Intel and Tencent debut AI-powered camera systems for retail

Samsung to build 5G and V2X networks for autonomous car tests at South Korea’s K-City

Samsung is collaborating with the Korea Transportation Safety Authority (KOTSA) to develop mobile network infrastructure for autonomous vehicles at the recently opened K-City test facility. K-City, for the uninitiated, is one of a number of “fake cities” that have emerged as test beds for the latest smart city technologies. Google parent company Alphabet last year offered a glimpse into Castle, a key test hub for its driverless car subsidiary, Waymo. Incidentally, Waymo launched its first commercial self-driving car service in Phoenix just yesterday. And Russia opened a tech-focused “town” called Innopolis back in 2012, where Yandex recently kickstarted tests for its autonomous taxis. Against that backdrop, the South Korean government announced K-City back in May, though it only partially opened for business last month.

Largest

At 320,000 square meters, Korea is touting K-City as the largest dedicated facility for testing self-driving cars in the world, built to replicate all manner of real-world scenarios, including bus lanes, bike lanes, highways, built-up urban areas, parking bays, and more. Situated about an hour’s drive south of Seoul in the city of Hwaseong, K-City cost around $10 million to build.

Meeshkan, a company whose Slack bot helps engineers monitor and train machine learning models without leaving the team chat app, has been named winner of the Slush 100 startup competition. Meeshkan works with popular frameworks like PyTorch and TensorFlow and is optimized for deep learning workflows. The competition took place today in Helsinki at Slush, one of the largest annual tech conferences held in northern Europe. “With our interactive machine learning product, out of the box and for free you get monitoring of all your machine learning jobs on Slack,” Meeshkan CEO Mike Solomon said during his pitch. “On top of that, you’re able to schedule as many jobs as you want right from Slack, pause long setting jobs that are executing, tweak parameters for the job that’s executing, fork a job just like you fork a repo on GitHub, and under the hood it will automatically spin up, provision a server, and send the job off and running.“ Meeshkan competed in the semifinals of Slush 100 against Aerones, a heavy-lift drone company that cleans wind turbines and wants to fight fires with drones, and Lifemote Networks, a SaaS service for internet services providers that predicts Wi-Fi troubleshooting using AI. More than 1,000 applications from 60 countries around the world were received for the startup competition, according to organizers. By supplying a tool that enables engineers and data scientists to train models without the full understanding of how to train and deploy a model from scratch, like a machine learning engineer, Meeshkan intends to help companies address the widespread shortage of data scientists. In a PricewaterhouseCoopers study released earlier this year, only 4 percent of business executives said their company has successfully implemented AI in their products or services, but that’s expected to change in the years ahead. The Slush tech conference was attended by 20,000 people. Among them: more than 1,000 investors and more than 3,000 startups. By category, the largest group of startups in attendance self-identified as AI, big data, or machine learning companies.

Hire by Google’s candidate discovery tool exits beta

Hire by Google, the hiring dashboard that’s part of Google’s enterprise-focused G Suite platform, launched a little over a year ago in June. Since then, it’s gained a feature — candidate discovery — that surfaces appropriate candidates for new gigs at a company, along with a veritable suite of AI-powered calendar scheduling, resume review, and phone call tools. Today, candidate discovery, which rolled out to select customers in beta earlier this year, is becoming generally available to all G Suite customers who pay for Hire.

Coinciding with candidate discovery’s wider launch, Google’s debuting a new capability that product manager Omar Fernandez says was informed by Hire’s beta testers. It allows customers to screen resumes with smart keyword highlighting based on their search criteria, and to re-engage qualified candidates in bulk.   “Throughout the beta period, we listened to customer feedback, and as a result [we introduced this] new feature in candidate discovery,” he wrote in a blog post. “Since the … release of candidate discovery … we’ve heard from many customers how it’s helped them quickly fill open roles at their companies … [One company] was able to fill one of its roles in 24 hours (the average time to hire is four weeks).” Two of those customers are OpenLogix, a global technology service firm, and Titmouse, an animation studio. OpenLogix uses candidate discovery to search a database of 30,000 prior candidates and create a prioritized list based on how well the candidate’s profile matches the title, job description, and location. Meanwhile, Titmouse taps it to manage thousands of applications submitted through the company’s careers page. Fernandez noted that candidate discovery is powered by Google’s Cloud Talent Solution (formerly Cloud Job Discovery), a development platform for job search workloads that factors in desired commute time, mode of transit, and other preferences in matching employers with job seekers. It also drives automated job alerts and saved search alerts. According to Google, CareerBuilder, which uses Cloud Talent Solution, saw a 15 percent lift in users who view jobs sent through alerts and 41 percent increase in “expression of interest” actions from those users. Hire by Google, for the uninitiated, is a full stack recruitment tool that lets hiring managers sift through job listings, interview and screen candidates, solicit applications, and more. It natively integrates with Gmail, Google Calendar, and Sheets, automatically filling in details such as contact information in invites and recording data captured across interviews. Moreover, thanks to artificial intelligence (AI), it’s able to recommend appropriate time slots for meetings and interviews, analyze key terms in job descriptions, and highlight candidates’ phone numbers and log calls. News of candidate discovery’s general availability follows on the heels of Google’s job search feature for military veterans, which launched in August. It aims to make it easier for service members to find civilian jobs that align with their occupation, in part by finding jobs in their area that require skills similar to those used in their military role. Companies that use Cloud Talent Solution can implement the job search feature on their own career sites.

