Support vector machine - Support vector machine A support vector machine (SVM) is a supervised learning technique first discussed by Vladimir Vapnik. An SVM is a maximum-margin hyperplane that lies in some space. Given training examples labeled either "yes" or "no", a maximum-margin hyperplane splits the "yes" and "no" training examples, such that the distance from the closest examples (the margin) to the hyperplane is maximized. The use of the maximum-margin hyperplane is motivated by statistical learning theory, which provides a probabilistic test error bound which is minimized when the margin is maximized. If there exists no hyperplane that can split the "yes" and "no" examples, an SVM will choose a hyperplane that splits the examples as cleanly as possible, while still maximizing the distance to the nearest cleanly split.
Kernel trick - algorithm can easily be transformed into a non-linear algorithm. This non-linear algorithm is the linear algorithm operating in the range space of φ. However, because kernels are used, the φ function is never explicitly computed. This is desirable, because the high-dimensional space may be infinite-dimensional (as is the case when the kernel is a Gaussian). The kernel trick has been applied to several algorithms in machine learning and statistics, including: Support vector machine Principal components analysis Fisher's Linear discriminant analysis The coiner of the term kernel trick is unknown. See also: Kernel Hilbert space.
James Mercer - the basis of the kernel trick (applied by Aizerman), which allows linear algorithms to be easily converted into non-linear algorithms. For example, see support vector machine. Mercer died in London, England..
Vladimir Vapnik - in statistics from the Institute of Control Science in Moscow, in 1964. At AT&T Bell Labs (later Shannon Labs) from 1991 through 2001?, Vapnik and his colleagues developed the theory of the support vector machine. They demonstrated its performance on a number of problems of interest to the machine learning community, including handwriting recognition. He is currently at NEC Laboratories in Princeton, New Jersey, and also Princeton University..
Fortran - refer to the language as "Fortran". Fortran is mainly used for scientific computing and numerical analysis. Although originally a procedural language, recent versions of Fortran have included some features to support object-oriented programming. The first FORTRAN compiler was developed for the IBM 704 in 1954-57 by an IBM team led by John W. Backus. This was an optimizing compiler, because the authors reasoned that no one would use the language if its performance was not comparable to assembly language. The language was widely adopted by scientists for writing numerically intensive programs, which encouraged compiler writers to produce compilers that generate faster code. The inclusion of a complex number data type in the language made Fortran especially suited to scientific computation. There are many vendors of high performance Fortran compilers today. Many.
Alliant Computer Systems - machines were announced in 1985, starting with the FX series. The FX series consisted of a number Computational Elements, or CEs, which included one, two or four Weitek 1064 FPU's and several custom designed support chips. These were controlled by the Interactive Processors, IPs, which were based on Motorola 68008's with 4MB of local RAM, connecting everything together using a crossbar system. Like many early multiprocessing systems, the FX series ran an adapted version of BSD Unix on the IPs, known as Concentrix. Systems were named for the number of CEs inside, the FX/1, FX/2 and FX/4. Alliant machines were fairly small, the FX/1 was about the size of a large full-height FX, while the FX/4 was smaller than a VAX 11/750, about the size of a large photocopier. A second.
Applesoft BASIC programming language - 6502-based computers: it used line numbers, spaces were not necessary in lines, plus it had some killer features that Integer BASIC lacked: Atomic strings. A string is no longer an array of characters (like in C); it is instead a garbage-collected object (like in Scheme and Java). This allows for string arrays; DIM A$(10) got you a vector of ten string variables. Multidimensional arrays. Single-precision floating point variables with an 8-bit exponent and a 31-bit significand. Along with this came a trigonometry library. High-resolution graphics. CHR$, ASC, STR$, and VAL functions for converting between string and numeric types No more writing LET Why weren't many action games written in Applesoft BASIC? Integer variables had to be converted to reals before math could be performed on them; they were then converted back.
Ardent - workstations, notably DEC machines, but eventually shut down completely in February 1995. Ardent started as Dana Computer in November 1985 in Silicon Valley. Their aim was to produce a desktiop supercomputer dedicated to graphics, parallel computing machines that could support up to four processor units. Each processor unit consisted of a MIPS R3000 CPU connected to a custom vector processor. The vector unit held a whopping 8,192 64-bit registers that could be used in any way from 8192 1-word to 32 256-word registers. This compares to modern SIMD systems which allow for perhaps eight to sixteen 128-bit registers with a small variety of addressing schemes. After learning that the name Dana was already in use by a local disk drive company, they became Ardent. Their business plan called for their Titan.
