Brain scans aid machines to see more like us

• Autores: Matt Reynolds
• Localización: New scientist, ISSN 0262-4079, Nº. 3119, 2017, pág. 14
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • Artificial intelligence is taking image recognition tips from a real expert: the human brain. David Cox at Harvard University and his group analyzed how regions of the brain's visual cortex responded to images containing four different types of object: humans, animals, buildings and food. The data came from a volunteer who viewed more than 1,200 images while an fMRI machine measured their brain response. The different objects had their own corresponding pattern of brain activity, and the strength of the signals indicated how difficult each image was to classify. The team used this information to train its machine learning algorithms. If an algorithm made a mistake on an "easy" image, it was more heavily penalized than if it made erred on a "difficult" image. This feedback essentially told the system what information it should base its classifications onto minimize errors. As a result, it performed better on images easily recognized by the brain, effectively making decisions in a more human-like way.