6 things a first-time CEO needs to know

CEO turnover is on the rise across corporate America. The number of top executives who have left their jobs in 2018 has reached a 10-year high, according to outplacement firm Challenger, Gray & Christmas.

Reasons for the trend vary, the firm says, from natural movement in a tight labor market, to economic uncertainty, to a desire by companies, in light of #MeToo, “to let go of leaders that do not fit their culture or otherwise act unethically.”

Whatever the causes, the high turnover has an interesting side effect: more opportunities for executives seeking their very first CEO gigs. These vice presidents, general managers, chief financial officers, et al, typically have spent years working toward a shot at the top spot. The current direction toward fresh blood in the corner office means more can get there.

And yet most have no idea about the challenges they’ll encounter.

It’s a subject close to my heart. After several years in leadership roles at Salesforce and Oracle, I became a first-time CEO at a smaller tech company in fall 2015. I left at the beginning of 2018 as part of a corporate restructuring that would have required me and my family to relocate. Five months later, I landed my second CEO post, at another tech firm.

The first realization that smacked me in the face as a rookie: It’s a really hard job. I had naively thought being CEO would be only incrementally more difficult than other positions I’d held. Running a business unit within a company carries a lot of responsibility, right?

Yes, but it’s not even close to the same. As CEO, the buck stops here on company success (or lack thereof), culture, brand satisfaction, product quality, funding, communication with the board of directors, and a host of other priorities that don’t really sink in until you’re in the big chair. It’s all a huge thrill, but it also requires a massive adjustment in how the freshly minted CEO thinks and acts.

I reflect frequently on the lessons I learned as a first-time CEO and how I can apply them at my new company. Here are six of the most important ones, in no particular order.

1. You’re under a microscope. Get used to it. On my third day as a CEO, I visited one of our offices. A manager in one of the groups had emailed me that he’d love to get together over coffee and share some ideas. He just happened to get to me before anyone else. I replied, “Sure, let’s do it.”

Within an hour, everyone in the office of 130 people knew I had gone out for coffee with this guy as one of my first acts as CEO. It was seen as a huge stamp of approval for him and his ideas. I thought I was just meeting a new employee.

The lesson: As a new CEO, everything you do or say, big or small, is magnified. Everyone in the company notices everything. A successful CEO recognizes this and uses their bully pulpit wisely, deliberately, and selectively.

2. It can be lonely. For the first time in their career, a new CEO doesn’t get to commiserate with other people in the company. Because of the microscope effect, if the CEO says something to one person, they must assume they’re saying it to everyone. They can’t, for instance, grab a beer with the head of products and complain about the head of sales. Inevitably, the comments will get out, and the sales leader will never recover. This reality is a hard adjustment for some new CEOs.

The lesson: A CEO will often feel like Tony Soprano. To paraphrase the mob boss from the hit HBO show: “You’ve got no idea what it’s like to be Number One. Every decision you make affects every facet of every other thing … And in the end you’re completely alone with it all.”

3. You have to walk the walk, not just talk the talk on transparency. My first CEO gig was with a company experiencing business challenges. I kept repeating the message to employees that we were going to turn this thing around. I’m not sure, however, that I was transparent enough about the facts of the situation and specifically why I felt some areas were broken. I could have built more trust by sharing more.

At my new company, I put up a dashboard at our monthly all-hands meeting with all 260 employees. It’s the same one I use with the executive team and the board. It has all the numbers and facts that show how the company is performing and how we need to.

The lesson: The first step in a transparent culture is, well, transparency. It’s the CEO’s job to make sure employees are given sufficient information so they’re not just along for the ride but truly are part of the ride.

4. You’re going to be busier than you’ve ever been. Deal with it. The amount of time CEOs spend watching over the business, meeting with investors and potential ones, talking with the media, etc., can be staggering. A first-time CEO must learn quickly that they can’t allow the calendar crush to steal their approachability. “I’m so busy” are words that can never slip from a CEO’s lips.

