Supporting the design of data visualisation for the visually impaired through reinforcement learning

Abstract

The aim of the research is to present a possible approach to help visually impaired people to make decisions while interacting with a data visualisation task. The main goal is to build a Machine Learning model (i.e. Reinforcement Learning) that can predict the visual behaviour of visually impaired people when they interact with a data visualisation. This work concerns partial vision, where the damage occurred in either the peripheral vision such as in the Tunnel vision (e.g. due to Glaucoma), or in the central vision (e.g. in age-related Macular Degeneration). Getting the desired results would help in designing accessible visualisation tasks which will assist the decision-making process for the relevant users. Initially, the model will be iteratively evaluated on existing visual search tasks from the literature; the tasks will consist of both visual impairment-and normal vision-related tasks. Once the model is tested, a new visualisation task that is suitable for the visually impaired will be designed and evaluated on human participants in order to help the cycle of design, development and testing, with the ultimate goal of supporting and transforming user experience for the visual impaired. The model will be iteratively refined using more advanced methods such as Deep Reinforcement Learning (DRL). Furthermore, rational analysis framework will inform the building of the model, as it uses rationality as an empirical tool to explain how and why people adapt to their environment.