What is Neural Networks?

What is Neural Networks?

An Artificial Neural Network (ANN) is an information processing paradigm that is inspired by the way biological nervous systems, such as the brain, process information. The key element of this paradigm is the novel structure of the information processing system. It is composed of a large number of highly interconnected processing elements (neurons) working in unison to solve specific problems. An ANN is configured for a specific application, such as pattern recognition or data classification, through a learning process. Learning in biological systems involves adjustments to the synaptic connections that exist between the neurones. This is true of ANNs as well.

Neural network simulations appear to be a recent development. However, this field was established before the advent of computers, and has survived at least one major setback and several eras.

Why use Neural Networks?

Neural networks, with their remarkable ability to derive meaning from complicated or imprecise data, can be used to extract patterns and detect trends that are too complex to be noticed by either humans or other computer techniques. A trained neural network can be thought of as an "expert" in the category of information it has been given to analyse. This expert can then be used to provide projections given new situations of interest and answer "what if" questions.

Other advantages include:

1. Adaptive learning: An ability to learn how to do tasks based on the data given for training or initial experience.
2. Self-Organisation: An ANN can create its own organisation or representation of the information it receives during learning time.
3. Real Time Operation: ANN computations may be carried out in parallel, and special hardware devices are being designed and manufactured which take advantage of this capability.
4. Fault Tolerance via Redundant Information Coding: Partial destruction of a network leads to the corresponding degradation of performance. However, some network capabilities may be retained even with major network damage.

Neural Networks versus Conventional Computers

Neural networks take a different approach to problem solving than that of conventional computers. Conventional computers use an algorithmic approach i.e. the computer follows a set of instructions in order to solve a problem. Unless the specific steps that the computer needs to follow are known the computer cannot solve the problem. That restricts the problem solving capability of conventional computers to problems that we already understand and know how to solve. But computers would be so much more useful if they could do things that we don't exactly know how to do.

Neural networks process information in a similar way the human brain does. The network is composed of a large number of highly interconnected processing elements(neurons) working in parallel to solve a specific problem. Neural networks learn by example. They cannot be programmed to perform a specific task. The examples must be selected carefully otherwise useful time is wasted or even worse the network might be functioning incorrectly. The disadvantage is that because the network finds out how to solve the problem by itself, its operation can be unpredictable.

On the other hand, conventional computers use a cognitive approach to problem solving; the way the problem is to be solved must be known and stated in small unambiguous instructions. These instructions are then converted to a high level language program and then into machine code that the computer can understand. These machines are totally predictable; if anything goes wrong is due to a software or hardware fault.

Neural networks and conventional algorithmic computers are not in competition but complement each other. There are tasks that are more suited to an algorithmic approach like arithmetic operations and tasks that are more suited to neural networks. Even more, a large number of tasks, require systems that use a combination of the two approaches (normally a conventional computer is used to supervise the neural network) in order to perform at maximum efficiency.

Related Reading: Applications of Neural Networks.

Related Reading: How Neural Network is used in Intelligent Houses?
Related Reading: How Neural Network is used in Intelligent Devices?

What is Artificial Intelligence?

What is Artificial Intelligence?

Artificial Intelligence a.k.a. A.I. is the science and engineering of making intelligent computer programs or machines. Artificial intelligence algorithms span several different branches of computer science and mathematics including: pattern recognition, predictive modeling, text mining and search, genetic programming, heuristics, inference, and ontology, and data analytics.

Artificial intelligence is commonly referred to as machine learning and was originally developed to enable computers to learn. Today, the technology is based on a number of advanced mathematical methods for optimization, regression and classification and finds application in a wide variety of fields including: gaming, speech recognition, computer vision, expert systems, heuristic classification, medical diagnostics, and credit card fraud.

What is Predictive Modelling?

Predictive modeling is the area of data analytics concerned with forecasting probabilities and trends. A predictive model is made up of a number of predictors, or variables, that are likely to influence future behavior or results. In marketing, for example, a customer's gender, age, and purchase history might predict the likelihood of a future sale.

Predictive modeling techniques are often iterative involving the collection of data, the formulation of a statistical model, and the approximation of an outcome. The process is refined and validated as more data becomes available. The model may employ a simple linear equation or a complex artificial intelligence algorithm, mapped out by sophisticated software.

Predictive modeling algorithms are used widely in information technology (IT). Applications of predictive modeling include: spam filtering, customer relationship management (CRM), capacity planning, disaster recovery, engineering, meteorology, insurance risk, credit score, and marketing.

What is Data Analytics?

Data analytics is the science or process of inspecting, cleaning, transforming, and modeling data with the goal of highlighting useful information. Data analytics software is used in many industries to allow companies and organization to make better business decisions and in the sciences to verify or disprove existing models or theories.

