Information-centric sensor networks (ICSNs) are a paradigm of wireless sensor networks that focus on delivering information from the network based on user requirements, rather than serving as a point-to-point data communication network. Introducing learning in such networks can help to dynamically identify good data delivery paths by correlating past actions and results, make intelligent adaptations to improve the network lifetime, and also improve the quality of information delivered by the network to the user. However, there are several factors and limitations that must be considered while choosing a learning strategy. In this paper, we identify some of these factors and explore various learning techniques that have been applied to sensor networks and other applications with similar requirements in the past. We provide our recommendation on the learning strategy based on how well it complements the needs of ICSNs, while keeping in mind the cost, computation, and operational overhead limitations.