The latest version of Keras, 2.0.9, has been released on GitHub with several RNN improvements, easier multi-GPU data parallelism, and a range of API changes, in addition to added bug fixes and performance improvements such as the native support for NCHW data layout in TensorFlow. Implementation changes in Keras 2.0.9 result in a different scaling and normalization behavior.
Google has unveiled a new interface “eager execution” making it easier to get started with TensorFlow. Eager execution is an imperative, define-by-run interface where operations are executed immediately as they are called from Python. Announcing the release on its official blog, Google said the benefits of eager execution interface includes fast debugging, support for dynamic models, support for custom and high-order gradients, and almost all of the available TensorFlow operations. Google is soliciting feedback on this experimental feature.
BI platform Yellowfin has announced the release of Yellowfin 7.4. While augmented data discovery is a notable feature in the product, the addition of ETL enables the company to add data science platforms such as H2O, incorporating data science components like Predictive Model Markup Language (PMML) and Portable Format for Analytics (PFA). This means Yellowfin is now an end-to-end platform for data scientists.
PASS 2017 Summit in data science news
With new advances to its SQL Server 2017 solution and Azure data services, Microsoft intends to form the ultimate hybrid data platform. The company made new on-premises and cloud announcements at PASS Summit 2017. Among the new tools, Microsoft announced the Azure SQL Database Managed Instance and Azure Database Migration Service that enable users to ‘lift and shift’ on-premises SQL Server workloads. Both services are available in a private preview. To help integrate some of these newly launched tools, Microsoft said it put features like integration with Python and R scripts into SQL Server 2017.
The PASS 2017 summit saw another significant announcement from Microsoft for future: SQL Operations Studio. The studio is a free, lightweight tool for “modern database development and operations on Windows, Mac or Linux machines for SQL Server, Azure SQL Database, and Azure SQL Data Warehouse.” It includes smart T-SQL code snippets, customizable dashboards and support for popular command line tools.
New Blockchain platforms in News
A renowned computer science professor from Cornell University is set to launch a new blockchain project called “thunder token.” Known for her work on the fundamentals of distributed systems, Elaine Shi claimed that thunder token will be able to achieve speeds 1,000x greater than existing technologies, while also addressing the common blockchain problem of scalability. Making the announcement at ethereum’s annual developer conference Devcon3, Shi said the new initiative is based on the thunderella protocol – a paper she co-authored with Cornell associate professor Rafael Pass. In thunder token, the protocol proposes a split set-up so that transactions are confirmed very quickly, with the blockchain only being used in the case of emergencies. The rest of the time, thunder token will use something a little less familiar – a system of agents that follows the direction of a “leader” to vote on which transactions are made according to the rules. It is not yet clear if the protocol will be purely private, or open to the public.
In a new approach towards open source that could benefit both developers as well as organizations, FundRequest has launched a new blockchain platform for the funding, claiming, and rewarding of open source contributions. With FundRequest, users will be given access to a decentralized ecosystem that provides code-enforced guarantees against corruption. Funding is only sent if a project’s functionality can be demonstrated, and will be withheld otherwise. FundRequest’s plugin will allow users to fund open requests on networks like Github with a fund button. After setting up a Github request ticket, users can use the FundRequest interface to fund them which generates a unique smart contract to manage the payment of funds.
AI platforms in News
At an artificial intelligence summit organized in Shenzhen, China, Hikvision unveiled its AI Cloud platform. The company said that Hikvision AI Cloud was developed to solve real world challenges existing in different vertical markets, and to create continuous value to end users. Hu Yangzhong, CEO of Hikvision who addressed the forum, noted how the ongoing trend of engineering AI algorithms into edge devices was making the edge more intelligent. “Edge computing uses local computing to enable analytics at the source of the data. With AI algorithms woven into the edge devices, only selected information such as an individual or a vehicle in a video image will be extracted and sent which significantly enhances the transition efficiency and reduces the network bandwidth, while still sustaining high quality and accuracy,” Hu said.
Drone detection technology developer Dedrone has announced the upgraded version of their software, named DroneTracker 3, which significantly expands the scope of current Dedrone features. DroneTracker 3 includes enhanced updates such as automated summary reporting, improved detection and reliability, enterprise-grade security and management, and an overall simplified set up which is easy to deploy. “Ultimately, DroneTracker 3 identifies how many drones are in an organization’s airspace, a question which was nearly impossible to answer prior to the launch of DroneTracker,” commmented Joerg Lamprecht, CEO and co-founder of Dedrone.