Adaptive workforce exploration in complex productivity landscapes

What the math gives to ML

The paper transfers the NK-model mechanism of adaptive specialization: interdependent binary attributes create a tunable ruggedness parameter k, while permanently assigned specialists exploit local task optima and stochastic generalists explore across tasks. Its useful neural-network asset is a measurable exploration-exploitation controller based on local loss-landscape ruggedness rather than a fixed mixture of shared and task-specific parameters. Implement this as a specialist-generalist optimizer or modular network, and test whether the optimal generalist fraction changes as estimated parameter-perturbation correlation decreases with increasing task interdependence.