Artificial Intelligence Glossarium: 1000 terms

Artificial Neural Network (ANN) (Искусственная нейронная сеть) – is a computational model in machine learning, which is inspired by the biological structures and functions of the mammalian brain. Such a model consists of multiple units called artificial neurons which build connections between each other to pass information. The advantage of such a model is that it progressively “learns” the tasks from the given data without specific programing for a single task.

Artificial neuron is a mathematical function conceived as a model of biological neurons, a neural network. The difference between an artificial neuron and a biological neuron is shown in the figure.

Artificial neurons are the elementary units of an artificial neural network. An artificial neuron receives one or more inputs (representing excitatory postsynaptic potentials and inhibitory postsynaptic potentials on nerve dendrites) and sums them to produce an output signal (or activation, representing the action potential of the neuron that is transmitted down its axon). Typically, each input is weighted separately, and the sum is passed through a non-linear function known as an activation function or transfer function. Transfer functions are usually sigmoid, but they can also take the form of other non-linear functions, piecewise linear functions, or step functions. They are also often monotonically increasing, continuous, differentiable, and bounded [56,57].

Artificial Superintelligence (ASI) (Искусственный сверхинтеллект) – is a term referring to the time when the capability of computers will surpass humans. “Artificial intelligence,” which has been much used since the 1970s, refers to the ability of computers to mimic human thought. Artificial superintelligence goes a step beyond and posits a world in which a computer’s cognitive ability is superior to a human.

Assistive intelligence (Вспомогательный интеллект) is AI-based systems that help make decisions or perform actions.

Association (Ассоциация) is another type of unsupervised learning method that uses different rules to find relationships between variables in a given dataset. These methods are frequently used for market basket analysis and recommendation engines, along the lines of “Customers Who Bought This Item Also Bought” recommendations.

Association for the Advancement of Artificial Intelligence (AAAI) (Ассоциация по развитию искусственного интеллекта) — An international, nonprofit, scientific society devoted to promote research in, and responsible use of, artificial intelligence. AAAI also aims to increase public understanding of artificial intelligence (AI), improve the teaching and training of AI practitioners, and provide guidance for research planners and funders concerning the importance and potential of current AI developments and future directions

Association Rule Learning (Правила обучения ассоциации) – A rule-based Machine Learning method for discovering interesting relations between variables in large data sets.

Asymptotic computational complexity (Асимптотическая вычислительная сложность) – In computational complexity theory, asymptotic computational complexity is the usage of asymptotic analysis for the estimation of computational complexity of algorithms and computational problems, commonly associated with the usage of the big O notation [58].

Asynchronous inter-chip protocols (Асинхронные межкристальные протоколы) are protocols for data exchange in low-speed devices; instead of frames, individual characters are used to control the exchange of data.

Attention mechanism (Механизм внимания) is one of the key innovations in the field of neural machine translation. Attention allowed neural machine translation models to outperform classical machine translation systems based on phrase translation. The main bottleneck in sequence-to-sequence learning is that the entire content of the original sequence needs to be compressed into a vector of a fixed size. The attention mechanism facilitates this task by allowing the decoder to look back at the hidden states of the original sequence, which are then provided as a weighted average as additional input to the decoder.

Attributional calculus (AC) (Атрибутивное исчисление) – A logic and representation system defined by Ryszard S. Michalski. It combines elements of predicate logic, propositional calculus, and multi-valued logic. Attributional calculus provides a formal language for natural induction, an inductive learning process whose results are in forms natural to people [59].

Augmented Intelligence (Дополненный (расширенный) интеллект) – is the intersection of machine learning and advanced applications, where clinical knowledge and medical data converge on a single platform. The potential benefits of Augmented Intelligence are realized when it is used in the context of workflows and systems that healthcare practitioners operate and interact with. Unlike Artificial Intelligence, which tries to replicate human intelligence, Augmented Intelligence works with and amplifies human intelligence [60]

Augmented reality (AR) (Дополненная реальность) — An interactive experience of a real-world environment where the objects that reside in the real-world are “augmented” by computer-generated perceptual information, sometimes across multiple sensory modalities, including visual, auditory, haptic, somatosensory, and olfactory.

Augmented reality technologies (Технологии дополненной реальности) are visualization technologies based on adding information or visual effects to the physical world by overlaying graphic and/or sound content to improve user experience and interactive features.

