Tag Archives: Computing

What are RTUs – Remote Terminal Units?

Nowadays, small computers make up remote terminal units or RTUs and SCADA units. Users program controller algorithms into these units, allowing them to control sensors and actuators. Likewise, they can program algorithms for logic solvers, power factor calculators, flow totalizers, and many more, according to actual requirements in the field.

Present RTUs are powerful computers able to solve complex algorithms or mathematical formula describing external functions. Sensing devices or sensors gather data from the field, sending the signals back to the RTU. By solving the algorithms in it using the input signals, the RTU then sends out control instructions to valves or other control actuators. As scan periods in RTUs are very small, the entire activity happens very fast, hardly taking a few milliseconds, with the RTU repeating the process.

Regulatory agencies certifying RTUs prefer use of dedicated hardware for solving certain safety related functions such as toxic gas concentration and smoke detection. Therefore, they make sure of the reliability of detection for safety related functions.

The RTU operates in a closed system. Sensors measure the process variables, while actuators adjust the process parameters and controllers solve algorithms for controlling the actuators in response to the measured variables. The entire system works together based on wiring or some form of communication protocol. This way, the RTU enables the field processes near it to operate according to design.

Before the controller in the RTU can solve the algorithm, it has to receive an input from the field sensor. This requires a defined form of communication between the RTU and the various sensors in the field. Likewise, after solving the algorithm, the RTU has to communicate with the different actuators in the field.

In practice, sensors usually feed into a master terminal unit or MTU that conditions their input, changing it to the binary form from the analog form, if necessary. This is because sensors may be analog or digital types. For instance, a switch acting as a sensor can send information about its state using a digital one or +5 V when it is open and a digital zero or 0 V when it is closed. However, a temperature sensor has to send an analog signal or a continuously varying voltage representing the current temperature.

The MTU uses analog to digital converters to convert analog signals from the sensors to a digital form. All communication between the MTU and the RTU is digital in nature, and a clock signal synchronizes the communication.

The industry uses RTUs as multipurpose devices for remote monitoring and control of various devices and systems, mostly for automation. Although industrial RTUs perform similar function as programmable logic circuits or PLCs do, the former operates at a higher level as RTUs are basically self-contained computer units, containing a processor and memory for storage. Therefore, the industry often uses RTUs as intelligent controllers or master controller units for controlling devices that automate a process. This process can be a part of an assembly line.

By monitoring the analog and digital parameters from the field through sensors and connected devices, RTUs can control them and send feedback to the central monitoring station for industries dealing with power, water, oil, and similar distribution.

What happens when you turn a computer on?

Working on a computer is so easy nowadays that we find even children handling them expertly. However, several things start to happen when we turn on the power to a computer, before it can present the nice user-friendly graphical user interface (GUI) screen that we call the desktop. In a UNIX-like operating system, the computer goes through a process of booting, BIOS, Master Boot Record, Bootstrap Loading, grub, init, before reaching the operating level.


As soon as you switch on the computer, the motherboard initializes its own firmware to get the CPU running. Some registers, such as the Instruction Pointer of the CPU, have permanent values that point to a fixed memory location in a read only memory (ROM) containing the basic input output system (BIOS) program. The CPU begins executing the BIOS from the ROM.


The BIOS program has several important functions, which begin with the power on self-test (POST) to ensure all the components present in the system are functioning properly. POST indicates any malfunction in the form of audible beeps. You have to refer to the Beep Codes of the motherboard to decipher them. If the computer passes the test for the video card, it displays the manufacturer’s logo on its screen.

After checking, BIOS initializes the various hardware devices. This allows them to operate without conflicts. Most BIOSs follow the ACPI create tables for initializing the devices in the computer.

In the next stage, the BIOS looks for an Operating System to load. The search sequence follows an order predefined by the manufacturer in the BIOS settings. However, the user can change this Boot Order to alter the actual search. In general, the search order starts with the hard disk, CD-ROMs, and thumb drives. If the BIOS does not find a suitable operating system, it displays an error. Otherwise, it reads the master boot record (MBR) to know where the operating system is located.

Master Boot Record

In most cases, the operating system resides in the hard disk. The first sector of the hard disk is the master boot record (MBR), and its structure is independent of the operating system. It consists of a special program, the bootstrap loader, and a partition table. The partition table is actually a list of all the partitions in the hard disk and their file system types. The bootstrap loader contains the code to start loading the operating system. Complex operating systems such as Linux use the grand unified boot loader (GRUB), which allows selecting of one of the several operating systems present on the hard disk. Booting an operating system using GRUB is a two-stage process.


