Daily News Digest Featured News

Blippar AR collaboration, spotting malware with machine learning, a robot delivers takeout

machine learning malware

News Summary

Augmented Reality

GateHouse Media and Blippar Collaborate on Multi-Platform Augmented Reality Project for Pearl Harbor’s 75th Anniversary (BusinessWire)

Blippar and GateHouse Media announce today their collaboration to use augmented reality in an exclusive editorial section commemorating the 75th anniversary of Pearl Harbor.

Using Blippar’s app technology, readers of “Remembering a Day of Infamy” will be able to access digital content from print pages via their smart devices. For example, in a story on Christmas 1941, readers can open the free Blippar app, point their mobile device at a page about President Roosevelt’s speech to the nation, and video of the speech will begin to play on their device.

Harman invests in Navdy’s ‘augmented reality’ HUD (Left Lane News)

Harman has acquired exclusive rights to sell Navdy’s ‘augmented reality’ head-up display system to automotive OEMs.

Seek Launches New Augmented Reality Mobile Platform (Mobile Marketing Watch)

Brands can launch augmented reality, location-based marketing campaigns through a new platform about which MMW was briefed earlier this week.

“We give businesses the ability to drop a treasure chest right outside their store, put whatever they want inside it, and attract customers to come open it, vying at a chance to win something,” Cheney said.


Augmented reality is also how companies like Snapchat are growing user engagement, and it’s how one Canadian association is making recycling even more rewarding.

The game’s slogan, “gotta catch ’em all,” probably rings true for membership directors and managers, who may feel like they need to catch all their members at the right moments.

Augmented reality can help you do that, Ziesenis says. It’s a creative way to update members with important information, and it takes full advantage of engaging users from their mobile devices.

Bell Integrator and Nostromo Bring Augmented Reality to the Enterprise (EconoTimes)

Bell Integrator, a global consulting, technology services and outsourcing company, and Nostromo, a pioneer of emerging technologies, announce the launch of a new Augmented Reality (AR) solution targeted to the enterprise.

The AR solution developed by Nostromo delivers a unique experience that revolutionizes the way organizations connect with their customers. Users can now engage with interactive content by pointing their cellphone camera at an AR trigger, which can be an image, physical object, or a geographic location. This solution takes mobile marketing to a whole new level, and the team at Nostromo is implementing this solution to their clients, resulting in tens of thousands of unique brand impressions, and increased revenue across the board!

Bell Integrator will empower its clients from the manufacturing, construction, eCommerce, travel, healthcare, and logistics industries to leverage the technology for their own industry-specific goals, taking over their own development, customization, and support.

There are many use cases for this emerging technology, such as 360° tours, location finders, virtual product demonstrations, education and training, product simulations and testing, incentive promotions, and interaction.

Up your cycling game with these augmented-reality glasses (Yahoo! Tech)

These smart cycling glasses, called Raptors, are the only ones that do not obstruct your view as you ride.
Navigation, speed, heart rate and other vital information will appear right in front of you, but the data will not get in your way, since it is displayed in an unobtrusive way.
The company that makes the glasses, Everysight, says Raptors can also record your ride and review how well you did on it afterward.

Artificial Intelligence/Machine Learning

Detecting Malware Pre-execution with Static Analysis and Machine Learning (SentinelOne)

It’s widely agreed in the industry that simple byte signatures aren’t enough to reliably detect malware anymore. Instead, modern anti-virus products heavily rely on some combination of static and dynamic analysis to feed features into predictive models which determine if a particular file is malicious or not.

One of the major benefits of static-based detection is that it can be performed before the file is executed (or pre-execution). This is obviously useful because it’s much easier to remediate malware if it’s never allowed to execute. An ounce of prevention is worth a pound of cure. A corollary of this benefit is that even corrupt and malformed executables which won’t execute can still be detected statically. Of course, any sort of detection which is mostly based on behavioral analysis will fail to detect these same samples because they don’t generate any behavior. It’s questionable if these types of files should even be considered malicious. Even still, there may be some value in detecting and removing malware which can’t actually harm you simply because it brings peace of mind and suits existing policies and procedures.

microsoft PED

In summary, the advantages of static detection are:

  • can be performed pre-execution
  • works on samples with dead C&C servers
  • works on invalid executables
  • computationally inexpensive

While the major disadvantages are:

  • doesn’t reliably determine behavior
  • doesn’t detect what happens in memory
  • less likely to detect novel threats
  • easier to develop counter-measures

Creating a predictive model starts with collecting a huge number and variety of malicious and benign files. Then, features are extracted from each file along with the file’s label (e.g. malicious or benign). Finally, the model is trained by feeding all of these features to it and allowing it to crunch the numbers and find patterns and clusters in the data. Depending on how good your hardware is, this may take many hours or days. In this way, when the features of a file with an unknown label are presented to the model, it can return a confidence score of how similar these features are to those of the malicious and benign sets.

machine learning malware

For various reasons which I’ll describe below, we settled on using a Random Forest for our model. Random forests are almost unreasonably effective and work decently well out of the box without much tuning, even with very large numbers of features.

