当前位置:网站首页>Scientists used CNN to translate raw information from brain activity

Scientists used CNN to translate raw information from brain activity

2021-09-15 06:07:03 Zhiyuan community

compile / Wenlong

A central goal of neuroscience is to decipher neural codes (neural code), Understand the neural representation of sensory characteristics and behavior , And the calculations that connect them . In deep learning , The neural network inspired by neuroscience has good pattern recognition ability , Even surpassing humans in the performance of some tasks .

therefore , Neuroscientists also want to use neural networks to decode neural activities in the brain .

lately , from UCL The participating international research teams use convolutional neural networks (CNN) Translated raw data from brain activity , Decoded many different behaviors and stimuli from multiple brain regions of different species .

The new method can accelerate the discovery of the relationship between brain activity and behavior , It also paves the way for a closer connection between deep learning and brain science .

The study on 8 month 2 Day to day 「Interpreting wide-band neural activity using convolutional neural networks」 The title is published in 《eLife》 On .

Kavli Chief researcher of the Institute of systems neuroscience Markus Frey say :「 Neuroscientists have been able to record larger and larger data sets from the brain , But understand the information contained in these data —— Read neural code , It's still a problem . in the majority of cases , We don't know what message is being transmitted .」

actually , Identifying the corresponding relationship between neural signals and external stimuli or behaviors has always been the main research method of neuroscience . however , Neural networks may provide a way to accelerate the discovery of new neural representations .

「 We hope to develop a method to automatically analyze many different types of raw neural data , Thus avoiding the need to manually decipher them .」

Based on convolutional neural network, the team proposed a method called DeepInsight Deep learning framework , Can cross stimulus 、 Behavior 、 Brain regions and recording techniques are generalized . Once trained , It can be analyzed to determine the neural coding elements that provide information for a given variable .

Accurate decoding and localization in unprocessed hippocampal records .( source : The paper )

They are for DeepInsight Tested , Explore the neural signals of the open field with mice , It is found that the network can accurately predict the location of animals 、 Head direction and running speed . Even without manual processing , The results are also more accurate than those obtained by traditional analysis .

Discover new representations of neural coding

In the past , Decoding neural activity is not easy , Strong background knowledge of coded variables is required , And the information we know often lacks integrity .

Professor of cell and developmental biology, University College London Caswell Barry say :「 The existing methods omit a lot of potential information in neural records , Because we can only decode the signals we already understand . Our network can access more neural code , To teach us to read some other signals .」

The team found , Their network can recognize new representations of neural codes , This is demonstrated by detecting previously unrecognized head orientation representations , This representation is encoded by interneurons in the hippocampus .

「 We can decode neural data more accurately than before , But the real progress is that the network is not limited by existing knowledge .」

The model can span a variety of recording techniques and brain regions .( source : The paper )

Predictive behavior

Researchers have shown that , Their network can not only predict different types of recorded behavior across brain regions , It can also be used to infer the hand movement of human participants .

Kavli Professor at the Institute of systems neuroscience and the Max Planck Institute of human cognition and brain science Christian Doeller say :「 This method allows us to more accurately predict human higher-level cognitive processes in the future , For example, reasoning and problem solving .」

Frey Add :「 Our framework enables researchers to quickly and automatically analyze unprocessed neural data , To save time , And this time can only be used for the most promising assumptions , Use more traditional methods .」

Thesis link :http://dx.doi.org/10.7554/eLife.66551

Reference Content :https://medicalxpress.com/news/2021-08-team-ai-decode-brain-behavior.html

版权声明
本文为[Zhiyuan community]所创,转载请带上原文链接,感谢
https://chowdera.com/2021/09/20210909121456477v.html

随机推荐