Instead, they find ways to maximize their time. For example, I hold all-hands meetings with an open Q&A. I’ll spend a morning with, say, the customer support team as they work, to see their challenges first hand and to listen. I’ve even been known to sit in the break room with my laptop, chatting up everyone who walks in. I also send a weekly email highlighting what I feel are top accomplishments, activities, and priorities.

The lesson: Most CEOs are outgoing and approachable by nature, but if they’re not careful, the demands of being CEO can make them significantly less approachable. They can’t allow themselves to fall into that trap.

5. Balance strategy and execution. While it’s the CEO’s job to set a strategic direction for the company, he or she can’t lose sight of execution. One of the first things I did in my second CEO gig was establish a team of 30 people – not just my direct reports but other leaders responsible for execution – and we meet twice a month to examine business operations and make sure execution is aligned with strategy.

The lesson: CEOs must consistently insert themselves into the execution conversation.

6. Set the right example for work-life balance. At 4 PM every Tuesday, I leave work to coach my kids’ youth hockey team. It’s an inconvenient time for a CEO to step away, but my family is my No. 1 priority. I explain to employees why I do this, how technology allows me to time shift my work to later at night, and that they are entitled to the same flexibility.

The lesson isn’t “work hard, play hard,” which sounds like a fraternity-type culture. It’s reinforcing the message that employees don’t need to be sitting on email from 4 to 5 if their kid has soccer practice. Go be with your kids and catch up on work later in the day. What’s good for the CEO is good for everyone else.

According to LinkedIn, 12,000 of its members share the title of CEO at companies with more than 50 employees. CEO is an exclusive club, and it can be daunting for those entering it for the first time. But by knowing what they’re getting into, and exercising the strong judgment and careful thought that got them to the top spot in the first place, rookie CEOs can knock it out of the park.

AI is moving B2B tech from consumerization to humanization

You make a call to a telecom operator over a billing-related query. Or you log on to a shopping site to check the status of your shipment. You are likely to be instantly greeted by an automated voice that will guide you, while a friendly chat box pops up on the shopping site to provide instant solutions. There’s nothing new about this; we’ve been hearing automated voice responses and seeing chatbots on websites and apps for quite some time now. So, what’s the difference? If the automated message on your screen was once “Error 856. Wrong Input,” it has now evolved to something relatable, conversational, and polite, like, “I’m sorry. Could you please repeat your issue?” AI technology is getting groomed, suiting up, and presenting itself with the best, most human foot forward. Technology is becoming increasingly human — and here’s why.

Consumerization of technology: a revolution in customer experience

Gone are the days when B2B software was clunky, frustrating and hard to use. Competition in the B2B space and the demand for instant solutions and quick fixes paved the way for the consumerization of B2B technology — bringing it close on the heels of B2C innovations. The impact of this has been so rapid and distinct that the lines have blurred between business and personal technology. Today, tech innovation in the B2B industry is driven by the determination to give every customer the best possible customer experience — delivering “wow” moments. The result? Customer expectations have changed: People now want to have their problems solved, but they’re not interested in being aware of the tech involved in the solution. They don’t want an overload of potentially irrelevant information tailored for the masses. They want personalized, intuitive, and accessible solutions for their needs. This has naturally led to an increased emphasis on AI. AI meets three critical needs in the customer service landscape today: personalization, contextual intelligence, and immediate responses. Today, AI has become a mainstay in the B2B industry — much like in the B2C space — and is bringing disruption like never before. Business software is addressing changing customer expectations by becoming seamlessly integrated with human customer service — so much so that the technology appears human in behavior.

Humanization and AI: The merger

Humanization is the next logical step toward a better user experience, and AI is rightly riding this wave, with solutions becoming increasingly like human interactions. What does the future look like? The ideal customer experience will seamlessly combine the ingenuity of AI along with the empathy of human engagement. This is the Holy Grail that customer experience and product team engineers are working toward. For our part at Freshworks, we recently launched a friendly canine partner, Freddy — our conversational AI omnibot. Freddy learns from the Freshworks records of customer interactions across marketing, sales, and support, and automatically replies to common queries in email, chat, voice-calls, and even social media with the appropriate content. This enables sales and support teams to focus on more complex, high-value inquiries. Being a SaaS firm, we are increasingly looking to understand and serve not only the immediate and direct customers of our products, but also the end-user of each business that engages with our software. Today, our lens has expanded to include the needs of the direct customer as well as the impact it has on the customer’s customer. And omnibots like Freddy, are possibly still only the beginning.

The future of AI: What’s to come?