Let’s say you own a vast ranch full of cats and dogs and you’ve built a robot which will help you take care of all of them. The robot needs to be able to determine if an animal is a cat or a dog so it knows which protocol to use when treating the animal. Of course, anyone who’s owned or operated a cat knows they are special snowflakes which require slightly different handling than dogs.

To train your random forest, you’ll first need to extract the features for each animal on your bizarre ranch and record them in something like a spreadsheet. And for each row of features, another spreadsheet contains the label (cat or dog) for that animal.

The process of training a random forest is really a process of creating many decision trees where each node in the tree represents an if-else of some feature value. In a random forest, the features for each decision tree are determined somewhat randomly, but in this example every feature is used for simplicity. As the decision tree is trained, each row of features is run through the decision tree by following all of the conditionals. The leaves of the tree contain the probabilities that a particular label would reach that particular leaf. Here’s an example decision tree of the above features. The leaves contain two probabilities one for C (cat) and one for D (dog).

random forest

Capital One Pursues ‘Explainable AI’ to Guard Against Bias in Models (The Wall Street Journal)

Capital One Financial Corp. is researching ways that machine-learning algorithms could explain the rationale behind their answers, which could have far-reaching impacts in guarding against potential ethical and regulatory breaches as the firm uses more artificial intelligence in banking.

The company is employing in-house experts to study “explainable AI,” a nascent field of research aimed at creating computer programs that can translate, in natural language, how a machine-learning model comes to a logical decision, which is imperative as organizations explore ways to leverage AI but realize that it can be rife with bias.

Artificial Intelligence Just Broke Steve Jobs’ Wall of Secrecy (Wired)

THE ARTIFICIAL INTELLIGENCE researcher Russ Salakhutdinov made headlines today when he said was going to start publishing journal articles and spending time talking to academics.

That wouldn’t be news, except Salakhutdinov works for Apple—a company famous for an extreme breed of corporate secrecy.

Salakhutdinov oversees Apple’s artificial intelligence group, and the only way he can recruit top researchers is to reassure them that once they get to Apple, they can continue to publish their work and share their ideas with the larger AI community.

Numenta Brings Brain Theory to Machine Learning in New Paper (Inside Big Data)

The earlier paper described a biological theory of how networks of neurons in the neocortex learn sequences. In this paper, the authors demonstrate how this theory, HTM sequence memory, can be applied to sequence learning and prediction of streaming data.

In the new paper, HTM sequence memory is compared with four popular statistical and machine learning techniques: ARIMA, a statistical method for time-series forecasting (Durbin & Koopman 2012); extreme learning machine (ELM), a feedforward network with sequential online learning (Huang, Zhu, & Siew, 2006); and two recurrent networks, long-short term memory (LSTM) (Hochreiter and Schmidhuber 1997) and echo state networks (ESN) (Jaeger and Hass 2004).
The results in this paper show that HTM sequence memory achieves comparable prediction accuracy to these other techniques. However, the HTM model also exhibits several properties that are critical for streaming data applications including:
Continuous online learning
Ability to make multiple simultaneous predictions
Robustness to sensor noise and fault tolerance
Good performance without task-specific tuning

Audi uses machine learning to refine self-parking technology (Motor Authority)

Audi has demonstrated how it is using machine learning to help refine its self-driving system.

The model, known as the Audi Q2 Deep Learning concept, features two mono cameras, facing forward and toward the rear, along with ten ultrasonic sensors positioned at points all around the vehicle. Data from these various sensors are processed by a central computer which then controls the accelerator and brake.

Over multiple parking maneuvers, an algorithm identifies successful actions, thus continually refining the parking strategy. Eventually, the model is able to park itself even in difficult situations.

Uber Acquires Tiny Mysterious Startup To Boost AI-Machine Learning (University Herald)

Uber this week has acquired Geometric Intelligence, a small two-year-old artificial intelligence startup that competes with the likes Google and Facebook in the emerging world of artificial intelligence.

