Another tech giant has revealed their know-how AI to the world
. Today, Yahoo has released the CaffeOnSpark AI source code, which can be used or modified by any academic researcher or large company.
Yahoo’s prowess in technology may be less well known. In fact, Open source Hadoop was incubated by Yahoo; Facebook (FB), Twitter (TWTR) and many other companies use this data-processing platform. So When it comes to AI, Yahoo is uniquely positioned because data is just as important as algorithms when it comes to training ai systems. And Yahoo owns one of the world’s most interesting and largest databases: the photo community site Flickr.
CafffeOnSpark, like many new open source AI projects, has its roots in deep learning. Deep learning, a branch of artificial intelligence, is best known for its ability to recognize things like human speech, photos and videos. Yahoo is using this technology to identify different content in photos to improve its search results on Flickr. Unlike other sites, Flickr’s image search isn’t based on description or user input. Yahoo teaches computers how to recognize certain features in photos, such as certain colors or even objects or animals.
In recent months, Google has opened up the source code for its deep learning network TensorFlow, Microsoft has opened up a similar network CNTK, Facebook has shared ai hardware designs, and Chinese search giant Baidu has made its deep learning training software public.
Each of these open source technology projects has its own focus and bias. Yahoo, for example, aims to run deep learning on existing systems and avoid transferring data from one place to another. Andy Fent, Vice president of architecture at Yahoo, explained that training deep learning systems to recognize images requires a huge amount of data. You have to put as many examples as possible in the algorithm, such as a cat, so that the system “learns” to recognize common features between cats and other animals.
There are thousands of images on Flickr, enough to train artificial intelligence. But the team didn’t want to transfer the images from Flickr to another layer of servers running deep learning software. So they developed a way to run deep learning software on top of existing architectures.
According to its name, CaffonSpark combines two existing technologies: Caffe, a deep learning framework, and Spark, a large-scale data processing system. What Yahoo did was find a way to run Caffe at the Spark level. Once found, Caffe can run not only on Spark, but both together on Hadoop. Yahoo’s development will not only make it easier for AI developers to use more familiar tools and avoid the cumbersome process of transferring data, it will also make it easier for deep learning to process content from several servers at once. Feng goes out of his way to tell us that Google’s TensorFlow is not yet there, and That Yahoo is one step ahead.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, Giiso’s research and development products include editing robots, writing robots and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
Feng said that after the team posted about CaffeOnSpark on its blog last year, a number of companies wanted Yahoo to open up the Source code. So they found that companies already had a lot of information on their servers, but they didn’t want to transmit it.