"Network Function Virtualization" (NFV) is an emerging technology and 5G key enabler. It promises operating expenditure savings and high flexibility in managing the network by decoupling the network functions, like firewalls, proxies etc., from the physical equipments on which they run. In order to reap the full benefits of NFV, some challenges still need to be overcome, namely those related to resource management, security and anomaly detection. Recently, Machine learning (ML) has been applied in different fields and has demonstrated amazing results. Utilizing Machine learning to address the challenges faced by NFV is a promising research field that requires further investigation. In this paper, we shed light on this domain by discussing the potential and challenges of ML application to NFV and by surveying existing works.