In cellular telephone networks, each new caller must be allocated a channel (within the provider's frequency band) that does not interfere with those allocated in nearby areas. This quickly becomes a serious problem when many callers congregate in particular areas, such as entertainment venues.
Reinforcement Learning (RL) has previously been applied to this problem:
Singh and Bertsekas (1997). "Reinforcement Learning for Dynamic Channel Allocation in Cellular Telephone Systems"" in Advances in Neural Information Processing Systems (NIPS), pp. 974-980.
However, in that work, the critical features given to the learning system were determined by hand. The goal of this project is to allow a deep neural network to discover some of the key features needed to support RL. In general, the system will receive a more raw form of data than that used by Singh and Bertsekas and then exploit Deep Learning's ability to discover salient patterns (a.k.a. features).
This project will use Google's Tensorflow system for Deep Learning and must combine it with a reliable RL system and a simple simulator for multi-channel telephone networks.
IMPORTANT: If you sign up for this project, please send a) your CV (including a transcript with all of your college grades, and b) a brief explanation of WHY you want to do this particular project to Prof. Keith Downing (firstname.lastname@example.org)