Decoding Attention with Deep Learning Networks
Abstract
Skilled performance in challenging workplace settings often deeply depends on an operator~s situational awareness. However, in almost all dynamic multitask work environments, the total demands of the component tasks greatly exceeds the individual~s attentional capacity at a given moment. This necessitates allocating attention sequentially to component tasks, with frequent shifts between sub-tasks likely being critical for successful performance. In the current proposal, we seek to develop a state-of-the-art Artificial Intelligence modelling approach to decode the current focus of an individual~s attention during a task on the basis of EEG recordings. That is, rather than rely on observable behaviors such as eye movements to infer where and when attention was allocated in the field, our project seeks to ~read out~ from brain activity where attention is directed in the time periods before any observable behavior is made. This approach allows us to more deeply characterize the ~attentional fixation patterns~ that underlie skilled task performance, even in situations where external markers are ambiguous (e.g., eye position is hard to interpret when attention may be divided to two items simultaneously). We have previously been successful at decoding covert attention deployment by applying modern machine learning models to the Alpha band (8-12hz) of EEG. However, there is likely much more information in other frequencies as well as in phase/frequency coupling that is not being exploited by our current model. Therefore, we will develop a novel Deep learning neural net approach to decoding the spatial focus of attention that can learn and extract a much wider set of features present in the EEG signal to potentially greatly improve our ability to accurately and precisely decode the position of attention. A secondary goal of the project is to use this AI-based Attention decoding to characterize and evaluate the attentional fixation patterns that may help differentiate experts from novices in a simulated multitasking work environment.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 17, 2018
- Source ID
- N000141912012
Entities
People
- Edward K Vogel
Organizations
- Office of Naval Research
- United States Navy
- University of Chicago