Microsoft claims lead in AI speech recognition: Redmond's research group has managed to reach human parity for speech recognition, edging out IBM, in the latest stage of one of the more active battlefronts in AI development. Microsoft used the Switchboard data set, a large collection of recorded telephone conversations that's a standard tool in the speech recognition community's toolbox, to achieve its results, according to a blog post:
After our transcription system reached the 5.9 percent word error rate that we had measured for humans, other researchers conducted their own study, employing a more involved multi-transcriber process, which yielded a 5.1 human parity word error rate. This was consistent with prior research that showed that humans achieve higher levels of agreement on the precise words spoken as they expend more care and effort. Today, I’m excited to announce that our research team reached that 5.1 percent error rate with our speech recognition system, a new industry milestone, substantially surpassing the accuracy we achieved last year. A technical report published this weekend documents the details of our system.
Last year, Microsoft said it had managed a 6.3 percent error rate, but in March IBM said it had achieved a 5.5 percent rate. Microsoft attributed its latest gains to improvements in its neural net-based acoustic and language models, its Cognitive Toolkit 2.1, and GPU-powered infrastructure on Azure.
POV: Constellation expects speech recognition to play an increasingly important role not only in consumer but enterprise contexts, as a means of interaction with applications.
But it's important to place Microsoft's achievement in the proper context, which the company notably does itself in the blog. The Switchboard test only goes so far—challenges remain in training systems to understand speech in noisy environments, or accented speech, for example.
Moreover, Microsoft has "much work to do in teaching computers not just to transcribe the words spoken, but also to understand their meaning and intent," the blog notes. "Moving from recognizing to understanding speech is the next major frontier for speech technology."
IBM targets food safety with its blockchain tech: Big Blue has formed a pact with some of the world's largest food sellers and producers around using blockchain to promote food safety. It's the latest move by IBM to apply blockchain in verticals beyond its origins in finance. (Note: While blockchain is the term most commonly used in the industry, Constellation believes that synchronous ledger technology may be a more appropriate term. Go here for Constellation VP and principal analyst Steve Wilson's deep-dive on the topic.
The companies working with IBM on the new initiative are Dole, Driscoll’s, Golden State Foods, Kroger, McCormick and Company, McLane Company, Nestlé, Tyson Foods, Unilever and Walmart, according to a statement.
Contaminated food is responsible for 420,000 deaths per year, according to the World Health Organization, and it can be very difficult and time-consuming to trace the source of the problem. That's where blockchain comes in, IBM says:
Blockchain is ideally suited to help address these challenges because it establishes a trusted environment for all transactions. In the case of the global food supply chain, all participants - growers, suppliers, processors, distributors, retailers, regulators and consumers - can gain permissioned access to known and trusted information regarding the origin and state of food for their transactions. This can enable food providers and other members of the ecosystem to use a blockchain network to trace contaminated product to its source in a short amount of time to ensure safe removal from store shelves and stem the spread of illnesses.
POV: IBM also announced the general availability of its Blockchain V1 platform, which builds on the work done by the open-source Hyperledger project. Big Blue is one of Hyperledger's most prominent contributors.
As for the food safety announcement, IBM has put together a very impressive list of launch partners to say the least, as all are major players in the increasingly complex global food supply chain.
But that complexity is related to more than logistical matters; traceability in the food supply chain isn't just about figuring out the source of contamination after the fact. Factors such as food fraud—a problem that costs producers billions per year—and consumer demand for concrete provenance of the food they purchase are just two other key factors. If blockchain's principal underlying value proposition is trust, food safety is an excellent area for IBM to target its resources.
Databricks, Skytap land sizable funding rounds for big data analytics and specialized cloud: Investors are opening up their pocketbooks for a couple of the hotter enterprise startups. First up, big data analytics vendor Databricks has landed $140 million in a Series D funding round led by Andreesen Horowitz. That brings its total funding to date—it was founded in 2013—to $247 million.
Meanwhile, Skytap has received a $45 million investment from Goldman Sachs. The Seattle startup's product is a cloud infrastructure service aimed at migrating legacy on-premises applications to the cloud quickly by essentially containerizing the customer's existing deployment model, from networking to storage to middleware and so on.
POV: Databricks' funding round is a substantial one, says Constellation VP and principal analyst Doug Henschen. The company was founded by the creators of Apache Spark and is a leader of that community, but is far from the only one provided cloud-based Spark services, he adds. Spark has the most active open-source big data community, with more than 1,000 contributors in 2015.
"But Databricks is holding the proverbial tiger by the tail," Henschen says. While the company says it has more than 500 organizations using its hosted Spark service, it's likely that the bulk of mainstream Spark adoption is being driven by Amazon first and foremost, as well as IBM, Microsoft, Google and others, he adds.
Databricks touts the fact that it tends to be the first to deliver new Apache Spark releases and features, but a key appeal of broader public cloud providers of Spark services is their ability to offer extensive infrastructure and developer services, as well as Hadoop and database services along with Spark's analytical capabilities, Henschen says: "Blending services from multiple cloud providers isn't difficult, but I think the broad market will be attracted to cloud providers that can offer more of a one-stop shop for data management, big data analysis and cloud application development services."
As for Skytap, "there is plenty of opportunity in moving to the public cloud and it's good to see startups going after that opportunity, not just the large IT titans," says Constellation VP and principal analyst Holger Mueller. Goldman's backing is of interest as well, since it shows the overall investment opportunity in the cloud infrastructure market, he adds.