AI is making life easier for us and our customers, with its gamut of features such as enhanced security, predictive analytics, automation, assistance in deploying code, and more. However, the innovations to look for are the advances in chatbots, voice recognition, and natural language processing in AI, which are gaining momentum due to the increased need for technology to be more human. AI is here to stay, but this does not mean that humans are replaceable. It merely means that an integration is underway. And when this integration becomes seamless, technology will become closer to becoming a part of our identity. We’ll welcome that with open arms. STS Prasad, the Senior VP of Engineering at Freshworks, is a technology entrepreneur, also known for scaling technology organizations and platforms for business growth.

Uber confidentially files for IPO a day after Lyft

(Reuters) — Uber Technologies has filed paperwork for an initial public offering, according to three people with knowledge of the matter, taking a step closer to a key milestone for one of the most closely watched and controversial companies in Silicon Valley. The ride-hailing company filed the confidential paperwork on Thursday, one of the sources said, in lock-step with its smaller U.S. rival, Lyft, which also announced on Thursday it had filed for an IPO. The simultaneous filings extend the protracted battle between Uber and Lyft, which as fierce rivals have often rolled out identical services and matched each other’s prices. Uber is eager to beat Lyft to Wall Street, according to sources familiar with the matter, a sign of the company’s entrenched competitiveness. Its filing sets the stage for one of the biggest technology listings ever. Uber’s valuation in its most recent private financing was $76 billion, and it could be worth $120 billion in an IPO. Its listing next year would be the largest in what is expected to be a string of public debuts by highly valued Silicon Valley companies, including apartment-renting company Airbnb and workplace messaging firm Slack. Ongoing market volatility, however, could alter companies’ plans. The IPO will be a test of public market investor tolerance for Uber’s legal and workplace controversies, which embroiled the company for most of last year, and on Chief Executive Dara Khosrowshahi’s progress in turning around the company. Khosrowshahi took over just over than a year ago, and has repeatedly stated publicly he would take Uber public in 2019. In August, he hired the company’s first chief financial officer in more than three years. Together, Uber and Lyft will test public market investor appetite for the ride-hailing business, which emerged less than a decade ago and has proven wildly popular, but also unprofitable. Uber in the third quarter lost $1.07 billion and is struggling with slowing growth, although its gross bookings, at $12.7 billion, reflect the company’s enormous scale. Its revenue for the quarter was $2.95 billion, a 5 percent boost from the previous quarter. Its bookings grew just six percent for the quarter. Uber has raised about $18 billion from an array of investors since 2010, and it now faces a deadline to go public. An investment by SoftBank that closed in January, which gave the Japanese investor a 15 percent stake in Uber, included a provision that requires Uber to file for an IPO by Sept. 30 of next year or the company risks allowing restrictions on shareholder stock transfers to expire. Uber has not formally chosen underwriting banks, although Morgan Stanley and Goldman Sachs are likely to get the lead roles, sources told Reuters. Lyft hired JPMorgan Chase & Co, Credit Suisse and Jefferies as underwriters. The Wall Street Journal reported Uber’s filing earlier on Friday.

History of Scandal

Becoming a public company will bring a heightened level of investor scrutiny and exposure to Uber, which suffered a string of scandals when the company was led by co-founder and former CEO Travis Kalanick, who resigned last year. The controversies included allegations of sexual harassment, obtaining the medical records of a woman raped by an Uber driver in India, a massive data breach, and federal investigations into issues including possibly paying bribes to officials and illicit software to evade regulators. Khosrowshahi and his leadership team have worked to reset the workplace culture and clean up the messes, including settlements with U.S. states over the data breach and with Alphabet’s self-driving car unit, Waymo, which had sued Uber for trade-secrets theft. Uber today is a different company than the vision its founders pitched to early investors, which helped it become the most highly valued venture-backed company in the United States. After concessions in China, Russia and Southeast Asia, where Uber sold its business to a local competitor, and the prospect of another merger in the Middle East, Uber is far from being the dominant global ride-hailing service it set out to be. Still, Uber operates in more than 70 countries, while Lyft is in the U.S. and Canada, although the smaller company is plotting a global expansion. Uber has also added a number of other businesses, which are growing but have yet to show sustainable profits, in a bid to become a one-stop mobility app. Those include freight hauling, food delivery and electric bike and scooter rentals. Meanwhile, its self-driving car unit is costing the company about $200 million a quarter, according to investors, but Uber’s program has retrenched since one of its autonomous cars killed a pedestrian in March.

