Neuroscientific insights about computer vision models: a concise review.

Area of Science:

• Computer Vision
• Computational Neuroscience
• Artificial Intelligence

Background:

• The study of biologically-inspired computational models began with the artificial neuron in 1943.
• Replicating the complex biological visual system has historically proven challenging.
• Modern computer vision models, while not always directly bio-inspired, embed biological vision principles.

Purpose of the Study:

• To explore principles from visual neuroscience and the biological visual pathway present in contemporary computer vision models.
• To provide insights for developing future bio-inspired computer vision systems.
• To trace the historical evolution of computer vision models from artificial neurons to advanced architectures.

Main Methods:

• Historical literature survey tracing the development of computer vision models.
• Analysis of architectural and functional similarities between biological vision and artificial models.
• Discussion of biologically plausible neural networks and bio-inspired unsupervised learning.

Main Results:

• Contemporary computer vision models, including deep neural networks and vision transformers, exhibit embedded principles of biological vision.
• A historical perspective reveals connections from early artificial neurons to modern deep convolutional neural networks (CNNs) and spiking neural networks (SNNs).
• Biologically plausible neural networks and bio-inspired unsupervised learning are increasingly adapted for computer vision.

Conclusions:

• Understanding the resonance between biological vision and current AI models is crucial for future advancements.
• The historical trajectory shows a gradual integration of biological principles into artificial vision systems.
• Future bio-inspired computer vision models can benefit from insights derived from this survey.