theory techniques for generating prime numbers; the idea is to find them all, one by one, by increasing order, and after finding each prime, rule out all its multiples; the smallest number not ruled out yet is the next prime. It's got a beautiful simplicity. ↩
Footnotes