Google’s Gradient Ventures leads $7 million investment in Wise Systems to automate routing for shippers

Wise Systems, which develops autonomous routing and dispatch software for delivery fleets, has raised $7 million in a series A round of funding led by Gradient Ventures — Google’s AI-focused investment fund. Additional participants in the round include Neoteny, E14 Fund, Trucks Venture Capital, and Fontinalis Partners. Founded out of Cambridge, Massachusetts in 2014, Wise Systems targets retailers, distributors, shippers, and couriers with software that automates many of the processes involved in getting goods from A to B. This includes scheduling and monitoring routes, with a built-in mechanism that rearranges stops and adjusts routes in real time based on anticipated delays caused by traffic and other factors. It leans heavily on machine learning smarts to monitor data points and improve delivery fleets’ performance over time. Feeding into this is a mobile app for drivers, which lets them log arrival and departure times, capture signatures, and record notes manually, while on the customer-facing end a hub tracks delivery status and real-time arrival schedules. Wise Systems had previously raised around $1 million, and with another $7 million in the bank it plans to double down on its operational growth and invest in R&D for AI-powered delivery management. “Autonomous dispatch and routing is the next-generation technology that logistics professionals need to meet the increasingly complex requirements of the rapidly evolving economy,” noted Wise Systems CEO Chazz Sims. “Our technology positions fleets to meet today’s and tomorrow’s needs, reshaping delivery in the $10 trillion logistics and transportation industry.” Google parent company Alphabet has operated a number of investment funds for some time already, but last July Google itself announced a new venture fund targeting early-stage AI startups. Since its inception, Gradient Ventures has invested in fewer than a dozen startups, including biomedical startup BenchSci.

Big money

The delivery and logistics industry is ripe for investment, with a number of big technology companies pushing resources into the sphere. Walmart recently launched Spark Delivery, a crowdsourced pilot delivery program similar to Amazon Flex, while Amazon announced that it is looking to create a network of independent delivery fleets to bolster its existing network of third-party providers. Last year, Uber launched its Uber Freight trucking business as it sought to expand beyond its consumer ride-hailing and food delivery businesses. As it happens, Uber revealed last week that it will begin using machine learning to provide shippers with accurate rates up to two weeks in advance. Meanwhile, Google has also invested in the delivery and logistics realm — in October it joined a $40 million investment in same-day delivery platform Deliv. Given activity elsewhere in the technology industry, it should come as little surprise that Google’s AI-focused venture fund would seek to invest in a company like Wise Systems. “With their applications of machine learning already delivering bottom-line savings and top-line expansion features to customers, we are excited to support Wise Systems’ continued growth,” added Gradient Ventures managing partner Anna Patterson. “Wise’s large customers are redefining their abilities and re-setting expectations of last mile delivery for the on-demand age.”

iOS 12 passes 70% adoption in 77 days while Android Pie remains a mystery

Uptake of Apple’s iOS 12 operating system for iPhones, iPads, and iPod touches continues to grow, a new Apple update revealed today, as a full 70 percent of all iOS devices are now on the latest major release. The milestone was reached December 3, or 77 days after the operating system’s September 17 debut. Once again, Apple has provided two sets of statistics — one for all devices, and a second, higher number of 72 percent adoption for devices “sold in the last four years.” In both cases, the gains came largely at the expense of iOS 11, which lost around 10 percent of its users to iOS 12, while earlier versions of iOS continue to hold onto a less than 10 percent share — a shrinking 9 percent of all devices, versus a stable 7 percent of devices sold in the last four years. Unlike Apple, which has provided continued updates showing the growth of iOS 12, Google’s Android developers page hasn’t been updated since late October, when Android 9 Pie apparently represented under 0.1 percent of the Android userbase. Google released Pie in early August 2018, and most recently put out an updated version just yesterday, though relatively few Android devices appear to be running the release. Android 9’s numbers could go up over the next month or two, however, as new devices are sold through the holidays with the latest OS preinstalled.

Microsoft unveils autoscaling for Azure Kubernetes Service and GPU support in Container Instances

