• deep reinforcement learning
• graph convolutional networks
• network slice placement

The Network Slice Placement problem consists of placing chains of Virtual Network Functions onto a physical infrastructure, subject to a variety of constraints and requirements, and Deep Reinforcement Learning (DLR) algorithms are a promising approach to tackle this problem because of their ability to learn from experience and the effectiveness of Deep Learning to extract relevant features. This work exploits DRL and Graph Convolutional Networks, obtaining good results in terms of the requests acceptance ratio, and producing an agent capable of generalizing to different load levels and connectivity patterns in the physical infrastructure. This work also include the realization of a simulator specifically designed for using RL in networking resources allocation problems.