Stage one of the GRUB is a tiny program and its only task is to call stage two, which contains the main code for loading the Linux Kernel and the file system into the RAM. The Kernel is the core component of the operating system, remains in the RAM throughout the session, and controls all aspects of the system through its drivers and modules. The last step of the kernel boot sequence is the init, which determines the initial run-level of the system. Unless otherwise instructed, it brings the computer to the graphical user interface (GUI) for the user to interact.

Tinker Board: Raspberry Pi Competitor from ASUS

The community of single board computer users is passionate and the DIY enthusiasts are growing daily. While they are infatuated with the amazingly tiny package called the Raspberry Pi (RBPi), they are constantly clamoring for more performance and connectivity features. This demand has produced several competitors to the RBPi, and the tech giant, ASUS Computers is now providing one in the form of a Tinker Board.

The ASUS Computers product is a mini-PC based on the ARM core, and its actual model number is the ASUS 90MB0QY1-M0EAY0. However, it is easier to remember it as the Tinker Board. The smart name from ASUS for the product is the exact demographic of its intention, offering a tiny, all-in-one product for makers and tinkerers, to use in media servers, fun projects, and embedded applications. For instance, the Tinker Board allows one to build a personal NES Mini alternative.

Although a 64-bit ARM Cortex-A47 quad-core processor, the Broadcom BCM2837, powers the RBPi3 at 1.2 GHz, a 32-bit ARM Cortex-A17, the quad-core Rockchip RK3288 processor powers the Tinker Board, operating at 1.8 GHz. ASUS claims the Tinker Board is almost twice as fast as the RBPi3 model B. Additionally, against the 1 GB RAM configuration of the RBPi3, the Tinker Board offers 2 GB of RAM.

The Tinker Board has other advantages as well. The hardware includes the complete H.264 4K video decode capability, supported by a far stronger graphics performance from the ARM Mali-T764 with a graphics core of the Rockchip RK3288. The audio capabilities are also better with the Asus minicomputer offering audio sample rates at 192K/24-bit, while the RBPi3 offers only 48K/16-bit, which necessitates an add-on board for HD audio from the RBPi3.

The integrated, Gigabit Ethernet port at full speed on the Tinker Board gives it a substantial boost over the 100 Mb LAN on the RBPi3. Similar to that available on the RBPi3B, the Tinker Board also has an 802.11b/g/n Wi-Fi and Bluetooth 4 capability. In addition, it has support for SDIO 3.0, and offers swappable antennas for the built-in 802.11 b/g/n Wi-Fi module.

Similar to the RBPi3B, the Tinker Board also supports the Debian Linux (modified by ASUS) operating system and KODI, with its slick media streaming interface. Similar to the RBPi, the Tinker Board also comes with no on-board storage, and you have to use a micro SD card. However, the additional capabilities on the Tinker Board make it about twice as expensive as compared to the market price of the RBPi3B.

Physically, both single board computers are of the same size, with mounting holes in the same position. Obviously, ASUS wants the Tinker Board to be a drop-in replacement for the RBPi3. The same configuration of the GPIO pins for both boards lends further support to this credence.

The RBPi concept has spawned a whole new era of tiny computer devices, selling in several schools, colleges, and universities. Many other device manufacturers have since piled on and released their own version of the credit-card sized powerhouse.

In this chaotic, crowded environment, the specifications of the Tinker Board, although not ground breaking, could play nicely in the existing RBPi-based projects.

How Can I Protect My Raspberry Pi?

By connecting the Single Board Computer to the Internet, you actually run the risk of compromising your Raspberry Pi or RBPi to different types of attacks from malicious persons. However, as several advantages of an Internet connection far outweigh such risks from attackers, there is merit in looking for ways to mitigate them. Spain Hardware from Madrid is venturing on a Kick Starter project to enable hardware protection for the RBPi.

When your RBPi requires secure communication, you can rely on the PiSec module, from Spain Hardware, to provide the necessary assistance. PiSec, being a protecting module, uses its own hardware to protect and encrypt all the inputs and outputs on the RBPi. PiSec protects the RBPi from all angles – SD card, USB, and Ethernet, offering a strong hardware base security that includes Elliptic curves and AES-256 XTS.

