In this paper, an approach for rapid broadband discrete nanomechanical mapping of soft samples using atomic force microscope is proposed. Nanomechanical mapping is needed to investigate, the spatial distribution of nanomechanical properties with dynamic evolution-provided that the mapping is fast enough. To enhance the mapping efficiency, we propose to significantly reduce the number of measurements by only implementing the technique at locations of interest, which further enables the broadband nanomechanical property study of soft samples undergoing dynamic processes. Firstly, an online searching learning-based optimization scheme is proposed to achieve the rapid probe engagement and withdrawal with minimum probe-sample interaction forces at each sample location. Then, a decomposition-based learning approach is used to achieve the rapid probe transition between sample locations. The proposed technique is demonstrated through multiple-location viscoelasticity measurements on a Polydimethylsiloxane (PDMS) sample in AFM experiment.