Distracted by Previous Experience: Integrating Selection History, Current Task Demands and Saliency in an Algorithmic...

Area of Science:

• Cognitive psychology
• Computational neuroscience
• Visual attention research

Background:

• Attention is influenced by prior learning and experience, leading to selection history bias.
• Existing models often focus on stimulus-driven or goal-driven control, but the interplay with learned biases is complex.

Purpose of the Study:

• To develop an algorithmic-level model predicting how stimulus-driven saliency, goal-driven control, and selection history interact to guide visual attention.
• To quantitatively assess the contribution of selection history to attentional bias using a novel computational approach.

Main Methods:

• Developed an algorithmic model integrating saliency maps, a history map encoding learning, and task control parameters.
• Utilized coded features instead of image pixels for model input.
• Tested the model against reaction time (RT) data from a psychophysical experiment.

Main Results:

• The model accurately predicted reaction time distribution parameters using an integrated priority map.
• Analysis of map weights confirmed the significant influence of selection history on attention guidance.
• The model captured individual differences in RTs and response probabilities, outperforming reduced models.

Conclusions:

• Integrating stimulus saliency, task goals, and selection history is crucial for quantitatively describing human visual attention.
• Selection history demonstrably biases attention, and its effects can be computationally modeled.
• Adding intertrial effects further improved the model's predictive accuracy, highlighting lingering biases.