In machine learning, there is always a trade-off between exploration and exploitation. When in an unusual circumstance, one should explore to gather data and the longer the period of exploitation, the more exploration should be done in new directions.