Having divided the ML lifecycle into those categories, the types of bias that can occur in each category were identified, which on the one hand allowed the definition of the said biases, and on the other hand a
If 𝑁<𝐾N<K, then 𝐻̂(𝐗𝑛)H^(Xn) by definition must be underestimated, and we see the bias towards synergy in the naive O-information estimation. We also see that this bias decreases to zero very quickly as soon as N increases above K. As the quantile binning eliminates ...