Original Blog Entry: Patterns that are not visible

hypersoniq — Wed, 05 Aug 2026 16:50:18 GMT

In a recent exercise on a biology problem, the glaring inefficiency of lottery systems became apparent. The problem asked to scan 50 entries of 100 nucleotides (letters A, G,C and T) to find a motif (pattern) of a fixed length, Even if the pattern is not exact due to single nucleotide polymorphism (over evolutionary time, a C may be changed to another letter. So this introduced the concept of edit distance, or the number of changes needed to turn one string into another, this was then the constraint... find all matches within a certain edit distance. Now, the problem supplied the fixed motif length, but the authors brought up a great point... what if you do not know even the length of the pattern? That hit home for the lottery problem... look at how many parameters we are left to guess... how many draws to use? Spotting a trend before it goes away? What about patterns that may hold but some of the numbers change, completely obscuring the pattern to begin with? I admit that the CS degree trains one to find deterministic answers, and should cause one to reject the idea of a lottery system to begin with... but Biology introduces messy, noisy data, non deterministic answers that must look to probability to make accurate calculations, best guesses while sticking to biological realities that can't just be glossed over by coding tricks. Some of these techniques might be used in this problem domain. I still need to figure out which ones and exactly how, but after over 20 years of a pointless chase, this may be the inspiration needed. If patterns that are similar but not exact can be identified, this could also answer such questions as how many draws to use or how there might be other less obvious structures instead of direct followers to apply a Markov decision process to. Not rushing back to the lottery lab anytime soon, but problems that seem impossible in one domain may be solved by applying techniques from a completely different domain.... [ More ]