Auto Associative Memory (Автоассоциативная память) is a single layer neural network in which the input training vector and the output target vectors are the same. The weights are determined so that the network stores a set of patterns. As shown in the following figure, the architecture of Auto Associative memory network has “n’ number of input training vectors and similar “n’ number of output target vectors [61].

Autoencoder (Автокодер) – а type of Artificial Neural Network used to produce efficient representations of data in an unsupervised and non-linear manner, typically to reduce dimensionality [62].

Automata theory (Теория автоматов) – The study of abstract machines and automata, as well as the computational problems that can be solved using them. It is a theory in theoretical computer science and discrete mathematics (a subject of study in both mathematics and computer science). [63] Automata theory (part of the theory of computation) is a theoretical branch of Computer Science and Mathematics, which mainly deals with the logic of computation with respect to simple machines, referred to as automata [64].

Automated control system (Автоматизированная система управления) – a set of software and hardware designed to control technological and (or) production equipment (executive devices) and the processes they produce, as well as to control such equipment and processes.

Automated planning and scheduling (Also simply AI planning.) (Планирование ИИ) – A branch of artificial intelligence that concerns the realization of strategies or action sequences, typically for execution by intelligent agents, autonomous robots and unmanned vehicles. Unlike classical control and classification problems, the solutions are complex and must be discovered and optimized in multidimensional space. Planning is also related to decision theory [65].

Automated processing of personal data (Автоматизированная обработка персональных данных) – processing of personal data using computer technology.

Automated reasoning (Автоматизированное мышление) – An area of computer science and mathematical logic dedicated to understanding different aspects of reasoning. The study of automated reasoning helps produce computer programs that allow computers to reason completely, or nearly completely, automatically. Although automated reasoning is considered a sub-field of artificial intelligence, it also has connections with theoretical computer science, and even philosophy [66].

Automated system (Автоматизированная система) is an organizational and technical system that guarantees the development of solutions based on the automation of information processes in various fields of activity.

Automation (Автоматизация) is a technology by which a process or procedure is performed with minimal human intervention.

Automation bias (Предвзятость автоматизации) – When a human decision maker favors recommendations made by an automated decision-making system over information made without automation, even when the automated decision-making system makes errors [67].

Autonomic computing (Автономные вычисления) is the ability of a system to adaptively self-manage its own resources for high-level computing functions without user input.

Autonomous (Автономность) – A machine is described as autonomous if it can perform its task or tasks without needing human intervention.

Autonomous artificial intelligence (Автономный искусственный интеллект) is a biologically inspired system that tries to reproduce the structure of the brain, the principles of its operation with all the properties that follow from this.

Autonomous artificial intelligence systems (Системы автономного искусственного интеллекта) – simulate the work and structure of the brain (thinking, creativity, emotions, will, freedom of choice and decision-making, search for new knowledge and making optimal decisions, memory, etc.). Such systems are also called adaptive artificial intelligence or neuromorphic artificial intelligence.

Autonomous car (Also self-driving car, robot car, and driverless car.) (Автономный автомобиль) – A vehicle that is capable of sensing its environment and moving with little or no human input [68].

Autonomous robot (Автономный робот) — A robot that performs behaviors or tasks with a high degree of autonomy. Autonomous robotics is usually considered to be a subfield of artificial intelligence, robotics, and information engineering [69].

Autonomous vehicle (Автономное транспортное средство) is a mode of transport based on an autonomous driving system. The control of an autonomous vehicle is fully automated and carried out without a driver using optical sensors, radar and computer algorithms.

Autoregressive Model (Авторегрессионная модель) – An autoregressive model is a time series model that uses observations from previous time steps as input to a regression equation to predict the value at the next time step. In statistics and signal processing, an autoregressive model is a representation of a type of random process. It is used to describe certain time-varying processes in nature, economics, etc. [70].

Auxiliary intelligence (Дополнительный интеллект) – systems based on artificial intelligence that complement human decisions and are able to learn in the process of interacting with people and the environment.

Average precision (Средняя точность) – A metric for summarizing the performance of a ranked sequence of results. Average precision is calculated by taking the average of the precision values for each relevant result (each result in the ranked list where the recall increases relative to the previous result) [71].

Ayasdi (Платформа Ayasdi) is an enterprise scale machine intelligence platform that delivers the automation that is needed to gain competitive advantage from the company’s big and complex data. Ayasdi supports large numbers of business analysts, data scientists, endusers, developers and operational systems across the organization, simultaneously creating, validating, using and deploying sophisticated analyses and mathematical models at scale.

