Architecture (Несколько текстов для зачёта), страница 5

The role of connection strengths between neurons in the brain is crucial; scientists believe they determine, to a great extent, the way in which the brain processes the information it takes in through the senses. Neuroscientists studying the structure and function of the brain believe that various patterns of neurons firing can be associated with specific memories. In this theory, the strength of the connections between the relevant neurons determines the strength of the memory. Important information that needs to be remembered may cause the brain to constantly reinforce the pathways between the neurons that form the memory, while relatively unimportant information will not receive the same degree of reinforcement.

A

Connection Weights

To mimic the way in which biological neurons reinforce certain axon-dendrite pathways, the connections between artificial neurons in a neural network are given adjustable connection weights, or measures of importance. When signals are received and processed by a node, they are multiplied by a weight, added up, and then transformed by a nonlinear function. The effect of the nonlinear function is to cause the sum of the input signals to approach some value, usually +1 or 0. If the signals entering the node add up to a positive number, the node sends an output signal that approaches +1 out along all of its connections, while if the signals add up to a negative value, the node sends a signal that approaches 0. This is similar to a simplified model of a how a biological neuron functions—the larger the input signal, the larger the output signal.

B

Training Sets

Computer scientists teach neural networks by presenting them with desired input-output training sets. The input-output training sets are related patterns of data. For instance, a sample training set might consist of ten different photographs for each of ten different faces. The photographs would then be digitally entered into the input layer of the network. The desired output would be for the network to signal one of the neurons in the output layer of the network per face. Beginning with equal, or random, connection weights between the neurons, the photographs are digitally entered into the input layer of the neural network and an output signal is computed and compared to the target output. Small adjustments are then made to the connection weights to reduce the difference between the actual output and the target output. The input-output set is again presented to the network and further adjustments are made to the connection weights because the first few times that the input is entered, the network will usually choose the incorrect output neuron. After repeating the weight-adjustment process many times for all input-output patterns in the training set, the network learns to respond in the desired manner.

A neural network is said to have learned when it can correctly perform the tasks for which it has been trained. Neural networks are able to extract the important features and patterns of a class of training examples and generalize from these to correctly process new input data that they have not encountered before. For a neural network trained to recognize a series of photographs, generalization would be demonstrated if a new photograph presented to the network resulted in the correct output neuron being signaled.

A number of different neural network learning rules, or algorithms, exist and use various techniques to process information. Common arrangements use some sort of system to adjust the connection weights between the neurons automatically. The most widely used scheme for adjusting the connection weights is called error back-propagation, developed independently by American computer scientists Paul Werbos (in 1974), David Parker (in 1984/1985), and David Rumelhart, Ronald Williams, and others (in 1985). The back-propagation learning scheme compares a neural network’s calculated output to a target output and calculates an error adjustment for each of the nodes in the network. The neural network adjusts the connection weights according to the error values assigned to each node, beginning with the connections between the last hidden layer and the output layer. After the network has made adjustments to this set of connections, it calculates error values for the next previous layer and makes adjustments. The back-propagation algorithm continues in this way, adjusting all of the connection weights between the hidden layers until it reaches the input layer. At this point it is ready to calculate another output.

V

IMPLEMENTATIONS AND FUTURE TECHNOLOGY

Neural networks have been applied to many tasks that are easy for humans to accomplish, but difficult for traditional computers. Because neural networks mimic the brain, they have shown much promise in so-called sensory processing tasks such as speech recognition, pattern recognition, and the transcription of hand-written text. In some settings, neural networks can perform as well as humans. Neural-network-based backgammon software, for example, rivals the best human players.

While traditional computers still outperform neural networks in most situations, neural networks are superior in recognizing patterns in extremely large data sets. Furthermore, because neural networks have the ability to learn from a set of examples and generalize this knowledge to new situations, they are excellent for work requiring adaptive control systems. For this reason, the United States National Aeronautics and Space Administration (NASA) has extensively studied neural networks to determine whether they might serve to control future robots sent to explore planetary bodies in our solar system. In this application, robots could be sent to other planets, such as Mars, to carry out significant and detailed exploration autonomously.