Assembly language - language Assembly is a human-readable notation for the machine language that a specific computer architecture uses. Machine language, a mere pattern of bits, is made readable by replacing the raw values with symbols called mnemonics. So, while a computer will recognize what the IA-32 machine instruction 10110000 01100001 does, for programmers it is easier to remember the equivalent assembly language representation mov %al,$0x61 (it means to move the hexadecimal value 61 (97 decimal) into the register 'al'.) Unlike in high-level languages, there is (to a close approximation) a 1-to-1 correspondence between simple assembly and machine language. Transforming assembly into machine languages is accomplished by an assembler, the other direction by a disassembler. Every computer architecture has its own machine language, and therefore its own assembly language (the example above is from.
Boosting - Boosting Boosting is a machine learning technique for performing supervised learning. Boosting occurs in stages, by incrementally adding to the current learned function. At every stage, a weak learner (i.e., one that can have an accuracy as bad as slightly greater than chance) is trained with the data. The output of the weak learner is then added to the learned function, with some strength (proportional to how accurate the weak learner is). Then, the data is reweighted: examples that the current learned function get wrong are "boosted" in importance, so that future weak learners will attempt to fix the errors. There are several different boosting algorithms, depending on the exact mathematical form of the strength and weight. One of the most common boosting algorithms is AdaBoost. Most boosting.
Compiler optimization - SSA-based Optimizations 5.2.4 Back End Optimizations 5.2.5 Functional Language Optimizations 5.2.6 Other Optimizations 6 Links Problems with Optimization Further problems with optimizing compilers are: Usually, an optimizing compiler simply takes the intermediate representation of a program code and replaces it with a better version. In other words, high-level redundancy in the source program (such as an inefficient algorithm) remains unchanged. Modern third-party compilers usually have to support several objectives. In so doing, these compilers are the jack of all trades yet the master of none. A compiler typically only deals with a small part of an application at a time, at most a module at a time and usually a procedure, the result being that it is unable to consider important contextual information. This is where so-called "post pass" optimizers come.
Convex Computer - time. Convex was formed in 1982 by Bob Paluck and Steve Wallach in Richardson, Texas. Their product concept was not particularly original: they planned on producing a machine very similar in architecture to the Cray Research vector processor machines, but with a somewhat lower performance, and at a much lower price point. In order to lower costs, the Convex designs were not as technologically aggressive as Cray's, and were based on more mainstream chip technology, attempting to make up for the loss in performance in other ways. Their first machine was the C1, released in 1985. The C1 was very similar to the Cray-1 in general design, but used a slower memory and main CPU. They offset this by increasing the capabilities of the vector units, including 128 64-bit registers, double.
Cray X-MP - designed, built and sold by Cray Research. The company's first parallel vector processor machine, it was the 1982 successor to the Cray-1 and a fourth generation machine. The principal designer was Steve Chen. The X-MP shared the 'horseshoe' design of the earlier machine. The processors ran on a 8.5 ns clock compared to 12.5 ns for the Cray-1A, delivering around 55 MFLOPS per processor and 235 MFLOPS for the four processor 1982 machine. The processors also had better chaining support, parallel arithmetic pipes, and shared memory access with multiple pipelines per processor. The system initially ran COS with UniCOS (a System V derivation) running through the guest operating system facility. Unicos became the main OS from 1984. The X-MP was sold with one, two or four processors and from one to.
Timeline of computing 1950-1979 - 500 BC-1949, 1950-1979, 1980-1989, 1990-present 1950 First commercial computer: Konrad Zuse leases his Z4 machine to ETH Zuerich. 1950 Floppy disk invented at the Imperial University in Tokyo by Doctor Yoshiro Nakamats, the sales license for the disk was granted to IBM. 1950 The British mathematician and computer pioneer Alan Turing published a paper describing what would come to be called the Turing Test. The paper explored the nature and potential development of human and computer intelligence and communication. 1951 High level language compiler invented by Grace Murray Hopper. 1951 Whirlwind, the first real-time computer built at MIT by the team of Jay Forrester for the US Air Defence System, became operational. This computer is the first to allow interactive computing, allowing users to interact with it using a keyboard and.
