Scientists Use Biological Data to Improve Computer Vision
Computer vision, the ability of artificial intelligence to analyze visual information, has made significant advancements thanks to scientists from the Massachusetts Institute of Technology (MIT) and IBM. They have discovered a method to enhance computer vision by mimicking the functioning of a real living brain.
The foundation of artificial intelligence is neural networks, mathematical models consisting of interconnected “neurons” that process information. While these models resemble the neural networks in the human brain responsible for visual perception, they are far less sophisticated.
Researchers have determined that when neural networks effectively solve computer vision problems, they form strong neural chains that function similarly to those in the brain. However, computer systems are still far from reaching the capabilities of human vision.
James Dicarlo, a Professor at MIT, proposed that incorporating specific characteristics of the human brain into these models could improve computer vision. To test this idea, Dicarlo and his team utilized data from neurons in the lower temporal cortex of monkey brains that were collected while the primates were shown different images.
By training a neural network to imitate the behavior of these neurons during the recognition of objects in images, the researchers created an artificial model. They then compared its performance with a similar neural network model that was trained without the biological data.
The results showed that the new model aligned more closely with the neural activity in the lower temporal cortex and exhibited behavior more similar to human vision. Additionally, the model demonstrated greater resistance to “adversary attacks,” which are small distortions in images that typically deceive neural networks.
This study reinforces the idea that the exchange of ideas between neurobiology and computer science can lead to progress in both fields. According to Dicarlo, “Everyone benefits from the exciting cycle of interaction between natural and artificial intelligence.”