An important advantage that neural networks have over traditional computer systems is that they can sustain damage and still function properly. This design characteristic of neural networks makes them very attractive candidates for future aircraft control systems, especially in high performance military jets. Another potential use of neural networks for civilian and military use is in pattern recognition software for radar, sonar, and other remote-sensing devices.

Motherboard

Motherboard, in computer science, the main circuit board in a computer. The most important computer chips and other electronic components that give function to a computer are located on the motherboard. The motherboard is a printed circuit board that connects the various elements on it through the use of traces, or electrical pathways. The motherboard is indispensable to the computer and provides the main computing capability.

Personal computers normally have one central processing unit (CPU), or microprocessor, which is located with other chips on the motherboard. The manufacturer and model of the CPU chip carried by the motherboard is a key criterion for designating the speed and other capabilities of the computer. The CPU in many personal computers is not permanently attached to the motherboard, but is instead plugged into a socket so that it may be removed and upgraded.

Motherboards also contain important computing components, such as the basic input/output system (BIOS), which contains the basic set of instructions required to control the computer when it is first turned on; different types of memory chips such as random access memory (RAM) and cache memory; mouse, keyboard, and monitor control circuitry; and logic chips that control various parts of the computer’s function. Having as many of the key components of the computer as possible on the motherboard improves the speed and operation of the computer.

Users may expand their computer’s capability by inserting an expansion board into special expansion slots on the motherboard. Expansion slots are standard with nearly all personal computers and offer faster speed, better graphics capabilities, communication capability with other computers, and audio and video capabilities. Expansion slots come in either half or full size, and can transfer 8 or 16 bits (the smallest units of information that a computer can process) at a time, respectively.

The pathways that carry data on the motherboard are called buses. The amount of data that can be transmitted at one time between a device, such as a printer or monitor, and the CPU affects the speed at which programs run. For this reason, buses are designed to carry as much data as possible. To work properly, expansion boards must conform to bus standards such as integrated drive electronics (IDE), Extended Industry Standard Architecture (EISA), or small computer system interface (SCSI).

Central Processing Unit

I

INTRODUCTION

Central Processing Unit (CPU), in computer science, microscopic circuitry that serves as the main information processor in a computer. A CPU is generally a single microprocessor made from a wafer of semiconducting material, usually silicon, with millions of electrical components on its surface. On a higher level, the CPU is actually a number of interconnected processing units that are each responsible for one aspect of the CPU’s function. Standard CPUs contain processing units that interpret and implement software instructions, perform calculations and comparisons, make logical decisions (determining if a statement is true or false based on the rules of Boolean algebra), temporarily store information for use by another of the CPU’s processing units, keep track of the current step in the execution of the program, and allow the CPU to communicate with the rest of the computer.

II

HOW A CPU WORKS

A

CPU Function

A CPU is similar to a calculator, only much more powerful. The main function of the CPU is to perform arithmetic and logical operations on data taken from memory or on information entered through some device, such as a keyboard, scanner, or joystick. The CPU is controlled by a list of software instructions, called a computer program. Software instructions entering the CPU originate in some form of memory storage device such as a hard disk, floppy disk, CD-ROM, or magnetic tape. These instructions then pass into the computer’s main random access memory (RAM), where each instruction is given a unique address, or memory location. The CPU can access specific pieces of data in RAM by specifying the address of the data that it wants.

As a program is executed, data flow from RAM through an interface unit of wires called the bus, which connects the CPU to RAM. The data are then decoded by a processing unit called the instruction decoder that interprets and implements software instructions. From the instruction decoder the data pass to the arithmetic/logic unit (ALU), which performs calculations and comparisons. Data may be stored by the ALU in temporary memory locations called registers where it may be retrieved quickly. The ALU performs specific operations such as addition, multiplication, and conditional tests on the data in its registers, sending the resulting data back to RAM or storing it in another register for further use. During this process, a unit called the program counter keeps track of each successive instruction to make sure that the program instructions are followed by the CPU in the correct order.