“B”

Backpropagation (Обратное распространение ошибки) – Backpropagation, also called “backward propagation of errors,” is an approach that is commonly used in the training process of the deep neural network to reduce errors.

Backpropagation through time (BPTT) (Обратное распространение во времени) – A gradient-based technique for training certain types of recurrent neural networks. It can be used to train Elman networks. The algorithm was independently derived by numerous researchers.

Backward Chaining (Обратная цепочка (или обратное рассуждение)) – Backward chaining, also called goal-driven inference technique, is an inference approach that reasons backward from the goal to the conditions used to get the goal. Backward chaining inference is applied in many different fields, including game theory, automated theorem proving, and artificial intelligence [72].

Bag-of-words model (Модель мешка слов) — A simplifying representation used in natural language processing and information retrieval (IR). In this model, a text (such as a sentence or a document) is represented as the bag (multiset) of its words, disregarding grammar and even word order but keeping multiplicity. The bag-of-words model has also been used for computer vision. The bag-of-words model is commonly used in methods of document classification where the (frequency of) occurrence of each word is used as a feature for training a classifier [73].

Bag-of-words model in computer vision (Модель мешка слов в компьютерном зрении) — In computer vision, the bag-of-words model (BoW model) can be applied to image classification, by treating image features as words. In document classification, a bag of words is a sparse vector of occurrence counts of words; that is, a sparse histogram over the vocabulary. In computer vision, a bag of visual words is a vector of occurrence counts of a vocabulary of local image features.

Baldwin effect (Эффект Балдвина) – the skills acquired by organisms during their life as a result of learning, after a certain number of generations, are recorded in the genome.

Baseline (Базовый уровень) – A model used as a reference point for comparing how well another model (typically, a more complex one) is performing. For example, a logistic regression model might serve as a good baseline for a deep model. For a particular problem, the baseline helps model developers quantify the minimal expected performance that a new model must achieve for the new model to be useful.

Batch (Пакет) – The set of examples used in one gradient update of model training.

Batch Normalization (Пакетная нормализация) – A preprocessing step where the data are centered around zero, and often the standard deviation is set to unity.

Batch size (Размер партии) – The number of examples in a batch. For example, the batch size of SGD is 1, while the batch size of a mini-batch is usually between 10 and 1000. Batch size is usually fixed during training and inference; however, TensorFlow does permit dynamic batch sizes.

Bayes’s Theorem (Теорема Байеса) – A famous theorem used by statisticians to describe the probability of an event based on prior knowledge of conditions that might be related to an occurrence.

Bayesian classifier in machine learning (Байесовский классификатор в машинном обучении) is a family of simple probabilistic classifiers based on the use of the Bayes theorem and the “naive” assumption of the independence of the features of the objects being classified.

Bayesian Filter (Фильтрация по Байесу) is a program using Bayesian logic. It is used to evaluate the header and content of email messages and determine whether or not it constitutes spam – unsolicited email or the electronic equivalent of hard copy bulk mail or junk mail. A Bayesian filter works with probabilities of specific words appearing in the header or content of an email. Certain words indicate a high probability that the email is spam, such as Viagra and refinance [74].

Bayesian Network (Байесовская сеть) – also called belief network, or probabilistic directed acyclic graphical model, is a probabilistic graphical model (a statistical model) that represents a set of variables and their conditional dependencies via a directed acyclic graph [75].

Bayesian optimization (Байесовская оптимизация) – A probabilistic regression model technique for optimizing computationally expensive objective functions by instead optimizing a surrogate that quantifies the uncertainty via a Bayesian learning technique. Since Bayesian optimization is itself very expensive, it is usually used to optimize expensive-to-evaluate tasks that have a small number of parameters, such as selecting hyperparameters.

Bayesian programming (Байесовское программирование) – A formalism and a methodology for having a technique to specify probabilistic models and solve problems when less than the necessary information is available.

Bees algorithm (Алгоритм пчелиной колонии) — A population-based search algorithm which was developed by Pham, Ghanbarzadeh and et al. in 2005. It mimics the food foraging behaviour of honey bee colonies. In its basic version the algorithm performs a kind of neighbourhood search combined with global search, and can be used for both combinatorial optimization and continuous optimization. The only condition for the application of the bees algorithm is that some measure of distance between the solutions is defined. The effectiveness and specific abilities of the bees algorithm have been proven in a number of studies.

