Learning curve for collective behavior of zero-intelligence agents in successive job-hunting processes with a diversity of Jaynes-Shannon's MaxEnt principle

Inspired by the unsupervised learning or self-organization in the machine learning context, here we attempt to draw learning curve' for the collective behavior of job-seeking zero-intelligence’ labors in successive job-hunting processes. Our labor market is supposed to be opened especially for university graduates in Japan, where the students have several successive chances n=0,1,2,… to obtain their positions within an academic (business) year. In this sense, the cumulative unemployment rate' in our model system is regarded as an error-measurement in the collective intelligence of students, and the job-hunting stage n-dependence of the error constructs a learning curve. In our simple toy-model of probabilistic labor market, the diversity of students' behavior is built-in by means of the Jaynes-Shannon's MaxEnt (Maximum Entropy) principle. Then, we discuss the speed of convergence for the error-measurement, where we consider a scenario in which the students do not use any information about the result of job-hunting processes in the previous stage. Our approach enables us to examine the existence of the condition on which macroscopic quantity, say, stage-wise unemployment rate’ becomes `scale-invariant’ in the sense that it does not depend on the job-hunting stage n. From the macroscopic view point, the problem could be regarded as a human resource allocation.