Given: 10 nodes (people), it's known that 3 of them are criminals There are n (< 10) matrices 10x10 of interaction of different type between each two people(some of them are cooperations of different types, some of them a betrayers of different types), it's needed to determine criminals, with a playoffs a3, a2, a1, a0 for guesssing 3, 2, 1, and 0 of them (assuming they have high level of cooperation to reach single goal, but also trying not to reveal themselves)
My naive plan:
- create a weighted sum of these matrixes
- outcome with criminals selection algorithm by result matrix
- run a ML algorithm on it
- which selection algorithm to choose?
- in case we have complex selection algorithm we have to use reinforcement learning only, am i right?
- in case we find good selection algorithm and we can use linear regression, how do we represent input and output rather than matrix expansions?
Another plans are welcomed