PARIS -- AI Net -- Your average data scientist apparently spends about 80% of his or her time on managing data, leaving few hours in the ordinary working day for anything more exciting. Fortunately, the wages are supposed to be good. But there is concern within telcos about the management burden and the unnecessary replication of industry efforts. Initiatives that allow operators to pool resources could help.
Step up Acumos. Launched officially in May last year, the open source initiative is the first project of the Deep Learning Foundation (DFL), itself a part of the Linux Foundation. Perhaps best described as an open marketplace for artificial intelligence (AI) applications, it has been welcomed by industry analysts as a potential catalyst for AI development in the telecom sector. "Third parties can publish machine learning algorithms and service providers -- as well as companies from other industries -- can 'try before they buy' on their own datasets," said James Crawshaw, a senior analyst with Heavy Reading (a sister company to Light Reading), during a discussion about Acumos last year.
France's Orange is one of several operator members now trying to persuade other telcos of the benefits. Philippe Dooze, a data analyst with Orange Labs, the French operator's research division, says it provides a framework to help data scientists with their daily work. "It eases the creation and update of models, as well as their storage and lifecycle management," he said during AI Net, an industry conference held in Paris this week. Importantly, Acumos has also been designed to aid the sharing and deployment of models.
The membership list remains relatively small, however, even if it does feature some big telecom organizations. On the service provider side, AT&T is the only member besides Orange. From the vendor community, Acumos advocates can reel off a much longer list of names, including Amdocs, B.Yond (a self-described AI specialist), Ciena, DiDi (China's Uber equivalent), Ericsson, Huawei, Intel, Nokia, Red Hat, Tech Mahindra, Univa and ZTE. Chinese web giants Baidu and Tencent are also involved. But Acumos looks desperately in need of more telecom operator members.
A forthcoming release might spur interest. Athena, the current release, includes the actual marketplace as well as an online "design studio," where developers can build models based on what already exists. Athena also provides support for microservice creation, along with the Java, Python and R programming languages. In June, the Linux Foundation will release Boreas (the name of another Greek god), offering support for license management, new types of model and training within Acumos. A third release named Clio, after the muse of history in ancient Greece, is planned for the end of the year.
Some telcos may still be nervous about joining initiatives geared toward the crowdsourcing of machine learning algorithms. Daniel Bar-Lev, a senior figure at the MEF, says his own industry association ran into resistance from telco legal teams when it tried to gather datasets from operators for an AI proof of concept. Even if telcos are not pooling the datasets on which Acumos algorithms would feed, the lawyers might fret about anything related. The risk of malicious code in models could be another "handbrake" on development, writes Crawshaw in a detailed blog about Acumos.
More prosaically, operators may feel they cannot commit overworked data scientists -- a much sought-after commodity in the industry these days -- to an open source project outside the business. The paradox is that Acumos should ultimately free up data scientists for more valuable activities. It is the classic chicken-and-egg situation: Until operators see evidence of the benefits, they might hold off on joining; yet the benefits may never materialize unless membership grows.
The links between Acumos and ONAP, the Linux Foundation's open source project focused on the orchestration and automation of virtualized networks, could also be an issue for the likes of Spain's Telefónica, which has thrown its weight behind Open Source MANO, a rival initiative in the network management and orchestration area. Dooze plays down suggestions that Acumos and ONAP will be more closely integrated in the Boreas release, but he says this could happen when Clio turns up. "Perhaps for the end of the year something will be created between Acumos and ONAP, but I cannot say more on that," he told conference attendees in response to questions. It is also notable that AT&T and Orange, the only two operator members of Acumos, were the first two operators to join ONAP.
All that said, the Boreas updates should go some way toward stimulating interest, and it cannot hurt to have a telco data scientist evangelizing for the project. "Thanks to Acumos, you will be able to help your data scientists in their daily works, and federate them around shared processes," said Dooze in Paris. As Crawshaw writes in his blog: "It would be nice to see more service providers becoming DFL members."
- South Korea Stutters Towards a National 5G Launch on Friday
- The Acumos AI Project Is On Track but Lacking 'Apps'
- Snakes on a Control Plane Scare Telefónica
— Iain Morris, International Editor, Light Reading