Due to the hazardous nature of landslides, timely and effective landslide susceptibility and detection results are needed for pre-disaster land management and spatial planning as well as post-disaster response, respectively. Also, interpretable models and outputs can contribute to the understanding of landslide processes and triggering factors by researchers. Nowadays, artificial intelligence (AI) has helped researchers and decision makers to bolster natural disaster management. However, major challenge in building AI-based models relates to the cost of searching and labeling the training data as well as the quality of the training data. Recently, training data minimization and historical natural hazard datasets have shown potential for the rapid construction of models. That is, the utilization of a small amount of data from a target area and historical landslide inventories from other areas has the potential to be an emerging low-cost approach for constructing landslide models. The purpose of this thesis was to enhance our understanding of using active and transfer learning for rapid landslide assessment as well as develop a framework that overcomes the limitations of active and transfer learning while providing the opptunity for plausible interpretation of models and results. Overall, this thesis provided new insights for rapid landslide assessment by using active and transfer learning. Meanwhile, the proposed framework can help decision makers to prevent and respond to natural hazards by providing timely and accurate suggestions for land management, spatial planning, and post-disaster response.