Convolutional operation – the following two-step mathematical operation: Element-wise multiplication of the convolutional filter and a slice of an input matrix. (The slice of the input matrix has the same rank and size as the convolutional filter); Summation of all the values in the resulting product matrix302.
Corelet programming environment (CPE) is a scalable environment that allows programmers to set the functional behavior of a neural network by adjusting its parameters and communication characteristics303.
Corpus of texts is a large dataset of written or spoken material that can be used to train a machine to perform linguistic tasks304.
Correlation analysis is a statistical data processing method that measures the strength of the relationship between two or more variables. Thus, it determines whether there is a connection between the phenomena and how strong the connection between these phenomena is305.
Correlation is a statistical relationship between two or more random variables306.
Cost – synonym for loss. A measure of how far a model’s predictions are from its label. Or, to put it more pessimistically, a measure of how bad a model is. To determine this value, the model must define a loss function. For example, linear regression models typically use the standard error for the loss function, while logistic regression models use the log loss307,308.
Co-training essentially amplifies independent signals into a stronger signal. For instance, consider a classification model that categorizes individual used cars as either Good or Bad. One set of predictive features might focus on aggregate characteristics such as the year, make, and model of the car; another set of predictive features might focus on the previous owner’s driving record and the car’s maintenance history. The seminal paper on co-training is Combining Labeled and Unlabeled Data with Co-Training by Blum and Mitchell309.
Counterfactual fairness is a fairness metric that checks whether a classifier produces the same result for one individual as it does for another individual who is identical to the first, except with respect to one or more sensitive attributes. Evaluating a classifier for counterfactual fairness is one method for surfacing potential sources of bias in a model. See «When Worlds Collide: Integrating Different Counterfactual Assumptions in Fairness» for a more detailed discussion of counterfactual fairness310.
Coverage bias – this bias means that the study sample is not representative and that the data set in the array has zero chance of being included in the sample311.
Crash blossom is a sentence or phrase with an ambiguous meaning. Crash blossoms present a significant problem in natural language understanding. For example, the headline Red Tape Holds Up Skyscraper is a crash blossom because an NLU model could interpret the headline literally or figuratively312.
Critic – synonym for Deep Q-Network313.
Critical information infrastructure – objects of critical information infrastructure, as well as telecommunication networks used to organize the interaction of such objects314.
Critical information infrastructure of the Russian Federation is a set of critical information infrastructure objects, as well as telecommunication networks used to organize the interaction of critical information infrastructure objects with each other315.
Cross-entropy is a generalization of Log Loss to multi-class classification problems. Cross-entropy quantifies the difference between two probability distributions. See also perplexity316.
Crossover (also recombination) in genetic algorithms and evolutionary computation, a genetic operator used to combine the genetic information of two parents to generate new offspring. It is one way to stochastically generate new solutions from an existing population, and analogous to the crossover that happens during sexual reproduction in biological organisms. Solutions can also be generated by cloning an existing solution, which is analogous to asexual reproduction. Newly generated solutions are typically mutated before being added to the population317.
Cross-Validation (k-fold Cross-Validation, Leave-p-out Cross-Validation) is a collection of processes designed to evaluate how the results of a predictive model will generalize to new data sets. k-fold Cross-Validation; Leave-p-out Cross-Validation318.
Cryogenic freezing (cryonics, human cryopreservation) is a technology of preserving in a state of deep cooling (using liquid nitrogen) the head or body of a person after his death with the intention to revive them in the future319.
Cyber-physical systems are intelligent networked systems with built-in sensors, processors and drives that are designed to interact with the physical environment and support the operation of computer information systems in real time320.
«D»
Darkforest is a computer go program, based on deep learning techniques using a convolutional neural network. Its updated version Darkforest2 combines the techniques of its predecessor with Monte Carlo tree search. The MCTS effectively takes tree search methods commonly seen in computer chess programs and randomizes them. With the update, the system is known as Darkforest3321.
Dartmouth workshop – the Dartmouth Summer Research Project on Artificial Intelligence was the name of a 1956 summer workshop now considered by many (though not all) to be the seminal event for artificial intelligence as a field322.
Data analysis is obtaining an understanding of data by considering samples, measurement, and visualization. Data analysis can be particularly useful when a dataset is first received, before one builds the first model. It is also crucial in understanding experiments and debugging problems with the system323.
Data analytics is the science of analyzing raw data to make conclusions about that information. Many of the techniques and processes of data analytics have been automated into mechanical processes and algorithms that work over raw data for human consumption324.
Data augmentation in data analysis are techniques used to increase the amount of data. It helps reduce overfitting when training a machine learning325.
Data Cleaning is the process of identifying, correcting, or removing inaccurate or corrupt data records326.
Data Curation – includes the processes related to the organization and management of data which is collected from various sources327.
