: Researchers use the FIM to measure how distinguishable models are based on their predictions. In sloppy models, FIM eigenvalues are distributed roughly evenly over many decades.
Below are several major papers and resources that define the field:
: A few parameter combinations ("stiff") tightly constrain model behavior, while others ("sloppy") can vary by orders of magnitude without changing the output.
In scientific literature, a "sloppy" model refers to a complex multiparameter system where model behavior is highly sensitive to only a few "stiff" parameter combinations, while the majority of "sloppy" directions in parameter space have almost no effect on model predictions.