At Microsoft Connect(); 2018 today, Microsoft launched a bevy of updates across its cloud, artificial intelligence (AI), and development product suites. That’s no exaggeration — the company made its Azure Machine Learning service generally available, enhanced several of its core Azure IoT Edge offerings, and debuted Cloud Native Application Bundles alongside the open source ONNX Runtime inferencing engine. And that only scratched the surface of today’s small mountain of announcements. Microsoft also took the wraps off improvements to virtual nodes in Azure Kubernetes Service (AKS) and autoscaling for AKS. And it made available JavaScript and Python support in Azure Functions, its serverless compute service that allows customers to run code on-demand without having to explicitly provision or manage infrastructure, in addition to a new Azure API Management Consumption tier. First off, Azure Kubernetes Service (AKS), which let developers elastically provision quick-starting pods inside of Azure Container Instances (ACI), is available in public preview starting today. Customers can switch it on within the Azure portal, where it cohabitates in the virtual network with other resources. Also launching today: the aforementioned cluster autoscaling for AKS. It’s based on the upstream Kubernetes cluster autoscaler project and automatically adds and removes nodes to meet workload needs (subject to minimum and maximum thresholds, of course). It works in conjunction with horizontal pod autoscaling and enters public preview this week. Dovetailing with those announcements, Microsoft debuted graphics processing unit (GPU) support in ACI, which lets developers run demanding jobs — such as AI model training — containerized in accelerated virtual machines. And the Seattle company detailed the Azure Serverless Community Library, an open source set of prebuilt components based on common use cases, such as resizing images in Blob Storage, reading license plate numbers with OpenALPR, and conducting raffles, all of which can be deployed on Azure subscriptions with minimal configuration. GPU support in ACI launches in preview today, and the Azure Serverless Community Library is available on GitHub and the Serverless Library website. Rounding out today’s news, Microsoft took the wraps off a new consumption plan for Linux-based Azure Functions, which was teased at Microsoft Ignite earlier this year. Now it’s possible to deploy Functions built on top of Linux using the pay-per-execution model, enabling serverless architectures for developers with code assets or prebuilt containers. Finally, Microsoft launched Azure API Management (APIM) — a turnkey solution for publishing APIs to external and internal customers — in public preview in a consumption-based usage plan (Consumption Tier). Effectively, it allows APIM to be used in a serverless fashion, with instant provisioning, automated scaling, and pay-per-play pricing. Along with the new APIM plan is the general availability of Python support (specifically Python 3.6) on Azure Functions runtime 2.0, and JavaScript support for Durable Functions, an extension of Azure Functions and Azure WebJobs that manages state, checkpoints, and restarts in serverless environments.

Cambridge Consultants’ DeepRay uses AI to reconstruct frames from damaged or obscured footage

Rain. Smoke. Dirt. Debris-produced distortion would normally spell doom for a videographer, but researchers at global research and development firm Cambridge Consultants say they’ve harnessed artificial intelligence (AI) to reconstruct footage from damaged or obscured frames in real time. In one test on airfields and aviation stock video, it was able to accurately reproduce aircraft on a runway.The AI system, dubbed DeepRay, will be fully detailed at the upcoming 2019 Consumer Electronics Show in January. It calls to mind Adobe’s distortion-correcting system for front-facing smartphone cameras, and an Nvidia technique that can “fix” corrupt images containing holes. But unlike most previous AI, DeepRay handles live video.“Never before has a new technology enabled machines to interpret real-world scenes the way humans can — and DeepRay can potentially outperform the human eye,” Tim Ensor, commercial director for artificial intelligence at Cambridge Consultants, told VentureBeat in a phone interview. “The ability to construct a clear view of the world from … video, in the presence of continually changing distortion such as rain, mist, or smoke, is transformational.”DeepRay — a product of Cambridge Consultants’ Digital Greenhouse internal incubator — leverages a machine learning architecture called a generative adversarial network (GAN) to effectively invent video scenes as it attempts to remove distortion. Broadly speaking, GANs are two-part neural networks consisting of generators that produce samples and discriminators that attempt to distinguish between the generated samples and real-world samples. In DeepRay’s case, a total of six networks — a team of generators and discriminators — compete against each other.Research in GANs has advanced by leaps and bounds in recent years, particularly in the realm of machine vision. Google’s DeepMind subsidiary in October unveiled a GAN-based system that can create convincing photos of food, landscapes, portraits, and animals out of whole cloth. In September, Nvidia researchers developed an AI model that produces synthetic scans of brain cancer, and in August, a team at Carnegie Mellon demonstrated AI that could transfer a person’s recorded motion and facial expressions to a target subject in another photo or video. More recently, scientists at the University of Edinburgh’s Institute for Perception and Institute for Astronomy designed a GAN that can hallucinate galaxies — or high-resolution images of them, at least.Ensor contends that only in the past two years has it been possible to train multi-network GANs at scale, in large part thanks to advances in purpose-built AI chips such as Google’s Tensor Processing Units (TPUs).“We’re excited to be at the leading edge of developments in AI. DeepRay shows us making the leap from the art of the possible, to delivering breakthrough innovation with significant impact on our client’s businesses,” Ensor said. “This takes us into a new era of image sensing and will give flight to applications in many industries, including automotive, agritech and healthcare.”At CES, the DeepRay team will demonstrate a neural network trained on Nvidia’s DGX-1 platform that, running on a standard gaming laptop, can remove distortion introduced by an opaque pane of glass. The dataset consists of 100,000 still images, but Ensor said that the team hasn’t characterized the system’s performance with larger sample sizes.“As with all [AI models], it continues to improve with training,” he explained, “and it will degrade gracefully.”In a press release today, the Linux Foundation’s Cloud Native Computing Foundation, which oversees Kubernetes, said the latest release includes simplified cluster management with Kubeadm, a container storage interface (CSI), and CoreDNS as Default DNS. At just 10 weeks since the last release, CNCF says this is one of the service’s shortest development cycles.Read more: How developers want to use Kubernetes and microservices to make the web faster, stable, and more openKubernetes is a foundational part of an effort by CNCF to change the way developers write applications for internet-based services. Cloud-based computing calls for breaking out separate features, or “microservices,” and placing them in containers that have all the necessary pieces for an application to run in one package.The development philosophy holds that breaking applications into smaller, self-contained units can significantly reduce costs and the time needed to write, deploy, and manage each one.Kubernetes, originally developed by Google but donated to the Linux Foundation, is used to manage deployment of microservices. In the press release, CNCF notes of the three main features added to Kubernetes 1.13: “These stable graduations are an important milestone for users and operators in terms of setting support expectations.”

