Deep learning systems continue to gain widespread adoption in the enterprise, tackling photo and voice recognition, customer service interactions, and even spotting abnormalities in medical records. But while the artificial intelligence (AI) models, which rely on massive data sets to "train" themselves on recognizing patterns and making predictions—throughout multiple iterations—timing is still an obstacle. Developing an accurate deep learning model can take up to days, or even weeks.
On Tuesday, a new deep learning model developed by IBM Research—Distributed Deep Learning—made big strides in the field: It achieved a record for image recognition accuracy of 33.8%.
The model, which used a massive data set of 7.5 million images, achieved "record communication overhead and 95% scaling efficiency on the Caffe deep learning framework over 256 GPUs in 64 IBM Power systems," according to IBM—all in just seven hours.
For some context, Facebook AI Research could scale at 89% on Caffe2, at higher communication overhead. And Microsoft's accuracy for a similar task was 29.8% accuracy—over a much longer period of 10 days.
According to IBM, this is a "milestone in making Deep Learning much more practical at scale—to train AI models using millions of photos, drawings or even medical images—by dramatically increasing the speed and making significant gains in image recognition accuracy possible as evidenced in IBM's initial results."
The milestone was achieved by addressing the problem of performance bottlenecks, which are the result of fast GPUs trying to synch up simultaneously. According to IBM, "this contention typically causes massive deep learning models on popular open source Deep Learning frameworks to run over days and weeks." To address it, IBM used "dozens of servers connected to hundreds of GPU accelerators popular in gaming systems with near perfect scaling."
Speeding up deep learning training models could prove to come in very useful for businesses by helping create models quicker that could analyze things like medical images, spot fraud, or boost speech recognition.
IBM Research has developed is a beta version of its Deep Learning software for IBM Systems, making it available to PowerAI v4 developers and customers for TensorFlow and Caffe. You can test it out here.
The 3 big takeaways for TechRepublic readers
- IBM Research—Distributed Deep Learning—made a big stride in deep learning this week, achieving a record for image recognition accuracy of 33.8%.
- The achievement beat the previous industry record of 29.9% accuracy in 10 days set by Microsoft.
- This new algorithm could reduce the training time associated with deep learning from days or hours to minutes or seconds, according to IBM.
- Machine learning: The smart person's guide (TechRepublic)
- Google Cloud Machine Learning Engine: The smart person's guide (TechRepublic)
- Google is using machine learning to create a news feed from your searches (ZDNet)
- Facebook's machine learning director shares tips for building a successful AI platform (TechRepublic)
- How to prepare your business to benefit from AI (TechRepublic)
- 3 ways to massively fail with machine learning (and one key to success) (TechRepublic)
- The Complete Machine Learning Bundle (TR Academy)
- IT leader's guide to the future of artificial intelligence (Tech Pro Research)
Hope Reese has nothing to disclose. She doesn't hold investments in the technology companies she covers.
Hope Reese is a journalist in Louisville, KY. Her writing has been featured in The Atlantic, The Boston Globe, The Chicago Tribune, Playboy, Undark Magazine, VICE, Vox, and other publications.