A temporal hierarchical feedforward model explains both the time and the accuracy of object recognition

Area of Science:

• Computational Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• Human brain excels at object recognition, with accuracy and speed influenced by stimulus properties.
• Existing computational models explain recognition accuracy but often lack biological plausibility regarding processing time.
• Temporal dynamics of neural processing are crucial but often overlooked in object recognition models.

Purpose of the Study:

• To develop a biologically plausible computational model that accounts for both object recognition accuracy and processing time.
• To investigate the role of temporal feature representation in improving recognition performance.
• To link neural processing dynamics with decision-making mechanisms in object recognition.

Main Methods:

• Modified a hierarchical spiking neural network (spiking HMAX) to represent stimuli temporally via spike trains.
• Coupled the spiking HMAX model with an accumulation-to-bound decision-making model.
• Simulated object recognition tasks and compared model performance with human psychophysical data.

Main Results:

• The proposed model accurately predicts human recognition accuracy and response times in psychophysical tasks.
• Temporal feature representation significantly improved the accuracy of the biologically plausible decision-making model.
• The decision bound mechanism effectively adjusted the speed-accuracy trade-off for different recognition demands.

Conclusions:

• Temporal representation of features is informative and enhances the performance of biologically plausible decision-making systems.
• The developed model offers a novel framework for understanding and simulating object recognition, integrating neural dynamics and decision processes.
• The findings suggest that timing information in neural signals is critical for efficient and accurate cognitive functions like object recognition.