According to TechCrunch, the 15 employees of the AI startup will now join the new Uber AI Labs in San Francisco, serving as the vast central AI lab for the ride-sharing mobile app.

The ubiquitous ride-sharing app envisions a future in which a fleet of vehicles can make the most complex decision without the help of a driver.

Microsoft launches its latest artificial intelligence chatbot on Kik (Daily Mail)

In March, Microsoft was forced to shut down its chatbot, Tay, after the system became corrupted with hate speech.
But the firm looks to be taking a second shot at a chatbot, with the launch of its new bot, Zo.

Unity poaches Uber’s machine learning head to tackle AI in AR/VR (TechCrunch)

Unity has hired Dr. Danny Lange to take on the next generation of AI and machine learning problems at the game engine startup. For Lange, this comes after just over a year as Uber’s head of machine learning. Previously Lange worked on machine learning products for Amazon and Microsoft.

HPE’s big data solutions add machine learning and natural language into the mix (Silicon Angle)

One of the concepts Veis talked about was “analyze in place,” or ways customers can get the value out of their datalake. With Vertica, they get the performance of a Vertica front end with the economy of Hadoop. Vertica’s new Frontloader was released in September; it added MS Azure support, in addition to AWS. So if companies want to ship most or all of their Vertica to the public cloud, they now have that option, Veis explained.

The latest version of IDOL has a new feature called ‘natural language question answering’ — users can ask the computer questions, and it answers back in a humanistic way. Unlike the consumer products Siri and Amazon Echo, which use a verbal interface, IDOL creates an “answer server,” drawing from 500 different data sources and a thousand different file formats. As is the nature of machine learning, it will get smarter over time, using cognitive functions.

Machine learning enables predictive modeling of 2-D materials (EurekAlert!

In a study published in The Journal of Physical Chemistry Letters, a team of researchers led by Argonne computational scientist Subramanian Sankaranarayanan described their use of machine learning tools to create the first atomic-level model that accurately predicts the thermal properties of stanene, a two-dimensional (2-D) material made up of a one-atom-thick sheet of tin.

Microsoft tunes R Server 9.0 for machine learning (InfoWorld)

Today, Microsoft rolled out R Server 9.0, outfitted with features that reflect R’s usefulness in machine learning and as a server.

Algorithms aren’t much good without a data source, so R Server adds support for Spark 2.0, the long-awaited upgrade to the in-memory data framework.

Machine-learning background helps PhD student at U of T design faster test for leukemia patients (University of Toronto)

A new, rapid gene-expression test could help clinicians determine the best management for patients with acute myeloid leukemia (AML) by making it possible to accurately predict a patient’s response to chemotherapy within one to two days of diagnosis.

Azure Machine Learning Cheat Sheet (Microsoft)

The most well-known concept of the Regression Algorithm is linear regression, which relates one – or multiple – independent variables to a single dependent variable to make assessments and future predictions. Least squares linear regression (or Bayesian regression in a quadratic setting) is the most used form of linear regression –because it minimizes the sum of the squares of the points outside of a tested curve to make a model that is representative of the data points in a certain set.

azure linear regression

Starbucks To Add 12,000 Cafes; Use Artificial Intelligence In Ordering App (Consumerist)

Additionally, the company says that it will soon unveil a new ordering system, dubbed My Starbucks Barista, that adds artificial intelligence to the company’s mobile order and pay app.

With the app customers will be able to place their orders via voice command or messages.

Computer Vision/Machine Vision

Many algorithms also present potential solutions to a classification problem. A common concept involves clustering data points, using the K-Means algorithm. Each cluster is defined by a cluster center, which assigns each data point to a cluster. The optimal part about this model is how it recalculates the cluster center after assigning each data point.

Autonomous robot successfully delivers takeout for the first time (Vision-Systems)

Navigating the sidewalks of London using a multi-component vision system, an autonomous robot developed by Starship Technologies has successfully delivered takeout to a customer in London, the first time such a task has been completed by a robot.

Just Eat, a UK food delivery service, sent the robot to Turkish restaurant Taksim Meze to pick up an order of falafel and lamb cutlets. The robot then made its way back to the customer in a locked compartment, according to CNET.

Privacy groups: Amazon Go takes invasive technologies to a ‘whole new level’ (The Inquirer)

PRIVACY GROUPS have spoken out about Amazon Go and said that the cashier-less shopping initiative takes privacy invasion to a “whole new level”.

