z-bo 2020-11-04 284

　　Recently, researchers from the Neurocomputing and Brain-Computer Interaction Team of the Brain-Inspired Intelligence Research Center, Institute of Automation, Chinese Academy of Sciences, proposed a structured neural decoding model to achieve high-quality natural images, human faces and other complex visual stimuli based on brain activity patterns. reconstruction. 

　　For a long time, philosophers and scientists have tried to speculate, observe, understand and decipher how the brain works so that people can perceive and explore the natural world. Among them, as an important information channel for humans to perceive the world, the processing mechanism of the human brain visual system has attracted the attention of researchers. Many studies have attempted to use neural information coding and decoding methods, that is, to reveal the mechanism of brain visual information processing by constructing a quantitative relationship between external stimuli and neural activities. In the early stage, researchers proposed a method of visual nerve information decoding based on Bayesian deep learning (TNNLS 2018), and can reconstruct the simple visual stimuli perceived by the subjects (such as handwritten numbers, letters) based on the recorded brain activity signals And other patterns). However, the reconstruction of complex natural visual stimuli is a difficult problem. In response to this problem, researchers propose a structured neural information decoding method. This method reveals the relationship between multiple typical computer vision models (such as VGG, ResNet) and the human brain ventral visual pathways in terms of hierarchical feature expression through multi-task feature decoding. By using the relationship between this hierarchical feature and the human brain visual cortex signal expression, the new method can reconstruct the complex natural images and facial stimuli perceived by the subjects based on the small amount of human brain fMRI data collected. This research provides a new perspective for understanding the decoding process of the brain and promotes the advancement of non-invasive brain-computer interface technology. Neural information coding and decoding is a core research problem in the field of brain-computer interface, and it is also an effective way to explore the principles behind the complex functions of the human brain to promote the development of brain-like intelligence.