How Apple’s HomePod uses AI and 6 mics to hear users through ambient noise

Apple’s HomePod hasn’t won much praise for the capabilities of its integrated digital assistant Siri, but it does have one undeniably impressive feature: the ability to accurately hear a user’s commands from across the room, despite interference from loud music, conversations, or television audio. As the company’s Machine Learning Journal explains today, HomePod is leveraging AI to constantly monitor an array of six microphones, processing their differential inputs using knowledge gained from deep learning models.One of the biggest challenges in discerning a user’s commands over ambient noise is overcoming the HomePod itself: Apple’s speaker can perform at very high volumes, and its microphones are immediately adjacent to the noise sources. Consequently, the company explains, there’s no way to completely remove the HomePod’s own audio from the microphones — only part of it.Instead, Apple used actual echo recordings to train a deep neural network on HomePod-specific speaker and vibration echoes, creating a residual echo suppression system that’s uniquely capable of cancelling out HomePod’s own sounds. It also applies a reverberation removal model specific to the room’s characteristics, as measured continuously by the speaker.Another interesting trick uses beamforming to determine where the speaking user is located, focus the microphones on that person, and apply sonic masking to filter out noises from other sources. Apple built a system that makes judgments about local speech and noise statistics based solely on the microphones’ current and past signals, focusing on the speech while trying to cancel out interference. It then trained the neural network using a variety of common noises that ranged from diffuse to directional, speech to noise, so that the filtering could apply to a large number of interference sources.

HomePod’s other impressive capability is determining which of multiple speaking people is the correct target for commands, to steer the beamforming mics and isolate noise. One trick is using the required “Hey Siri” trigger phrase to determine who and where commands are coming from, but Apple also developed techniques to separate competing talkers into individual audio streams, then use deep learning to guess which talker is issuing commands, sending only the stream focused on that talker for processing.

The Machine Learning Journal’s entry does a great job of spotlighting how AI-assisted voice processing technologies are necessary but not sufficient to guarantee a great experience with far-field digital assistants. While all of the techniques above do indeed yield quick, reliable, and accurate Siri triggering, HomePod’s limited ability to actually respond fully to requests was a frequent target of complaints in reviews. If there’s any good news, it’s that the issues appear to be in Siri’s cloud-based brain rather than in the HomePod’s hardware or locally run services, so server-side patches could dramatically improve the unit’s functionality without requiring users to buy new hardware.

Intel open-sources HE-Transformer, a tool that allows AI models to operate on encrypted data

As any data scientist will tell you, datasets are the lifeblood of artificial intelligence (AI). That poses an inherent challenge to industries dealing in personally identifiable information (e.g., health care), but encouraging progress has been made toward an anonymized, encrypted approach to model training.

Today at the NeurIPS 2018 conference in Montreal, Canada, Intel announced that it has open-sourced HE-Transformer, a tool that allows AI systems to operate on sensitive data. It’s a backend for nGraph, Intel’s neural network compiler, and based on the Simple Encrypted Arithmetic Library (SEAL), an encryption library Microsoft Research also released in open source this week.

The two companies characterized HE-Transformer as an example of “privacy-preserving” machine learning.

“HE allows computation on encrypted data. This capability, when applied to machine learning, allows data owners to gain valuable insights without exposing the underlying data; alternatively, it can enable model owners to protect their models by deploying them in encrypted form,” Fabian Boemer, a research scientist at Intel, and Casimir Wierzynski, Intel’s senior director of research, wrote in a blog post.

