Baidu open sources its Mobile deep learning, IBM’s HPC research and more in today’s data science news.
Open source announcements in News
Baidu has open sourced Mobile Deep Learning (MDL), a convolution-based neural network customized for mobile devices. MDL can identify objects in an image (taken from smartphone camera) in fractions of second, and give the suggestion to Baidu to carry forward the search process. Coming at faster speed and reduced complexity, MDL supports both iOS and Android, though it may run better on Apple. The codes are available now at Github taking nearly 4 MB space. Last year, Baidu had open sourced PaddlePaddle deep learning package, and developers suggest PaddlePaddle will be best model to use with MDL.
Google has open sourced Abseil, a set of libraries from the very building blocks of its internal codebase. “These libraries are the nuts-and-bolts that underpin almost everything that Google runs,” the company said, adding that Abseil was developed over the last decade to support important projects like gRPC, Protocol Buffers, and TensorFlow. Abseil includes C++ and Python utilities. While the C++ libraries are now available on GitHub under Apache license, Google will soon make available a Python version of the library.
In Other Data Science News
Researchers at IBM’s Dublin research facility claim to have developed a deep learning model that could advance high-performance computing (HPC) by 12,000 percent. Using available conditions of wave, ocean currents and winds, the framework can help in forecasting wave conditions at real time. The research indicates that simulations can be done on lower-end computing devices like Raspberry Pi, and it does not have to require HPC infrastructure. The deep learning model can also be utilized to make the running HPC infrastructure train smartphones or other cheaper computing devices.
Canada’s largest bank, the Royal Bank of Canada (RBC), is trialing using blockchain technology for payments to and from the United States. It allows the bank to explore the potential of the tech without fully replacing the existing system. “We wanted to set it up as a shadow ledger so that we can demonstrate our leadership in exploiting that technology while at the same time recognizing that the technology is still early in its adoption phase,” RBC’s executive vice president Martin Wildberger said, adding that while the technology could prove “transformative and critical,” it still needs more time to mature.
MapR Technologies has enhanced the scope of its database MapR-DB to drive real-time analytics. The company announced that its latest database version expands the scope for self-service SQL data exploration with enhanced Drill integration, and also supports connectors to native Spark and Hive for real-time processing. The new MapR-DB version also aids real-time application integration with global data capture.