The visionaries at Google have set their sights on teaching their computers to see.
In an effort to bolster its already-impressive image and item recognition efforts, the search giant has acquired Moodstocks, a French startup that specializes in machine learning-based image recognition technology for smartphones.
"Our dream has been to give eyes to machines by turning cameras into smart sensors able to make sense of their surroundings," Moodstocks wrote in a statement.
Google's purchase of Moodstocks is the latest in a string of "acqui-hires" -- or talent acquisitions -- involving machine learning-based image recognition, a core component of the computer vision field. It follows Amazon's purchase of Orbeus earlier this year and Twitter's purchase of Magic Pony Technology just a few weeks ago, signaling what industry watchers describe as the race by technology pacesetters to commercialize the next computing frontier.
"Tech firms have invested a great deal of time and money to understand text on the internet," Mike Gualtieri, principal analyst at Forrester Research, wrote in an email. "It is only natural that they now want to understand images. The possibilities to arrive at meaning through the combination of text and images will start to provide AI not just with text (speech) but also eyes with image recognition."
Google, for example, has long been rumored to be working on a feature that allows users to search directly from their photos, using only an image as a search term. Google hopes tapping Moodstocks' talent -- which will join Google's existing research and development team in Paris -- will help it further advance in the computer vision field.
"We have made great strides in terms of visual recognition," wrote Vincent Simonet, head of Google's France research and development center, "but there is still much to do in this area."
Big data ushers in new era in image recognition
Google Photos -- as well as Facebook, Flickr and other companies that deal with the sorting and tagging of images -- already uses advanced image recognition technology behind the scenes, but "these approaches could still be faster, more performant and require fewer resources to train," agreed Tomasz Malisiewicz, Ph.D., lead software engineer in deep learning at Magic Leap Inc., a Florida startup that works on augmented reality technology.
"In this new deep-learning era of image recognition, the entire notion of a computer vision team is being reworked," Malisiewicz said. "When algorithm development was much more manual, having a team of Ph.D.s to lead the effort was important. But with the commoditization of deep-learning software, most developers can now train recognition models using this year's favorite neural network architecture."
What is critical in this new era in image recognition is data -- lots of it.
"The focus has shifted from algorithm design to large-scale data management, since the best-performing models require a large amount of data," Malisiewicz said.
Recreating the visual world
What's in it for companies other than the titans of big data? Analysts point out that the use cases for machine learning-based image recognition go well beyond simple categorization, such as finding personal family photos or narrowing down products in a shopping catalog.
The technology has the potential to transform healthcare, for example, by helping doctors diagnose diseases from medical images. In addition, augmented and virtual reality wearables that need to use cameras to sense the world around them can benefit from image recognition and improved tracking capabilities, Malisiewicz said.
Computer vision seeks mathematical models and computer algorithms that can automatically extract information from images and videos.
- Jason Corso Ph.D., associate professor of electrical engineering and computer science, University of Michigan
Machine learning-based image recognition will help enterprises automate operational functions like surveillance, field inspections and anything else that involves continuous or periodic inspections, Forrester's Gualtieri said.
The actionable data that results from advancements in computer vision opens up new potential revenue opportunities for enterprises, according to Matthew Zeiler, founder and CEO at Clarifai, an AI company that specializes in visual recognition.
"They're also able to glean insights from previously untapped information sources, like images and video that have no descriptions, metadata or hashtags," he added.
Other benefits are sure to come. Computer vision is still a relatively young field and exploring it is vital, said Jason Corso, Ph.D., associate professor of electrical engineering and computer science at the University of Michigan.
"It is critical to note that these advances and business cases barely scratch the surface of the potential market and problem space. Ours is a visual world; harnessing its richness is a critical horizon for future technological advances," he said.
CIO news roundup for week of July 4
Google's continued foray into computer vision wasn't the only big story. Here's what else grabbed headlines:
- Turner exits Microsoft. CEO Satya Nadella announced in an email on Thursday that long-time COO Kevin Turner is leaving the company at the end of the month. Turner, who has been with Microsoft for 11 years and was seen as a prospective candidate to replace former CEO Steve Ballmer, is set to join Chicago-based market maker Citadel Securities as its CEO in September. Turner's responsibilities, which touched on everything from sales and marketing to product development, will be divided among five executives, according to the email. "Kevin's departure is a mutual win-win," R. Ray Wang, principal analyst at Constellation Research, told TechCrunch: Turner leaves for a CEO position and Nadella continues to put his "stamp on the new Microsoft," he said. Turner's exit comes a year or so after the departure of four other top Microsoft executives in what was described then as a "top-level shake-up."
- Forecast calls for more cloud. According to a Gartner report, cloud computing is fast becoming "the default vehicle for software deployment." The no-cloud stance of some companies will by 2020 be as rare as a "no-internet" policy is today. That doesn't mean everything gets ported to the public cloud. Hybrid cloud, a mix of public, private and on-premises services, will likely be the most common usage of the cloud, Gartner said. The consultancy also predicted that by 2019, more than 30% of the 100 largest vendors' new software investments will have shifted from cloud-first to cloud-only.
- Where's the future of autonomous cars headed? News of a second crash involving a Tesla vehicle injected more uncertainty into that question. The National Highway Traffic Safety Administration is investigating the crash involving a Tesla Model X, in Pennsylvania, to determine whether the autopilot mode was engaged during the crash. The accident follows the May 7 fatal highway crash involving a Tesla Model S vehicle, which was on autopilot mode and failed to activate the brakes when a tractor-trailer made a turn in front of it.
- Preparing for a quantum computing future. Google is testing new software designed to prevent future quantum computers from cracking current internet encryption techniques. Quantum computers today are small and experimental, containing only a handful of quantum bits, and don't pose much of a threat, Matt Braithwaite, a software engineer at Google, wrote in a blog post. "However, a hypothetical, future quantum computer would be able to retrospectively decrypt any internet communication that was recorded today, and many types of information need to remain confidential for decades. Thus even the possibility of a future quantum computer is something that we should be thinking about today," he said.
Assistant editor Mekhala Roy contributed to this week's news roundup.
Industries to be transformed by machine learning algorithms
Deep learning could be key to more human-like AI
Understanding the impact of deep machine learning
Online retailer Wayfair uses computer vision to learn your style