Data entry – the process of converting verbal or written responses to electronic form328.
Data fusion — the process of integrating multiple data sources to produce more consistent, accurate, and useful information than that provided by any individual data source329.
Data Integration involves the combination of data residing in different resources and then the supply in a unified view to the users. Data integration is in high demand for both commercial and scientific domains in which they need to merge the data and research results from different repositories330.
Data is a collection of qualitative and quantitative variables. It contains the information that is represented numerically and needs to be analyzed.
Data Lake is a type of data repository that stores data in its natural format and relies on various schemata and structure to index the data331.
Data markup is the stage of processing structured and unstructured data, during which data (including text documents, photo and video images) are assigned identifiers that reflect the type of data (data classification), and (or) data is interpreted to solve a specific problem, in including using machine learning methods (National Strategy for the Development of Artificial Intelligence for the period up to 2030)332.
Data Mining is the process of data analysis and information extraction from large amounts of datasets with machine learning, statistical approaches. and many others333.
Data parallelism is a way of scaling training or inference that replicates an entire model onto multiple devices and then passes a subset of the input data to each device. Data parallelism can enable training and inference on very large batch sizes; however, data parallelism requires that the model be small enough to fit on all devices. See also model parallelism334.
Data Processing Unit (DPU) is a programmable specialized electronic circuit with hardware accelerated data processing for data-oriented computing335.
Data protection is the process of protecting data and involves the relationship between the collection and dissemination of data and technology, the public perception and expectation of privacy and the political and legal underpinnings surrounding that data. It aims to strike a balance between individual privacy rights while still allowing data to be used for business purposes336.
Data Refinement is used to convert an abstract data model in terms of sets for example into implementable data structures such as arrays337.
Data Science is a broad grouping of mathematics, statistics, probability, computing, data visualization to extract knowledge from a heterogeneous set of data (images, sound, text, genomic data, social network links, physical measurements, etc.). The methods and tools derived from artificial intelligence are part of this family338,339.
Data set is a set of data that has undergone preliminary preparation (processing) in accordance with the requirements of the legislation of the Russian Federation on information, information technology and information protection and is necessary for the development of software based on artificial intelligence (National strategy for the development of artificial intelligence for the period up to 2030)340.
Data Streaming Accelerator (DSA) is a device that performs a specific task, which in this case is the transfer of data in less time than the CPU would do. What makes DSA special is that it is designed for one of the characteristics that Compute Express Link brings with it over PCI Express 5.0, which is to provide consistent access to RAM for all peripherals connected to a PCI Express port, i.e., they use the same memory addresses.
Data variability describes how far apart data points lie from each other and from the center of a distribution. Along with measures of central tendency, measures of variability give you descriptive statistics that summarize your data341.
Data veracity is the degree of accuracy or truthfulness of a data set. In the context of big data, its not just the quality of the data that is important, but how trustworthy the source, the type, and processing of the data are342.
Data Warehouse is typically an offline copy of production databases and copies of files in a non-production environment343.
Database is a «container» storing data such as numbers, dates or words, which can be reprocessed by computer means to produce information; for example, numbers and names assembled and sorted to form a directory344.
DataFrame is a popular datatype for representing datasets in pandas. A DataFrame is analogous to a table. Each column of the DataFrame has a name (a header), and each row is identified by a number345.
Datalog is a declarative logic programming language that syntactically is a subset of Prolog. It is often used as a query language for deductive databases. In recent years, Datalog has found new application in data integration, information extraction, networking, program analysis, security, and cloud computing346.
Datamining – the discovery, interpretation, and communication of meaningful patterns in data347.
Dataset API (tf. data) is a high-level TensorFlow API for reading data and transforming it into a form that a machine learning algorithm requires. A tf. data. Dataset object represents a sequence of elements, in which each element contains one or more Tensors. A tf.data.Iterator object provides access to the elements of a Dataset. For details about the Dataset API, see Importing Data in the TensorFlow Programmer’s Guide348.
Debugging is the process of finding and resolving bugs (defects or problems that prevent correct operation) within computer programs, software, or systems. Debugging tactics can involve interactive debugging, control flow analysis, unit testing, integration testing, log file analysis, monitoring at the application or system level, memory dumps, and profiling. Many programming languages and software development tools also offer programs to aid in debugging, known as debuggers349.
Decentralized applications (dApps) are digital applications or programs that exist and run on a blockchain or peer-to-peer (P2P) network of computers instead of a single computer. DApps (also called «dapps») are outside the purview and control of a single authority. DApps – which are often built on the Ethereum platform – can be developed for a variety of purposes including gaming, finance, and social media350.
Decentralized control is a process in which a significant number of control actions related to a given object are generated by the object itself on the basis of self-government351.
Decision boundary – the separator between classes learned by a model in a binary class or multi-class classification problems352.
