Monitoring a population of dependent processes under limited resources is critical for abnormal events detection. A novel online collaborative learning method is proposed to adaptively allocate the resources for exploitation of high-risk processes and exploration of dependent dynamics. Efficiency of the proposed method is proved through theoretical analysis and experiments.