PiSec, based on a True Random Number Generator, works by generating safe and strong encryption keys and certificates. Internally, PiSec uses a protected file system that it protects with an internal certificate making it impervious to unauthorized access. The processor on board the PiSec module makes use of Elliptic Curve Cryptography to reduce its own overhead and speed up the process of verification.

PiSec provides protection complying with certificates such as the AES 256-bit XTS Military Grade Encryption and X.509. Repeated attempts after a predefined number of unsuccessful attempts to gain access to the RBPi results in the PiSec automatically blocking access. This helps in preventing DoS or Denial-of-Service and brute force attacks.

Typically, you can use your RBPi right out of its box, including its Ethernet connection, the USB ports, and its SD card. You can use the SBC to collect, store, and transfer data, but the RBPi handles all this using clear text, which anyone can intercept and read. You can use your tiny but powerful computer in several ways, for instance, as a standalone PC as a storage system, data logger, and standalone server, a device to control complex systems/machines, or used with licensed software. In all these cases, it will certainly hurt your business if your data is exposed and someone sniffs the actuator or the sensor communication lines and steals your telemetry.

There are several ways to achieve security through software generated keys and certificates. However, relying on a hardware solution is a far better solution, as most of such software solutions do not use a true random generated number. PiSec offers this strong protection security to the entire RBPi, including all devices on its SPI bus, without overloading the processor of the RBPi, nor collapsing its OS. Being a hardware solution, it is simple enough to plug the PiSec on your RBPi, without any necessity of a learning curve or any previous experience on security.

Features of the PiSec include a TRNG or true random number generator. It obtains the random seed from on-board white noise generators that are FIPS and AIS 31 compliant, and with a very high entropy level. TRNG is crucial to creating strong secure keys and certificates.

PINE64 : A 64-bit Contender for the Raspberry Pi

Earlier, a DIY computing project could cost an enthusiast hundreds of dollars. Now, with single board computers such as the Raspberry Pi or RBPi or its latest kin, the Raspberry Pi Zero, anyone can start a new project at the cost of a cup of coffee. Seen from the other side of the fence, a competitor has to include a better choice of components, offer a better price or both. PINE64 Inc. has taken the third route.
PINE54 Inc. is attempting to improve on the legacy so far built up by the RBPi. According to the team, two mathematical constants make up the name of their board – Pi and Euler’s Number e. As it has a 64-bit processor, the name also includes the number 64 along with an A to differentiate it from future versions. The PINE A64 runs on an ARMv8 processor, the Cortex-A53, and is available for just $15.
PINE A64 measures 12.7×7.94cms and uses a 64-bit processor, a quad-core ARM Cortex A53 running at 1.2GHz. A dual-core Mali 400 MP2 handles the graphics. Memory includes a micro SD slot to handle cards up to 256GB and 2GB DDR3 SDRAM onboard. Ports available on the PINE A64 include one gigabit Ethernet, two USB 2.0, one HDMI 1.4 connector for 4K output, a stereo mini-jack connector and a charging circuit for a 3.7V Lithium battery.
PINE64 Inc. will also be offering separate modules to augment the functionality of PINE A64. The modules will add a touch panel port, a 5MP camera port, Bluetooth 4.0 and Wi-Fi connectivity and a 4-Lane MIPI video port. The board runs on 5V power via its micro USB connector, but can fall back on its internal battery with on-board power management.
According to Johnson Jeng, the co-founder of PINE64 Inc., the company has designed a simple, smart and affordable computer. People can use this to bring their next big ideas to life. The 64-bit quad-core single board computer is available at an exceptional price. It is compatible with several open-source platforms, enabling people to build a community of innovation and creativity.
Just like other ARM-based single board computers, you can set up PINE A64 to operate as a gaming console or a mini-computer. You can control your connected home or allow it run your own media center. PINE A64 can operate with Android 5.1, openHAB, Ubuntu Linux, OpenWRT and Kodi. Additionally, it supports Miracast and offers the H.265 video standard to give your 4Kx2K output.
The Raspberry Pi Foundation concentrates on delivering performance without increase in costs, and hence, prefers to retain the ARMv7 architecture for the RBPi family even when ARMv8 64-bit chips are readily available. According to Eben Upton, the founder of the RBPi series, a more powerful processor will certainly come with a boost in the prices.
With companies now launching new Systems-on-a-Chip or SoC platforms that are 64-bit and super-cheap, PINE64 Inc. has decidedly stolen a march over the RBPi series. Allwinner started this trend with the 64-bit Cortex A53 processor for their tablets and now PINE64 Inc. has used it to power their PINE A64, A64+ and A64+ 2GB boards.

