20th Dec.’ 17 – HeadlinesAlok Dhuri
Kibana’s new version, Twitter’s enterprise API, Google’s speech generation tool Tacotron 2, and a new model selection system Auto-Tuned Models (ATM) among today’s top stories in artificial intelligence and data science news.
Kibana 6.1.1 released
Open-source data visualization tool Kibana has released its version 6.1.1 with a reported fix for a high severity security vulnerability in the Time Series Visual Builder. According to the official release announcement, all administrators of Kibana 6.1.0 are asked to upgrade immediately to 6.1.1, while the versions prior to 6.1.0 will not be affected. “If you had any Kibana 6.1.0 instances on Elastic Cloud, we’ve automatically upgraded them, so no further action is required,” Tech Lead Court Ewing said in the release announcement. “For folks that cannot upgrade from 6.1.0 at this time, you can disable time series visual builder entirely by specifying
metrics.enabled: false in kibana.yml and restarting Kibana. Note, this will require a full ‘optimize’ run, which can take a few minutes.”
Just to recap, Kibana 6.1.0 had introduced a new feature for “math aggregations” in the Time Series Visual Builder which allowed users to apply mathematical operations to their TSVB results. Later it was found that the new feature had a vulnerability that could allow an attacker to execute arbitrary code on the Kibana server. Gauging the severity of the issue, Kibana decided to do away with this feature. “We do want to have this sort of math capability in Kibana at some point, but we need to take a more holistic view on its security before releasing it again,” the release added. For complete details about all other bug fixes in this release, refer to the release notes.
Twitter promised to streamline relations with developer community, and they have made a start!
Earlier this year, Twitter had unveiled its long-term vision to revamp and streamline its API platform leveraging its investment in Gnip which it acquired in 2014. As part of that broader plan, Twitter has announced a new enterprise-level API to provide access to real-time activities like tweets, retweets, likes and follows. To be specific, the API is designed to help developers build apps that can power customer service, chatbots and brand engagement on Twitter. Alongside this launch, Twitter is launching a suite of developer tools for Direct Messages out of beta. These features include Quick Replies, Welcome Messages, Buttons on messages, Custom Profiles, and Customer Feedback Cards. Brands like Samsung, MTV, TBS, Wendy’s, and Patrón have used these tools with their chatbots. Whereas Tesco and Evernote have been using these previously announced tools for customer service.
Generating Human-like Speech from Text
Google has unveiled a new method for training a neural network to produce realistic speech from text that requires almost no grammatical expertise. Named Tacotron 2, the new technique combines the best of Google’s previous projects on speech generation: WaveNet and Tacotron. It uses text and the corresponding narration to calculate all the linguistic rules that are specified to the systems, and while the text is converted into the original Tacotron-style “mel-scale spectrogram” for rhythm and emphasis, the words are generated using a WaveNet-style system. Google researchers have submitted the project for consideration at the IEEE International Conference on Acoustics, Speech and Signal Processing. The full paper is available at arXiv.
Auto-tuning data science
Auto-Tuned Models (ATM): Researchers propose automated machine learning technique for model selection in data science
A new paper called “ATM: A distributed, collaborative, scalable system for automated machine learning” was presented recently at the IEEE International Conference on Big Data, where researchers from MIT and Michigan State University proposed a new system that automates the model selection step in data science. The system, called Auto-Tuned Models (ATM), takes advantage of cloud-based computing to perform a high-throughput search over modeling options, and find the best possible modeling technique for a particular problem.
ATM tests thousands of models in parallel, evaluates each, and allocates more computational resources to those techniques that show promise. Poor solutions wither out in the process, while the best options rise to the top. So rather than blindly choosing the “best” one and providing it to the user, ATM displays results as a distribution, allowing for comparison of different methods side-by-side.
Researchers have open sourced ATM, including provisions to add new model selection techniques and improve on the platform. ATM can run on a single machine, local computing clusters, or on-demand clusters in the cloud, and can work with multiple data sets and multiple users simultaneously.
Soon, ‘smart’ robots will taste and smell!
Rewired and Aromyx have entered partnership to digitize and explore novel applications of taste and smell for smart robots. While Rewired is a venture studio for robotics development, Aromyx is the maker of EssenceChip, the digital platform for measuring taste and scent.
EssenceChip is a disposable biosensor that places the human taste and olfactory receptors into a biochip. This technology is targeted for use by the food and beverage, flavor and fragrance, consumer packaged goods, chemical and agricultural industries.
Rewired invests in machine perception technologies that will “unlock the next generation of smart robotics.” The studio focuses on the sensors, software, and systems that help autonomous machines interact with unpredictable environments and collaborate with humans. “The next generation of machines must be able to gather diverse data about their surroundings and holistically interpret that data in order to model the world and productively interact with it,” said Andy Hickl, Venture Partner at Rewired. “Applying Aromyx’s technology to robotics will help us detect and capture new modalities of data, that will inform decision making in autonomous machines and inspire the development of new learning models.”