“With advances in machine learning, artificial intelligence, and machine vision comes greater requirements for legal and consumer protection for us in the physical world. The outdated premise of data protection is that people come to technology. In today’s world, technology is hidden in our environment and in everyday objects.

“Companies who manufacture and promote this technology need to be upfront about the consequences, including negatives, for the societies in which they operate. Societies must understand not only the surface features of new technology but also the unexpected consequences of environments and services being dependent on personal data.

Echoing Tynan’s remarks, the Open Rights Group (ORG) has also called for Amazon to be upfront about how they are going to use the data that it slurp through its Go initiative, adding that this information should be binned once consumers’ have left the shop.

Two FSU professors earn prestigious honor from engineering association (Florida State University News)

Whalley was recognized for his contributions to architectural and compilation techniques to meet the constraints of embedded systems. He focuses on techniques for microprocessors embedded in products such as appliances, cars and smartphones.

Srivastava, also a distinguished research professor at Florida State, was recognized for his contributions to differential geometric and statistical techniques in shape analysis. Those techniques have proven valuable in several scientific disciplines, including medical diagnosis, human biometrics and computer vision.

MVTec successfully launches HALCON 13 at VISION and worldwide (Manufacturing Tomorrow)

MVTec Software GmbH (www.mvtec.com), the leading supplier of machine vision software, successfully launched the latest version of its flagship product HALCON.

In live demonstrations and expert presentations at its booth, MVTec exhibited market-ready solutions for many of the latest machine vision trends discussed at VISION. In addition to the Industrial Internet of Things (Industry 4.0) and the integration of machine vision and other disciplines, such as programmable logic controllers (PLC), these trends include embedded vision, the use of new technologies such as deep learning, as well as faster, more robust and easier imaging processes.

DSP applications focus on IoT and computer vision to stimulate new industrial growth (DigiTimes)

To improve the “smartness” of electronic devices, the crown jewel of the DSP technology application is computer vision and image recognition. DSP core used to process the input and analysis of large amounts of image data for automobile solutions in pedestrian detection and parking space tracking has attracted much media attention.

From the point of view of image resolution, HD, UHD, 4K and even 8K streaming of videos with frame rates from 30fps to 60 fps has become mainstream. The challenge is that large amounts of data have to be processed and stored. The DSP core from the fifth-generation CEVA XM6 has enhanced processing abilities and achieved low power loss, which is important to mobile devices. As a result of joint efforts with partners through the design and application of 3D images and neural network systems, the product has higher smart recognition quality for future smart applications.

Finnish researchers develop hyperspectral smart phone camera (IMV Europe)

Scientists from the VTT Technical Research Centre of Finland have developed a hyperspectral camera for smart phones. In what the researchers say is a world’s first, the mobile camera is anticipated to bring low-cost spectral imaging to consumer applications such as sensing food quality or monitoring health.

Optical spectral imaging offers a versatile way of sensing various objects and analysing material properties. Hyperspectral imaging provides access to the optical spectrum at each point of an image, enabling a wide range of measurements.

Datalogic Receives Top Honor At Inspect Awards 2017 (Newswire Today)

Datalogic, a global leader in Automatic Data Capture and Industrial Automation markets, and world-class producer of bar code readers, mobile computers, sensors, vision systems and laser marking equipment, was awarded first prize in the Inspect Awards 2017 for IMPACT+ OCR, a solution dedicated to the optical recognition of printed characters on product packaging in the food industry.

IMPACT+ OCR delivers rapid and easy variable data inspection and does not require programming skills for vision machine systems. It is extremely user-friendly software with a step-by-step configuration approach. Key features include multiple OCR reading regions, the ability to memorize different inspection recipes and a customizable operator interface.

Specifically suited for OCR applications in the food industry, IMPACT+ OCR (datalogic.com) guarantees the reading and checking of batch numbers, expiry dates, serial numbers, and any alphanumeric string. Combined with thermal transfer printers, IMPACT+ OCR ensures high printing quality, and increases the safety and traceability of food and beverage packaging processes.


Allen Taylor

About the author

Allen Taylor

An award-winning journalist and former newspaper editor, I currently work as a freelance writer/editor through Taylored Content. In addition to editing VisionAR, I edit the daily news digest at Lending-Times. I also serve the FinTech and AR/AI industries with authoritative content in the form of white papers, case studies, blog posts, and other content designed to position innovators as experts in their niches. Also, a published poet and fiction writer.

Add Comment

Click here to post a comment

Your email address will not be published. Required fields are marked *