Google develops a new AI algorithm that maps images of the brain's nervous system July 18, 2018 Source: ZDNet reports Author: little NetEase Technology News July 17th, according to VentureBeat report, drawing a bio-network structure diagram in the nervous system (this field is called connectomics) requires a lot of calculations. There are approximately 86 billion neurons in the human brain that are connected by 100 trillion synapses, and imaging one cubic millimeter of tissue can generate more than 1000 terabytes of data. Fortunately, artificial intelligence (AI) can help. In a paper published in the journal Nature Methods, scientists at the Google and Max Planck Institute for Neurobiology demonstrated a recursive neural network, a system tailored for connected omics analysis. Recurrent neural networks are machine learning algorithms that are often used for handwriting and speech recognition. Google researchers are not the first to apply machine learning to connected groups. In March of this year, Intel collaborated with the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Lab to develop the "next generation" brain image processing pipeline. But Google claims that their model accuracy is "order-level" higher than previous deep learning techniques. Figure: Google algorithm tracks 3D neurites in the songbird brain Google researchers use an edge detection algorithm that identifies the boundaries of neurites (the branches of the neuron ontology) and a recurrent convolutional neural network (a subclass of recurrent neural networks) that neural network Pixels in the neuron scan are gathered and highlighted. To track accuracy, the research team developed the "expected run length" (ERL), which is a random point for random neurons in the 3D image of the brain, and then the measurement algorithm tracks the distance of the neuron before the error. The team reported that the model performed "better than the previous algorithm" after a brain scan of a 1 million cubic micron zebra finch. Google researcher and paper authors Viren Jain and Michal Januszewski said: "The results of these automations combined with a small amount of extra manpower can help solve the remaining errors. Researchers at the Max Planck Institute can now study the connective groups of the songbird brain to get new insights into how the zebra finch sing and test how they learn the theory of singing." In addition to the papers, the research team also released the model's TensorFlow code on Github, as well as the WebGL 3D software they used to visualize data sets and improve refactoring results. They plan to continue to improve the system in the future with the goal of fully automating the synaptic resolution process and “contributing to the projects of the Max Planck Institute and other institutionsâ€. Notebook Ultrasound Scanner,B W Ultrasound Machine,Ultrasound Bone Densitometer,Ultrasound Machine Instrument Mianyang United Ultrasound Electronics Co., Ltd , https://www.uniultrasonic.com
Google develops a new AI algorithm that maps images of the brain's nervous system