Transfer learning to scale deep Q networks in the context of airline pricing

Dynamic Airline ticket pricing is a complex process, wherein airlines determine the best price for varied business contexts that encapsulate several factors. While most airlines use traditional revenue management (RM) systems to do this, studies have shown that deep reinforcement learning (DRL) models could maximize revenue by expanding price discovery. However, scaling these models to all routes of an airline would be cost-intensive. To help address this issue, we propose the application of transfer learning to share the knowledge gained from DRL, between similar routes, potentially helping airlines inch closer to putting a DRL-based pricing-model in production.