Neural habituation: A unified account of visual identification dynamics across tasks

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

People often fail to see things that appear right before their eyes. This can occur in many different situations and for many different types of information, ranging from failing to identify where or when something appeared to not noticing that it appeared it all. This project aims to explain failures of detection across a wide range of situations using a unified theoretical framework. Understanding the mechanisms of visual detection failure and its role in normal visual processing is important for many societal problems that involve continuously monitoring visual information, including forensic applications. For example, accurate screening of baggage at security checkpoints or monitoring security video requires an understanding of when human detection is likely to fail. Also, designing successful artificial vision systems requires an understanding of why some types of detection failure are actually necessary for the processing of continuous visual information. The brain's mechanism for separating what is different from what is the same in the constant stream of visual information may come at a cost in some situations. This is relevant to designing real-time computer vision systems that can process ongoing visual information accurately. The theory under investigation in this project is that visual detection failures are often due to neural habituation, which occurs when a neural representation within a network becomes temporarily over-activated after prolonged visual exposure. During the habituation period, the neural representation is temporarily less accessible, which can lead to a failure to consciously notice or detect that particular piece of information if it appears again during the habituation period. The hypothesis tested is that the brain uses transient neural habituation to track information that is constant versus changing. For example, by becoming habituated to recently viewed faces, the brain avoids confusing the attributes of those faces with a currently viewed face, helping to tell people apart. The cost is that people often fail to detect recently seen attributes if they re-appear during the habituation period. To probe visual detection and detection failures, scientists often use rapid serial visual presentation tasks, in which multiple visual images are presented in quick succession. This project aims to explain detection failure across a wide range of such tasks with a unified theoretical model of neural habituation. This effort will increase current understanding of how processes occurring at the level of neural networks can explain processes occurring at the level of visual detection and awareness.