Data analytics is distinguished from data mining by the scope, purpose and focus of the analysis. Data miners sort through huge data sets using sophisticated machine learning algorithms to identify undiscovered patterns and establish hidden relationships. Data analytics focuses on inference, the process of deriving a conclusion based solely on what is already known by the researcher.

What is Gradient Boosting?

Gradient boosting is a machine learning technique for regression problems, which produces a prediction model in the form of an ensemble of weak prediction models, typically decision trees. It builds the model in a stage-wise fashion like other boosting methods do, and it generalizes them by allowing optimization of an arbitrary differentiable loss function. Gradient boosting method can be also used for classification problems by reducing them to regression with a suitable loss function.

Related Reading: What is Neural Networks?
Related Reading: How AI is used to make Intelligent Homes?

Nest Intelligent Thermostat

How the Nest Intelligent Thermostat Works?

Your heating, air conditioning and the ductwork that carries and recycles air between rooms make up the HVAC (heating, ventilation and air conditioning) system for your home. You control your home's HVAC through your thermostat. All you have to do is select your heating and cooling options and to set your desired indoor temperature. The thermostat does the rest, switching systems on and off based on the temperature it detects in the room.

The Nest Learning Thermostat goes beyond this simple temperature detection to make a real impact in your HVAC energy consumption. In this article, we'll see what Nest can do, how it does what it does, who's behind it and what challenges it faces in the HVAC industry.

To understand Nest's value, let's first look at what other thermostats do. All thermostats let you set a desired temperature and monitor the current temperature. You can also switch between heat and AC.

Many thermostats rely entirely on your settings. As a result, you might only adjust the thermostat when you feel uncomfortable. In addition, you may not think to adjust it before you leave home to save on energy while the house is empty.

The Nest Learning Thermostat aims to solve this problem. Nest actually programs itself by learning your behavior patterns and desired temperatures for certain days and times during the week, and then building a schedule for your HVAC.

Nest Features: Saving Energy

Nest features what the company calls Nest Sense technology to learns your day-to-day routine and maintains your HVAC schedule automatically, based on what it learns.

Nest creates an Auto-Away mode based on what it's learned. This sets a temperature for minimal HVAC activity when you're not in the building. You can also set an Away mode manually if you wish.

While it's actively heating or cooling, Nest displays an estimated time for the system to reach the desired temperature.

Nest displays a green leaf any time the thermostat is running at energy-saving settings. This can teach you to make energy-saving decisions. For example, if Nest has learned that you typically run your AC until the house is 74 degrees Fahrenheit (23.3 Celsius), you could turn up the temperature until you see the green leaf, perhaps at 76 degrees Fahrenheit (24.4 Celsius), to save energy. The leaf will always appear at cooling settings of 84 degrees Fahrenheit (28.9 Celsius) or higher and heating settings of 62 degrees Fahrenheit (16.7 Celsius) or lower.

1. Nest lets you know what activity (between Auto-Away, your own adjustments and the weather) resulted in the greatest energy savings throughout the day.
2. Nest uses WiFi to connect to your Nest account at nest.com. This feature allows you to monitor and adjust the Nest remotely from the Web site.
3. Nest supports a mobile app available for Apple iOS devices (iPod Touch, iPhone, and iPad) and Android devices. The app turns your mobile device into a remote control for your Nest.
4. You can add Nest to any number of thermostats in a multi-thermostat building. They will work alongside other thermostats, but note that each of Nest's energy-saving features only applies to the rooms in its sensor range and to the HVAC components it controls.
5. Your Nest account can manage up to 10 Nest devices, whether they're in the same building or at multiple locations.

Nest Technology: Nest Sense

The core technology behind Nest is a combination of its sensors, computer and algorithms. The company calls this Nest Sense. Part of Nest Sense's job is to gather data to use in its calculations. This data goes beyond just measuring the temperature in the room. In fact, Nest gathers data from the following sources:

1. Three temperature sensors, designed to get a more precise measurement than a single sensor
2. Motion and light sensors that detect activity in the room at a wide 150-degree angle
3. A WiFi connection to get weather data about your area from the Internet

Using data from these sources, Nest Sense creates a schedule for your HVAC.

Related Reading: Artificial Intelligence used in Intelligent Devices
Related Reading: How Neural Networks are used in Intelligent Devices?

Intelligent Homes with Artificial Intelligence

What are these Intelligent Homes?

Whether it's a fully integrated house filled with interactive devices or a single component within an environment, there are three main pieces to creating an intelligent reactive device. First, it has to be able to detect what is going on. We humans use our senses to gather information about the environment around us and the others who may be in it. Sensors do the same thing for electronics, though a sensor may have a narrower scope of stimuli it can detect.

