Learning User Preference Architecture (2018)

My research topic was highly influenced from computer games. Normally when we play story mode, there is only a handful of difficulty levels available which are fixed and not curtailed to user. Similarly there are limited environment selections in other domains like E-Learning, Physical rehabilitation, etc. There are no architectures available which adapts the game environment depending on user experience. 

Thus my research is aimed at creating an architecture which has the capability to adapt the domain environment in ways which improves user performance and is applicable to a wide range of application domain. 

For this first we require domain specific user data which needs to be grouped into different user types. This data will train the Conditional GAN with user type as input.

When the real user provides feedback, a similarity measure is used to find out the user type. Then data is generated from the CGAN and used to train the domain platform. Once the platform is trained and is adapted for the user, the real user interacts. Based on the interaction, new feedback is generated, and the loop continues.