cool hit counter Stanford University: predicting the demographic composition of US communities with the help of deep learning and Google Street View_Intefrankly

Stanford University: predicting the demographic composition of US communities with the help of deep learning and Google Street View

Market researchers and political analysts have been studying these things for decades in response to whether people in Democratic and Republican constituencies prefer sedans or pickup trucks. Recently, though, Stanford researchers with an ambitious project - analyzing 50 million photos and geolocation data on Google Street View - have come to the same conclusion. With the help of newly developed artificial intelligence techniques, researchers are able to analyze large amounts of images, extract data that can be sorted and mined to predict things like income levels, political leanings, shopping habits, etc. in a given community.

In this Stanford University study, the computer collected millions of images of cars with information such as the manufacturer and specific model.

Erez Lieberman Aiden, a computer scientist at Baylor College of Medicine's Center for Genome Research who advised on the study, noted that "in a flash, we can do the same textual analysis of the images."

Mr. Lieberman Aiden said that computers, like humans, can understand the world in two very different ways, by reading and by looking. In this sense, 'the hands of the computer that were tied behind its back have been released'.

For AI, text is easier information to process because English words are discrete characters made up of 26 letters. This brings it closer to the natural language of a computer, rather than being confronted with a mess of images.

In recent years, image recognition technology, developed under the leadership of large technology companies, has seen great advances. And this study from Stanford University gives us a glimpse of the potential in this area.

By extracting information such as vehicle make, model and year from the images and then linking it to other data sources, the project was able to predict many "interesting facts", such as neighborhood pollution and voting patterns.

"The use of image data will give rise to a new set of tools for social analysis," said research leader Timnit Gebru. Details about this study have been published in stages.

For example, as recently as November, they published an article in the Journal of the National Academy of Sciences titled "Predicting the Demographic Composition of Communities Across the United States with the Help of Deep Learning and Google Street View".

With the full paper.

The PDF version will be shared to the 199IT premium exchange group, thanks for your support!

High-end knowledge exchange group now 2500+ users, the most valuable sharing, the most professional discussion.

199IT's top reports are highlighted and can be accessed directly by clicking on.


The WeChat public platform has now introduced a subscription number topping feature

Click on our home page

Capture the excitement in time

1、Thats an amazing amount of hair for a Tsinghua postdoc at an internet company
2、Python implementation of WeChat jumping high score tips
3、2017 JavaScript Framework Battle Report React Sub Battlefield
4、Too painful to do Python development on Windows Docker to understand
5、Were speaking out in Malaysia and we invite you to join us

    已推荐到看一看 和朋友分享想法
    最多200字,当前共 发送