Decision boundary in the case of backpropagation-based artificial neural networks or perceptrons, the type of decision boundary that the network can learn is determined by the number of hidden layers the network has. If it has no hidden layers, then it can only learn linear problems. If it has one hidden layer, then it can learn any continuous function on compact subsets of Rn as shown by the Universal approximation theorem, thus it can have an arbitrary decision boundary.
Decision intelligence (DI) is a practical discipline used to improve the decision making process by clearly understanding and programmatically developing how decisions are made and how the outcomes are evaluated, managed and improved through feedback.
Decision intelligence is a discipline offers a framework to assist data and analytics practitioners develop, model, align, implement, track, and modify decision models and processes related to business results and performance353.
Decision support system (DSS) is an information system that supports business or organizational decision-making activities. DSSs serve the management, operations and planning levels of an organization (usually mid and higher management) and help people make decisions about problems that may be rapidly changing and not easily specified in advance – i.e., unstructured and semi-structured decision problems. Decision support systems can be either fully computerized or human-powered, or a combination of both354.
Decision theory (also theory of choice) – the study of the reasoning underlying an agent’s choices. Decision theory can be broken into two branches: normative decision theory, which gives advice on how to make the best decisions given a set of uncertain beliefs and a set of values, and descriptive decision theory which analyzes how existing, possibly irrational agents actually make decisions355.
Decision threshold this indicator allows you to define the cut-off point for classifying observations. Observations with predicted values greater than the classification cutoff are classified as positive, and those with predicted values less than the cutoff are classified as negative356.
Decision tree is a tree-and-branch model used to represent decisions and their possible consequences, similar to a flowchart357.
Decision tree learning – uses a decision tree (as a predictive model) to go from observations about an item (represented in the branches) to conclusions about the item’s target value (represented in the leaves). It is one of the predictive modeling approaches used in statistics, data mining and machine learning358.
Decision Tree uses tree-like graph or model as a structure to perform decision analysis. It uses each node to represent a test on an attribute, each branch to represent the outcome of the test, and each leaf node to represent a class label359,360,361.
Declarative programming is a programming paradigm – a style of building the structure and elements of computer programs – that expresses the logic of a computation without describing its control flow362,363.
Decoder in general, any ML system that converts from a processed, dense, or internal representation to a more raw, sparse, or external representation. Decoders are often a component of a larger model, where they are frequently paired with an encoder. In sequence-to-sequence tasks, a decoder starts with the internal state generated by the encoder to predict the next sequence. Refer to Transformer for the definition of a decoder within the Transformer architecture364.
Decompression – used to restore data to uncompressed form after compression365.
Deductive classifier is a type of artificial intelligence inference engine. It takes as input a set of declarations in a frame language about a domain such as medical research or molecular biology366.
Deductive Reasoning, also known as logical deduction, is a reasoning method that relies on premises to reach a logical conclusion. It works in a top- down manner, in which the final conclusion is obtained by reducing the general rules that hold the entire domain until only the conclusion is left367.
Deep Blue was a chess supercomputer developed by IBM. It was the first computer chess player that beat the world cham- pion Garry Kasparov, after six-game match in 1997368.
Deep Learning (DL) is a subfield of machine learning concerned with algorithms that are inspired by the human brain that works in a hierarchical way. Deep Learning models, which are mostly based on the (artificial) neural networks, have been applied to different fields, such as speech recognition, computer vision, and natural language processing369.
Deep model is a type of neural network containing multiple hidden layers. Contrast with wide model370.
Deep neural network is a multilayer network containing several (many) hidden layers of neurons between the input and output layers, which allows modeling complex nonlinear relationships. GNNs are now increasingly used to solve such artificial intelligence problems as speech recognition, natural language processing, computer vision, etc., including in robotics371.
Deep Q-Network (DQN) in Q-learning, is a deep neural network that predicts Q-functions. Critic is a synonym for Deep Q-Network372.
DeepMind is an artificial intelligence company founded in 2010 and later acquired by Google in 2014. DeepMind developed Alphago program that beat a human professional Go player for the first time373,374.
Default logic is a non-monotonic logic proposed by Raymond Reiter to formalize reasoning with default assumptions375.
Degree of maturity is the degree of clarity (clarity) of the definition, management, measurement, control and implementation of a specific technological process376.
Demographic parity is a fairness metric that is satisfied if the results of a model’s classification are not dependent on a given sensitive attribute377.
Denoising it is the task of machine vision to remove noise from an image. It is a common supervised learning approach in which noise is artificially added to the dataset and the system removes it on its own378.
Dense feature is a feature in which most values are non-zero, typically a Tensor of floating-point values. Contrast with sparse feature379.
Dense layer – synonym for fully connected layer380.
Depersonalization of personal data – actions, as a result of which it becomes impossible, without the use of additional information, to determine the ownership of personal data by a specific subject of personal data381,382.