Behavior informatics (BI) (Информатика поведения) — The informatics of behaviors so as to obtain behavior intelligence and behavior insights.

Behavior tree (BT) (Дерево поведения) – A mathematical model of plan execution used in computer science, robotics, control systems and video games. They describe switchings between a finite set of tasks in a modular fashion. Their strength comes from their ability to create very complex tasks composed of simple tasks, without worrying how the simple tasks are implemented. BTs present some similarities to hierarchical state machines with the key difference that the main building block of a behavior is a task rather than a state. Its ease of human understanding make BTs less error-prone and very popular in the game developer community. BTs have shown to generalize several other control architectures [76].

Belief-desire-intention software model (BDI) (Модель убеждений, желаний и намерений) — A software model developed for programming intelligent agents. Superficially characterized by the implementation of an agent’s beliefs, desires and intentions, it actually uses these concepts to solve a particular problem in agent programming. In essence, it provides a mechanism for separating the activity of selecting a plan (from a plan library or an external planner application) from the execution of currently active plans. Consequently, BDI agents are able to balance the time spent on deliberating about plans (choosing what to do) and executing those plans (doing it). A third activity, creating the plans in the first place (planning), is not within the scope of the model, and is left to the system designer and programmer. [77]

Bellman equation (Уравнение Беллмана) – named after Richard E. Bellman, is a necessary condition for optimality associated with the mathematical optimization method known as dynamic programming. It writes the “value” of a decision problem at a certain point in time in terms of the payoff from some initial choices and the “value” of the remaining decision problem that results from those initial choices. This breaks a dynamic optimization problem into a sequence of simpler subproblems, as Bellman’s “principle of optimality” prescribes [78].

Benchmark (also benchmark program, benchmarking program, benchmark test) (Бенчмарк) – test program or package for evaluating (measuring and / or comparing) various aspects of the performance of a processor, individual devices, computer, system or a specific application, software; a benchmark that allows products from different manufacturers to be compared against each other or against some standard. For example, online benchmark – online benchmark; standard benchmark – standard benchmark; benchmark time comparison – comparison of benchmark execution times. [79].

Benchmarking (Бенчмаркинг) is a set of techniques that allow you to study the experience of competitors and implement best practices in your company.

BETA (БЕТА версия) A beta publication refers to a phase in online service development in which the service is coming together functionality-wise but genuine user experiences are required before the service can be finished in a user-centered way. In online service development, the aim of the beta phase is to recognize both programming issues and usability-enhancing procedures. The beta phase is particularly often used in connection with online services and it can be either freely available (open beta) or restricted to a specific target group (closed beta). [80]

Bias (Погрешность) is a systematic trend that causes differences between results and facts. Error exists in the numbers of the data analysis process, including the source of the data, the estimate chosen, and how the data is analyzed. Error can seriously affect the results, for example, when studying people’s shopping habits. If the sample size is not large enough, the results may not reflect the buying habits of all people. That is, there may be discrepancies between survey results and actual results.

Biased algorithm (Алгоритмическая предвзятость) – systematic and repetitive errors in a computer system that lead to unfair results, such as one privilege persecuting groups of users over others. Also, sexist and racist algorithms.

Bidirectional (BiDi) (Двунаправленность) – A term used to describe a system that evaluates the text that both precedes and follows a target section of text. In contrast, a unidirectional system only evaluates the text that precedes a target section of text.

Bidirectional Encoder Representations from Transformers (BERT) (Представления двунаправленного кодировщика от трансформаторов) – A model architecture for text representation. A trained BERT model can act as part of a larger model for text classification or other ML tasks. BERT has the following characteristics: Uses the Transformer architecture, and therefore relies on self-attention. Uses the encoder part of the Transformer. The encoder’s job is to produce good text representations, rather than to perform a specific task like classification. Is bidirectional. Uses masking for unsupervised training.

Bidirectional language model (Двунаправленная языковая модель) – A language model that determines the probability that a given token is present at a given location in an excerpt of text based on the preceding and following text.

Big data (Большие данные) is a term for sets of digital data whose large size, rate of increase or complexity requires significant computing power for processing and special software tools for analysis and presentation in the form of human-perceptible results.

Big O notation (Запись Big O notation) – A mathematical notation that describes the limiting behavior of a function when the argument tends towards a particular value or infinity. It is a member of a family of notations invented by Paul Bachmann, Edmund Landau, and others, collectively called Bachmann – Landau notation or asymptotic notation [81].