Statistical learning theory - convergence of the learning process? Theory of controlling the generalization ability of learning processes How can one control the rate of convergence (the generalization ability) of the learning process? Theory of constructing learning machines How can one construct algorithms that can control the generalization ability? The last part of the theory introduced an well-known learning algorithm: the support vector machine. Statistical learning theory contains important concepts such as the VC dimension and structural risk minimization. This theory is foundation of a real understanding of machine learning. This theory is related to mathematical subjects such as: reproducing kernel Hilbert spaces regularization networks kernelss References The Nature of Statistical Learning Theory, Vladimir Vapnik, Springer-Verlag, (1999), ISBN 0387987800 Statistical Learning Theory, Vladimir Vapnik, Wiley-Interscience, (1998), ISBN 0471030031.
Supervised learning - Supervised learning Supervised learning is a machine learning technique for creating a function from training data. The training data consists of pairs of input objects (typically vectors), and desired outputs. The output of the function can be a continuous value (called regression), or can predict a class label of the input object (called classification). The task of the supervised learner is to predict the value of the function for any valid input object after having seen only a small number of training examples (i.e. pairs of input and target output). To achieve this, the learner has to generalize from the presented data to unseen situations in a "reasonable" way (see inductive bias). (Compare with unsupervised learning.) In order to solve a given problem of supervised learning (e.g. learning to recognize.
RISC - minicomputer whose initial implementation required 3 racks of equipment for a single cpu, and was notable for the amazing variety of memory access styles it supported, and the fact that every one of them was available for every instruction. RISC design philosophy In the late 1970s research at IBM (and similar projects elsewhere) demonstrated that the majority of these "orthogonal" addressing modes were ignored by most programs. This was a side effect of the increasing use of compilers to generate the programs, as opposed to writing them in assembly language. The compilers tended to be fairly dumb in terms of the features they used, largely a side effect of attempting to be fairly small. The market was clearly moving to even wider use of compilers, diluting the usefulness of these orthogonal.
Plant improvement - Warning: The following article has been machine-translated from Italian, and the English version has been only partially checked. The genetic manipulation of plants has been practised for hundreds of years. Agronoms and horticulturalists have developed schemes of hybridisation among plants to introduce and to maintain some desired characteristics. The classical methods are slow and uncertain, they require sexual reproduction followed by repeated recrossings between progeny and progenitors and they also sometimes transfer unwanted characteristics. Scientific developments have allowed new techniques, mostly used today for producing many plants from one with particular characteristics. Obtaining just these characteristics, together with speeding the propagation of the plants, is the purpose of modern biotechnology. Among the most obvious characteristics looked for: Increased quality and yield of the crop Increased tolerance of environmental pressures (salinity, extrem.
Naive Bayesian classification - re-factored as: Thus, the probability ratio p(S D) / p(¬S D) can be expressed in terms of a series of likelihood ratios. The actual probability p(S D) can be easily computed from ln(p(S D) / p(¬S D)) based on the observation that p(S D) + p(¬S D) = 1. Taking the logarithm of all these ratios, we have: This technique of "log-likelihood ratios" is a common technique in statistics. In the case of two mutually exclusive alternatives (such as this example), the conversion of a log-likelihood ratio to a probability takes the form of a sigmoid curve: see logit for details. In real life, the naive Bayes approach is more powerful than might be expected from the extreme simplicity of its model; in particular, it is fairly robust in the presence.
Neural network - network of nodes - hence the term "neural network". Table of contents showTocToggle("show","hide") 1 Structure 1.1 Models 1.2 Calculations 1.3 Usefulness 2 Real life applications 3 Types of neural networks 3.4 Single layer perceptron 3.5 Multi-layer perceptron 3.6 Simple recurrent network 3.7 Hopfield network 3.8 Boltzmann machine 3.9 Support vector machine 3.10 Committee of machines 3.11 Self-organizing map 3.12 Instantaneously trained networks 3.13 Data representation 4 Relation to optimization techniques 5.
