I think a decision tree could be used to solve it. I’ll try to do an example using history data for branch instructions:
Consider a branch that follows the pattern 3 x “jump to a given address”, 1 x “do not jump (i.e. goes to the following address)” for a considerable amount of time. For simplicity, let’s represent it as “1110”. Let’s also consider we store the history from the last 8 times this branch executed and that we want to use this data to predict if the jump will occur. Then, a decision tree should map the following states:
11101110 -> 1
11011101 -> 1
10111011 -> 1
01110111 -> 0
Based on some HTM School videos I’ve watched (btw, you’re doing a fantastic job!) and rock, paper, scissors results, I think a Temporal Memory would learn this sequence very quickly, and should also adapt quicky when the pattern changes.