Depth – the number of layers (including any embedding layers) in a neural network that learn weights. For example, a neural network with 5 hidden layers and 1 output layer has a depth of 6383.
Depthwise separable convolutional neural network (sepCNN) is a convolutional neural network architecture based on Inception, but where Inception modules are replaced with depthwise separable convolutions. Also known as Xception. A depthwise separable convolution (also abbreviated as separable convolution) factors a standard 3-D convolution into two separate convolution operations that are more computationally efficient: first, a depthwise convolution, with a depth of 1 (n ✕ n ✕ 1), and then second, a pointwise convolution, with length and width of 1 (1 ✕ 1 ✕ n). To learn more, see Xception: Deep Learning with Depthwise Separable Convolutions384.
Description logic is a family of formal knowledge representation languages. Many DLs are more expressive than propositional logic but less expressive than first-order logic. In contrast to the latter, the core reasoning problems for DLs are (usually) decidable, and efficient decision procedures have been designed and implemented for these problems. There are general, spatial, temporal, spatiotemporal, and fuzzy descriptions logics, and each description logic features a different balance between DL expressivity and reasoning complexity by supporting different sets of mathematical constructors385.
Design Center is an organizational unit (the entire organization or its subdivision) that performs a full range or part of the work on creating products up to the stage of its mass production, and also has the necessary personnel, equipment and technologies for this386.
Developmental robotics (DevRob) (also epigenetic robotics) is a scientific field which aims at studying the developmental mechanisms, architectures, and constraints that allow lifelong and open-ended learning of new skills and new knowledge in embodied machines387.
Device is a category of hardware that can run a TensorFlow session, including CPUs, GPUs, and TPUs388.
DevOps (development & operations) is a set of practices, tools, and culture philosophies that automate and integrate the processes of software development teams and IT teams. DevOps emphasizes team empowerment, collaboration and collaboration, and technology automation. The term DevOps is also understood as a special approach to organizing development teams. Its essence is that developers, testers and administrators work in a single thread – they are not each responsible for their own stage, but work together on the release of the product and try to automate the tasks of their departments so that the code moves between stages without delay. In DevOps, responsibility for the result is distributed among the entire team389,390.
Diagnosis concerned with the development of algorithms and techniques that are able to determine whether the behaviour of a system is correct. If the system is not functioning correctly, the algorithm should be able to determine, as accurately as possible, which part of the system is failing, and which kind of fault it is facing. The computation is based on observations, which provide information on the current behaviour391.
Dialogflow API.AI is a platform that allows users to build brand-unique, natural language interactions for bots, applications, services, and devices. It features a Natural Language Understanding Tools to design unique conversation scenarios, design corresponding actions and analyze interactions with users392.
Dialogue system (also conversational agent (CA)) is a computer system intended to converse with a human with a coherent structure. Dialogue systems have employed text, speech, graphics, haptics, gestures, and other modes for communication on both the input and output channel393.
Dice coefficient is a measure to compare the similarity of two segmentations, e.g., by expert and by machine. It is the ratio of twice the number of common pixels to the sum of all pixels in both sets.
Dictation – speech (voice) text input.
Dictation system is a system for speech text input.
Digital Body Language encompasses all the digital activities performed by an individual. Every time a person performs a Google search, visits a web page, opens a newsletter or downloads a guide, they contribute to their digital body language. Digital body language is used in building marketing automation394.
Digital divide is a concept that has become especially widespread in the last decade due to the increased importance of introducing new digital technologies in society and overcoming existing differences in the field of information and knowledge that hinder the development of basic economic and social infrastructures, in particular the energy sector, telecommunications and education395.
Digital educational environment is an open set of information systems designed to support various tasks of the educational process. The word «open» means the ability and the right to use different information systems as part of the DSP, replace them or add new ones at your own discretion396.
Digital ethics is a form of ethics that includes systems of values and moral principles of electronic interaction between people, organizations and things.
Digital platform is a group of technologies that are used as a basis for creating a specific and specialized system of digital interaction397.
Digital rights are the rights of individuals as it pertains to computer access and the ability to use, create and publish digital media. Digital rights can also refer to allowed permissions for fair use of digital copyrighted materials. Digital rights are extensions of human rights like freedom of expression and the right to privacy. The extent to which digital rights are recognized varies from country to country, but Internet access is a recognized right in several countries398.
Digital Social Innovation (DSI) is innovation that uses digital technologies to enable or help carry out SI399.
Digital society (Global information society) is a new world knowledge society that exists and interacts, and is also closely integrated into a fundamentally and qualitatively new digital social, economic and cultural ecosystem, in which the free exchange of information and knowledge is implemented using artificial intelligence, augmented and virtual reality, which are additional interfaces for the interaction of people and machines (computers, robots, wearable devices, etc.)400.