Computer Translated Sign Language

There are many people in the world who cannot hear because their hearing ability is impaired. This disability also precludes them from holding audible conversations with others. For a long time, telephone calls dominated the long-distance communication scenario. However, over the past couple of decades, other means of communication have also evolved, such as emails and text-based messaging. Although these supplement voice calls largely, the problem of face-to-face communication with the deaf still remains.

Using sign language is one means of face-to-face communication that the hearing-impaired use and this is as efficient as their methods of communication using smartphones, tablets and computers. Similar to using any other language, two people can communicate face-to-face only when both are capable of using the sign language. Lately, communication between two individuals is now easier because of the use of translators in computers. This allows the user to understand even when they do not understand the spoken language.

Now MotionSavvy is using the same technology for translating sign language to another language that the user can understand. They are using a dedicated tablet, Uni, created to enable efficient two-way communication between those who can hear and those whose hearing is impaired.

Uni involves the use of two distinct technologies. First, it monitors sign language by using integrated cameras and interprets the signs using a special recognition software. Then it translates the signs into spoken words. The other part of the technology involves converting spoken words into text. This happens when the other person responds by speaking. Uni converts this speech into text, displaying it on the screen for the deaf person to read.

The World Federation of the Deaf claims there are more than 70 million deaf people in the world. With the technology offered by MotionSavvy, there is a dramatic potential to influence the lives of such people.

MotionSavvy is launching Uni with the ability to read at least 2000 signs initially. They will be issuing updates for adding more signs. However, they are offering SignBuilder software, with which users can configure new signs.

Uni is available in two versions – hardware and software. You can buy the hardware device that includes the software, or the software alone. You can use the software-only solution on a computer that has the Leap Motion controller. For both solutions, users need to pay a monthly subscription that allows them access to SignBuilder and CrowdSign.

The basic Uni Dictionary contains about 2000 signs. Although this confers the ability to hold meaningful conversations, individuals can add new vocabulary to Uni Dictionary with the help of SignBuilder. They can also share the new signs with others on the Uni network, by using the software CrowdSign. MotionSavvy expects the number of signs to grow exponentially with people using the two software programs.

At present, Uni is able to recognize signing by hands in front of its camera. Eventually, MotionSavvy expects to implement recognition of all facial emotions. Uni is working for people using signs such as CASE or SEE. MotionSavvy is working on improving recognition and adding more features to accommodate culturally strong ASL users as well.

Android vs. Linux – Which OS is better?

Is Android A Better OS Than Linux?

Android has established itself as an important operating system for mobile devices. Google developed Android as an open source OS based on the Linux kernel. Google selected the Linux kernel because of its proven driver model, existing drivers, process and memory management, networking support and several other core operating system services. However, the Google team had to make several changes to make Android capable of operating mobile devices successfully. Differences with standard Linux are highlighted here.

The target architecture

Although the Linux kernel supports several architectures, right now, Android supports only two: ARM and x86. The ARM platform is more prevalent on mobile phones while the Android-x86 targets mainly the Mobile Internet Devices or MIDs used for general-purpose desktop/laptop/server computing systems. This being the fundamental difference between the two Operating Systems, it provides a strong insight into further divergence between the two.

Modifications in the kernel

Android does not use the standard Linux kernel straightaway, but uses it with some enhancements. These include alarm driver, shared memory driver, inter-process communication interface, power management, low memory killer, kernel debugger and logger. Google has contributed all the kernel enhancements back to the open source community under GPL.

Bionic C library

The GNU C library used by most Linux distributions makes use of the Native POSIX Thread Library or NPTL, which offers high performance, especially in server applications. However, disk space footprint and memory requirements of NPTL are far too large for resource-limited systems such as mobile devices.

This led Google to create a new C library called Bionic. It has fast execution paths, avoids edge cases and remains a simple implementation. As mobile devices are single user systems, for security reasons Google has removed the settings for groups and passwords, keeping only a unique user id and group id. Bionic operates with the limited CPU and memory resources available on Android platforms.

