Speed versus accuracy in visual search: Optimal performance and neural architecture

Area of Science:

• Computational neuroscience
• Visual perception
• Machine learning

Background:

• Visual search involves balancing speed and accuracy, crucial for survival.
• Understanding how the brain achieves this trade-off is a key challenge.
• Spiking neural networks offer a biologically plausible framework for complex computations.

Purpose of the Study:

• To develop an optimal model for visual search that balances speed and accuracy.
• To investigate the neural architecture capable of implementing such a model.
• To compare model predictions with psychophysical data.

Main Methods:

• Proposed an ideal observer model for visual search using V1-type orientation-selective spiking neurons.
• Incorporated signal-to-noise ratio, error costs, and time costs as parameters.
• Modeled local and global gain-control mechanisms for scene complexity.
• Developed a five-layer spiking neural network to approximate the optimal model.

Main Results:

• The model accurately predicts error rates and response times across varying stimulus discriminability and scene complexity.
• The five-layer spiking network closely approximates the optimal model's performance.
• Demonstrated that known cortical mechanisms can support efficient visual search.

Conclusions:

• The proposed model provides a framework for understanding optimal speed-accuracy trade-offs in visual search.
• A biologically plausible spiking neural network can implement efficient visual search.
• Cortical mechanisms are sufficient for efficient visual search, aligning computational models with neural architecture.