The interactive environment must also have some sort of processing unit that interprets the data gathered by sensors. This is the brain of the interactive environment. It stores user profiles and matches them with preferences. When a sensor detects that a specific person has entered the environment, this processing unit determines the next course of action.

The final piece is some sort of actuator, switch or setting that the processing unit engages to change the environment to best suit the user's needs. This might be a thermostat setting, a sound system or even a haptic feedback system that alerts you to specific conditions.

An interactive environment could have a central processing unit through which all components operate. This would allow for a single point of operation. All the data gathered by sensors would pass through the processing unit, which would send out commands to adjust the environment as needed.

Another approach is to use multiple, independent systems within an environment. This means that you might have an intelligent thermostat and an intelligent sound system but the two aren't connected to each other. One potential advantage of this approach is that if one component's processing unit fails, the others should still work without a problem. Interactive environments may also use a combination of systems with some integrating with a central unit while others are independent.

Today's intelligent reactive devices learn from patterns. Let's look at a thermostat as an example. Let's say your ideal temperature is 70 degrees Fahrenheit (21 degrees Celsius). When you're home, that's what you want. But let's say you're away from your home during the day. You may not care if your home gets a little warmer or cooler than 70 degrees at that point. So you set the thermostat a little higher -- or lower, depending on the time of year -- than your normal comfort zone to save power. When you come back home, you reset the thermostat and wait for your house to get comfortable again.

Many modern thermostats have a programmable mode that lets you set temperatures for particular times during the day. You could program your thermostat to adjust to a different temperature after you leave and return to your preferences an hour or so before you get back. You'll still be saving energy, but you won't have to come home to a hot or cold house.

An intelligent reactive thermostat could learn these patterns by recording when you make adjustments to the temperature. If you make a pattern of adjusting the temperature - for example, if you like it toasty in the morning but cave-cold in the evening - the intelligent thermostat can keep a record of it and make these adjustments for you once it has figured out your preferences.

The Nest thermostat does just that. It also has a motion detector built in so that it can adjust these settings on the fly. Maybe you've got a day off - something a normal programmable thermostat would be unable to determine. The Nest could detect you as you move about the house and make sure to override its normal routine so that you remain comfortable.

The Nest also has a WiFi transmitter that allows it to check weather reports. This lets the system know if it will need to work harder to maintain the ideal temperature inside the house. This adds a second layer of artificial intelligence over pattern recognition - search and learning.

The thermostat is a comparatively simple application of an intelligent environment. For example, you might have a favorite chair you sit in. Sometimes you sit there when watching television. Other times you might be reading or listening to music. In a fully automated home, sensors might be able to determine when you sit in the chair. But what does the home do next?

In general, the way AI makes decisions involves sets of actions. When you watch television, those actions could include turning on the TV and any other home entertainment equipment you have. It may also involve closing the blinds to block outside light. You might like to watch movies in a dark room, so the house dims the lights inside as well.

But if you wanted to read a book, a dim room with a blaring television may not be the environment you had wished. Instead, you might want a nearby lamp to be on while you sit in a quiet room and read. In this case, the house would need to turn off any gadgets that make noise and turn the light on for you. But how does the house know which set of actions to follow?

It sounds like a simple problem -- after all, you know if you want to watch TV, read or listen to music. But the house has to learn. It might do this by observing your behavior over several days, looking for patterns and patterns within other patterns. Otherwise, it might turn on the lamp when you really wanted the television.

This is mainly a software problem. Programmers help AI become smarter by building in a feedback system so the program keeps track of how frequently it gets things right and wrong. It gradually builds a database keyed to your behaviors so that it can anticipate your needs based on past experience. It may still get things wrong once in a while.

Things get more complicated when there are multiple people living in one house- or working in the same building. The software for the intelligent environment will have to build databases for each person and tweak them over time. And then there's the question of prioritization - if two people have drastically different preferences, how does the intelligent house take that into consideration?

Related Reading: What Artificial Intelligence is all About?
Related Reading: How Neural Networks is used in Intelligent House?

New Inks for OLED Screens

Why develop New Inks for OLED Screens?

Organic light emitting diode, or OLED, displays seem to have it all: energy efficiency and a beautiful, crisp picture that refreshes rapidly. But it's difficult to make them on a large scale, so OLED televisions remain very expensive. Using a custom-made printer from Japanese manufacturer Dainippon Screen, DuPont says it can print a 50 inch-television in under two minutes, and testing of the displays shows their performance is reliable--the displays should last 15 years.

OLEDs on the market today rely on an expensive, small-scale technique called shadow-mask evaporation to pattern the light-emitting organic molecules that make up the pixels in these displays. One of the most promising alternatives to shadow-mask deposition is printing, which is compatible with manufacturing large screens. But printing is difficult to do reliably, and the resulting devices have lagged behind the performance and durability of those made using conventional techniques. It's been a major materials challenge to develop inks that won't bleed during the printing process, that have the desired electrical and optical properties, and that don't deteriorate over time.