The Dalvik Virtual Machine

Android uses a virtual machine to run applications. Most top cell manufacturers such as Samsung, Motorola and Nokia use J2ME, a mobile optimized version of the Java virtual machine. In contrast, Android uses the Dalvik Virtual Machine, which is a standard Java platform. The dex files used by Dalvik are more compact and optimized to perform well on mobile devices with slow CPUs, limited memory, no swap space and limited battery power.

File system

Most desktop/laptop/server applications use magnetic hard disks, which the standard Linux systems manage with the latest Ext journaling file system. However, magnetic drives are physically too large, too fragile and consume too much power. To provide a robust file system, embedded systems use solid-state memory devices such as NOR for code execution and NAND for storage. Block erasure and memory are important features of solid-state memory, which the Ext file system does not handle. Therefore, Android uses an optimized Linux flash file system called YAFFS and this deals with lifetime limitations, bad block management and error correction for maintaining data integrity in NAND flash systems.

Power management

Standard Linux systems manage power though APM or ACPI. Android does not use either, relying more on its own PowerManager module, which is a Linux power extension. The module has low-level drivers for controlling the peripheral supported such as screen display and backlight, keyboard backlight and button backlight.

Neo Smartpen N2 Connects with Bluetooth

Although computers and keyboards have taken out much of the efforts of writing, some situations still demand we keep this skill alive. Then, some people are unwilling to give up the feeling of writing with a pen to pounding on a keyboard. Engineers have tried to modernize the humble writing instrument with the Bluetooth pen of Livescribe. Now, an improved Smartpen N2 is in the market.

Neo Smartpen N2 has a sleeker design compared to that of Livescribe. According to the manufacturer, the pen has a shape users will find more comfortable and it is lighter than most smartpens in the market. Without the cap, Neo Smartpen N2 is only 22gms as against the Livescribe, which weighs 34gms. Although the difference is not much, to someone who writes extensively with a pen, this could count for a lot.

An ARM 9 dual-core Processor powers the Smartpen N2, which sports a built-in 90MB NAND flash drive. The pen connects via Bluetooth to a tablet or phone. However, it works even without them. N2 has a built-in camera that captures 120 pictures-per-second while recording about 1,000 pages of notes to store in its memory. Later, you can synchronize this content over to another device.

To conserve battery, Smartpen N2 turns itself off automatically when it detects idle time and turns on to be ready for writing. This convenient feature helps to conserve battery and the pen can write for about five continuous hours before it has to be recharged. A full recharge takes about two hours.

The entire Smartpen N2 writing system has three parts. The first is the pen itself, to be followed with the special paper, which records the motion of the pen. Then there is the app, which translates these motions into an image on the tablet’s screen. The app can also send the notes to popular services such as Dropbox or Evernote. Neo Notes app is available for free for Android and iOS phones and tablets.

Both Livescribe 3 and Smartpen N2 translate their ink notes scrawled on special paper for capturing them in digital form. However, the Equil Smartpen 2 uses a sensor that you can clip onto the top of any kind of paper you are writing on. While both Equil and Neo N2 are cross-platform compatible, apps available for Android and iOS, Livescribe 3 remains an iOS-only device.

Apart from recording written notes in the form of images, Neo N2 can also record voice memos in real-time, simultaneously as you write. Other features of this amazing pen include translating features that convert handwritten notes into text, after you have selected the language. Additionally, you only have to draw a check mark on the mail icon in the corner of your page and the app will email the page attached as a PDF. At the same time, the app will synchronize any new notes you make automatically to your Evernote account.

The only thing limiting the appeal of Neo Smartpen N2 is its need for special paper. Therefore, this is a device for serious writers only and not meant for scribblers.

IDEASTICK: Windows Goes Into Your Pocket Now

At last, users of the Operating System Windows will also be able to enjoy the simple portability that Linux users already have. Lenovo has come up with an oversized memory stick – the new Stick 300. Actually, instead of being just an oversized memory stick, Stick 300 is full-fledged Windows PC. Although the specifications are rather low-end, the ideastick from Lenovo makes it up with being portable and having a more appealing price tag.

Obviously, the tiny chassis cannot offer an exciting hardware. However, Stick 300 runs on an Intel Atom Processor, Z3735F, with 2GB of RAM and has 32GB of storage. And, with the ideastick Stick 300, you can transform your HDMI-TV or monitor into a fully functional Windows PC. Since Stick 300 is only 100x38x15mm in dimension, it is portable and affordable. You can easily take it along when on vacations and use it as a media hub.