B

Branching Instructions

The program counter in the CPU usually advances sequentially through the instructions. However, special instructions called branch or jump instructions allow the CPU to abruptly shift to an instruction location out of sequence. These branches are either unconditional or conditional. An unconditional branch always jumps to a new, out of order instruction stream. A conditional branch tests the result of a previous operation to see if the branch should be taken. For example, a branch might be taken only if the result of a previous subtraction produced a negative result. Data that are tested for conditional branching are stored in special locations in the CPU called flags.

C

Clock Pulses

The CPU is driven by one or more repetitive clock circuits that send a constant stream of pulses throughout the CPU’s circuitry. The CPU uses these clock pulses to synchronize its operations. The smallest increments of CPU work are completed between sequential clock pulses. More complex tasks take several clock periods to complete. Clock pulses are measured in Hertz, or number of pulses per second. For instance, a 100-megahertz (100-MHz) processor has 100 million clock pulses passing through it per second. Clock pulses are a measure of the speed of a processor.

D

Fixed-Point and Floating-Point Numbers

Most CPUs handle two different kinds of numbers: fixed-point and floating-point numbers. Fixed-point numbers have a specific number of digits on either side of the decimal point. This restriction limits the range of values that are possible for these numbers, but it also allows for the fastest arithmetic. Floating-point numbers are numbers that are expressed in scientific notation, in which a number is represented as a decimal number multiplied by a power of ten. Scientific notation is a compact way of expressing very large or very small numbers and allows a wide range of digits before and after the decimal point. This is important for representing graphics and for scientific work, but floating-point arithmetic is more complex and can take longer to complete. Performing an operation on a floating-point number may require many CPU clock periods. A CPU’s floating-point computation rate is therefore less than its clock rate. Some computers use a special floating-point processor, called a coprocessor, that works in parallel to the CPU to speed up calculations using floating-point numbers. This coprocessor has become standard on many personal computer CPUs, such as Intel’s Pentium chip.

III

HISTORY

A

Early Computers

In the first computers, CPUs were made of vacuum tubes and electric relays rather than microscopic transistors on computer chips. These early computers were immense and needed a great deal of power compared to today’s microprocessor-driven computers. The first general purpose electronic computer, the ENIAC (Electronic Numerical Integrator And Computer), was completed in 1946 and filled a large room. About 18,000 vacuum tubes were used to build ENIAC’s CPU and input/output circuits. Between 1946 and 1956 all computers had bulky CPUs that consumed massive amounts of energy and needed continual maintenance, because the vacuum tubes burned out frequently and had to be replaced.

B

The Transistor

A solution to the problems posed by vacuum tubes came in 1947, when American physicists John Bardeen, Walter Brattain, and William Shockley first demonstrated a revolutionary new electronic switching and amplifying device called the transistor. The transistor had the potential to work faster and more reliably and to consume much less power than a vacuum tube. Despite the overwhelming advantages transistors offered over vacuum tubes, it took nine years before they were used in a commercial computer. The first commercially available computer to use transistors in its circuitry was the UNIVAC (UNIVersal Automatic Computer), delivered to the United States Air Force in 1956.

C

The Integrated Circuit

Development of the computer chip started in 1958 when Jack Kilby of Texas Instruments demonstrated that it was possible to integrate the various components of a CPU onto a single piece of silicon. These computer chips were called integrated circuits (ICs) because they combined multiple electronic circuits on the same chip. Subsequent design and manufacturing advances allowed transistor densities on integrated circuits to increase tremendously. The first ICs had only tens of transistors per chip compared to the 3 million to 5 million transistors per chip common on today’s CPUs.