Related Reading: OLED: New Generation of TV's

OLED displays are made up of 12 to 15 layers of materials. In each pixel the red, green, and blue light-emitting materials are positioned side by side and sandwiched between materials that bring electrical current in and out of the device and that allow light to leave it. When these layers blend together during printing, device performance suffers.

DuPont has addressed this problem by using active molecules in the inks used to print each layer that are insoluble in the inks used to print the adjacent layers. This is more complex than it may sound. Each of these materials is a major development effort on its own, and having to tie it to the ones above and below it imposes a lot of constraints.

The company worked with Dainippon Screen to develop a multi-nozzle printer for the new inks. The Dainippon printer works like a garden hose. It generates a continuous stream of ink, rather than droplets, and moves over a surface at rates of four to five meters per second while patterning a display. DuPont says that its red, green, and blue OLED materials can make displays that will last 15 years.

OLED: A New Generation of Television's

OLED: A New Generation of Television's

Organic light-emitting diode (OLED) displays are energy-efficient and crisp, but high manufacturing costs have kept them from being as widely available as liquid-crystal displays (LCDs), especially in larger devices such as TVs. A new type of OLED electronics could help bring down manufacturing costs and make the technology much more widely available.

Transistor test: The glass slides in this image are patterned with transistors designed for driving OLED displays. The metal probes and wires are being used to test their performance. 

The pixels in OLED displays use transistors to stimulate organic molecules, which then emit different colors of light. OLED displays do not need the light-wasting filters that make LCDs such energy hogs. But LCDs dominate the market in large part because the amorphous-silicon transistor arrays used to drive LCDs can be made over areas as big as a single-car garage door and then sliced into smaller pieces to make displays for TVs and other devices. Manufacturing at this scale helps keep costs down.

OLED display makers can't use the same electronics because switching the pixels in an OLED requires relatively high currents that rapidly burn out amorphous-silicon transistors. Instead, today's OLED displays are built on more expensive polycrystalline-silicon transistor arrays. The largest OLED display on the market is a 15-inch model made by LG. It sells for just over $2,300; the same size LCD TV costs under $200.

Less-expensive OLED electronics could, in theory, be made by using organic materials for the electronics as well as the pixels. Transistors made using organic semiconductors provide the high currents needed to drive OLED pixels. But electrons move through conventional organic transistors slowly, which results in a display that doesn't refresh the picture fast enough. To speed these transistors up, engineers have altered the design, shrinking components to bring the source and drain electrodes closer together and make the channel smaller. This makes the device faster because the electrons don't have to travel as far through the organic material that makes up the channel, which can't conduct the electrons very fast. But making such high-resolution devices requires expensive lithography techniques.

A less expensive method, is to bring the source and drain electrodes of a transistor closer together by stacking components on top of one another instead of side by side. These transistors are made by depositing a film of aluminum on a glass substrate to act as the gate electrode, then oxidizing it to create a thin insulating layer on top. Next the researchers deposited an ultrathin, dilute layer of carbon nanotubes to act as a source electrode, followed by a layer of organic materials to act as the channel, and finally a top layer of gold as a drain electrode. Each of these films is very thin, enabling good performance without the need for high-resolution lithography techniques.

New electronic devices, also operate at a tenth of the voltage of conventional OLED electronics, which saves power. The researchers have not yet made large-area OLED displays driven by the vertical transistor arrays, but the transistors operate at suitable current and voltage to do so. While the researchers have so far been making these arrays on glass, the techniques used to make them are compatible with flexible substrates and could be used to make flexible OLED displays.

The vertical electronic structure was first proposed in 1994 by Yang Yang, professor of materials science and engineering at the University of California, Los Angeles, and Alan Heeger, professor of materials science and engineering at the University of California, Santa Barbara. Heeger shared the 2000 Nobel Prize in Chemistry for the discovery and development of conductive polymers like the ones used in the new device. In the mid-1990s, Yang and Heeger began developing these devices through a company called UNIAX that was subsequently acquired by DuPont. When the two did their original work, the performance of the available materials wasn't as good as it is today.

"Carbon nanotubes weren't available in 1994," says Yang. The thin nanotube layer allows very little current leakage, a problem that drained power from previous designs. The devices also switch much faster than was possible in the past. 

The OLED display architecture is being simplified in hopes of further reducing manufacturing costs and complexity. Instead of building a light-emitting pixel next to transistors, build low-power organic transistors that emit light themselves. It's possible to make light-emitting organic transistors if the active materials are electroluminescent, but these transistors only operate at high voltages, making them impractical. The vertical, nanotube-electrode-based architecture could greatly improve the efficiency of these devices.