Initially, Stick 300 will ship with Windows 8.1, but it will be eligible for a free upgrade to Windows 10. You do not need to bother about connectivity, as Stick 300 has both Wi-Fi 802.11 b/g/n and Bluetooth 4.0 built-in. It also has a micro SD card slot and a tiny speaker.

Stick 300 is comparable to other products in the market. This includes Compute Stick from Intel, which they had released in March and an Ubuntu Linux powered lower-end option. Therefore, if you are in the market looking for an ultra-portable Windows solution, Stick 300 is a simple and functional option.

On its side, the Stick 300 has a USB 2.0 socket, which allows you to use your keyboard and mouse wirelessly; that is, if your monitor or TV is not touch-enabled. Powering up is through a second micro USB port. The hardware included is good enough for browsing the web, watching Netflix and even doing some light gaming. Therefore, instead of lugging along a hefty notebook on the road, you can conveniently carry the Stick 300 and plug it into a TV in a hotel for a spot of catching up.

In comparison, Compute Stick from Intel is also an entire PC crammed inside an HDMI stick, which you can fit in your palm. The Compute Stick instead uses an Intel Atom quad-core processor. It has a full-fledged USB 2.0 port on one side and a micro USB port on the other for powering up. However, the Lenovo Stick 300 is $20 cheaper.

If you can do with somewhat lower specifications, there is another stick with the Ubuntu 14.04 LTS OS on it. Since it has 1GB RAM and 8GB internal storage, the Ubuntu stick is less expensive than those from Lenovo or Intel. For keeping it cool, the Ubuntu stick has vents on top and sides. It also has a tiny fan for circulating air.

You may have to use the included HDMI cable to connect the stick to your TV, as most TVs do not have much space surrounding the HDMI socket that will accommodate the width of your portable stick computer. In addition, since all these sticks have only one USB port, you will need a unifying wireless solution – such as from Logitech – to get both your keyboard and mouse connected.

Computers Can Beat Humans in Image Recognition

Every day, computers are getting smarter. So far, it is not clear whether the smartness is moving towards something as depicted in the Terminator movies, but computers are beating humans in chess, poker and Jeopardy. The next hurdle that computers have crossed is image recognition. Microsoft claims to have programmed a computer that can beat humans at recognizing images.

Although the final competition is going to be held on December 17, 2015, already there are claims that computers are better than humans are in visual recognition. The ImageNet Large Scale Visual Recognition Challenge will do judging for the final competition. The first claim about computers beating humans came from Microsoft. They claimed that while humans made 5.1% errors in recognizing images, computers failed only in 4.94% cases. After 5 days of Microsoft announcing their feat, Google announced that they have bettered the Microsoft claim by 0.04%. That means the competition is getting fiercer every day.

Since 2010, more than 50 institutions take part every year in the competition for image recognition. ImageNet runs this competition and they have hundreds of object categories and several millions of example images. So far, humans have scored the most, but this year a computer is expected to take the crown. Typically, contestants use the latest deep learning algorithms. Derived from different types of artificial neural networks, these deep learning algorithms mimic the way the human brain works to a varying degree.

Although no contestant actually offers their exact code, they provide papers that freely describe their algorithm in great detail – similar to the spirit of open source – explaining the advantages of their algorithm and why it is expected to work so well. As Microsoft explains in their paper, they are using deep CNNs or convolutional neural networks that have 30 weight layers. Google have revealed that they are using batch normalization techniques, and these do not allow neurons to saturate during initialization.

Usually, the conventional way of using neural units involves hand designing them and fixing while training. However, Microsoft has deviated from this path and made the neural units smarter. They have done this by making their form more flexible in nature. According to the principal researcher at the Visual Computing Group of Microsoft Research, Asia, each neural unit undergoes a particular form of end-to-end training that imparts the learning. The introduction of smarter units improves the model considerably.

However, the reason for the ability of current neural networks being able to beat human experts lies in the algorithm of Microsoft’s Deep Learning. This algorithm usually initializes and trains on 1.2 million training images and verifies its learning on 50-thousand validation images. For the final application of its learning, Deep Learning uses 100-thousand test images from the main image database. However, Microsoft did not actually follow this standard route.

As training of very deep neural networks is rather difficult, Microsoft used a robust initialization method. As with other contestants, Microsoft did buy Nvidia’s access to their arrays of graphic processing units. However, they also bought and configured their own supercomputer. They simulated parametric rectified linear neural units and that helped them finally to beat the human experts for image classification.