Visual Search as a Probabilistic Sequential Decision Process in Software Autonomous System

Software systems now can replace a vast amount of human resources with their quasi-autonomous features. Adaptive nature helps the system to adjust their behavior in response to their environmental changes. Most of these changes are unknown which put the system under uncertainty while making decisions. Decision-making in the real-time environment requires through observation for searching and analyzing different features of the environment in which the systems are running to understand how humans select relevant information. In this way, a human and a software system can jointly evaluate the situations and make the decision. Earlier visual search (VS) models believe that the eyes move to locations with maximum saliency and every eye movement is planned. It is hard to choose any optimal searching process when the operating characteristics of any visual system vary over time. Furthermore, this method does not account how the brain coordinates its different regions to process information over time. To address these limitations, we propose visual process as an instance of deep reinforcement learning which represents a step towards building autonomous systems to understand the visual world more accurately. In this paper, we begin with different visual search algorithms, including reinforcement learning, followed by probabilistic sequential decision making. In parallel, we focus on various existing areas of related research and their advantages and drawbacks within the domain. Finally, we propose a model which can address the shortcomings of the existing algorithms.