The “HE” in HE-Transformer is short for homomorphic encryption, a form of cryptography that enables computation on ciphertexts — plaintext (file contents) encrypted using an algorithm. It generates an encrypted result that, when decrypted, exactly matches the result of operations that would have been performed on unencrypted text.

HE is a relatively new technique — IBM researcher Craig Gentry developed the first fully HE scheme in 2009. And as Boemer and Wierzynski note, designing AI models that use it requires expertise in not only machine learning but encryption and software engineering.

HE-Transformer aids in the development process by providing an abstraction layer that can be applied to neural networks on open source frameworks such as Google’s TensorFlow, Facebook’s PyTorch, and MXNet. It effectively eliminates the need to manually integrate models into HE cryptographic libraries.

HE-Transformer incorporates the Cheon-Kim-Kim-Song (CKKS) encryption scheme and addition and multiplication operations, such as add, broadcast, constant, convolution, dot, multiply, negate, pad, reshape, result, slice, and subtract. Additionally, it supports HE-specific techniques, like plaintext value bypass, SIMD packing, OpenMP parallelization, and plaintext operations.

Thanks to those and other optimizations, Intel claims that HE-Transformer delivers state-of-the-art performance on cryptonets — learned neural networks that can be applied to encrypted data — using a floating-point model trained in TensorFlow.

“We are excited to work with Intel to help bring homomorphic encryption to a wider audience of data scientists and developers of privacy-protecting machine learning systems,” said Kristin Lauter, principal researcher and research manager of cryptography at Microsoft Research.

Currently, HE-Transformer directly integrates with the nGraph compiler and runtime for TensorFlow, with support for PyTorch forthcoming. Deep learning frameworks that are able to export neural networks to ONXX — such as PyTorch, CNTK, and MXNet — can be used by importing models into nGraph in ONXX and exporting them in a serialized format.

Boemer and Wierzynski said that future versions of HE-Transformer will support a wider variety of neural network models.

“Recent advances in the field have now made HE viable for deep learning,” they wrote. “Researchers can leverage TensorFlow to rapidly develop new HE-friendly deep learning topologies.”

Verizon plans Samsung 5G phone with Qualcomm modem for first half of 2019

Just days after helping three cellular carriers initiate 5G service in South Korea, Samsung made an interesting announcement in partnership with U.S. carrier Verizon: The companies plan to release “one of the first commercial 5G smartphones” in the first half of 2019, apparently powered not by Samsung’s own chipset, but instead by Qualcomm’s Snapdragon platform, including its X50 5G modem.

Today’s announcement is significant because Samsung has been working on its own 5G technologies, including mobile chips, and was a major contributor to the international 5G standard that was finalized late last year and earlier this year. Yet Verizon indicates that Samsung will be showing a “proof of concept” device this week, using Qualcomm’s upcoming flagship mobile platform, antenna modules, and related components.

The use of Qualcomm parts simultaneously reflects the San Diego company’s apparent strength in the emerging 5G chip market and the continued struggles of even its largest rivals to miniaturize the high-speed mobile chips. Facing thermal and battery issues with its first 5G chips, chief rival Intel recently pushed up development of a second-generation 5G chipset that is expected to be used in iPhones and other devices in 2020. Chinese developer Huawei has shown giant heatsinks to work around its own 5G modem’s thermal issues. By comparison, numerous companies have signed up to use Qualcomm’s parts.

Verizon says that its mobile 5G service “will go live in early 2019 and expand rapidly.” The company is already offering fixed 5G home broadband servicein Houston, Indianapolis, Los Angeles, and Sacramento, with reported typical speeds up to twice the company’s promised 300Mbps norm, approaching its 1Gbps peak.

Samsung’s 5G phone is unlikely to be the first mobile device on Verizon’s network. Verizon previously announced that Motorola will release the 5G Moto Mod, a Snapdragon X50-equipped 5G backpack, for the Moto Z3 in “early 2019.” The Moto Z3 is already available, making the addition of 5G capabilities as easy as snapping the accessory onto the back of the existing phone.

How Nvidia researchers are making virtual worlds with real-world videos

The team at Nvidia was initially inspired to take this approach by the work of Alexei Efros and other researchers at University of California, Berkeley, and by their creation of the Pix2Pix system. Nvidia created Pix2PixHD in response, working in tandem with AI practitioners from UC Berkeley. Earlier this year, UC Berkeley researchers also produced models that were able to dance and do flips and 20 other acrobatic moves.

“I think this is the first interactive AI rendering, and we’re really proud of the progress we’ve made. But it is early stage, and I think there’s going to be a lot of progress to make the output higher quality and more general so that we can handle more kinds of scenes. And so I’m really excited about where this is going to go in the future,” Catanzaro said.