Александр Чесалов – Глоссариум по искусственному интеллекту: 2500 терминов. Том 2 (страница 16)

Distribution series are series of absolute and relative numbers that characterize the distribution of population units according to a qualitative (attributive) or quantitative attribute. Distribution series built on a quantitative basis are called variational421.

Divisive clustering – see hierarchical clustering422,423.

Documentation generically, any information on the structure, contents, and layout of a data file. Sometimes called «technical documentation» or «a codebook». Documentation may be considered a specialized form of metadata424.

Documented information – information recorded on a material carrier by means of documentation with details that make it possible to determine such information, or, in cases established by the legislation of the Russian Federation, its material carrier425.

Downsampling – overloaded term that can mean either of the following: Reducing the amount of information in a feature in order to train a model more efficiently. For example, before training an image recognition model, downsampling high-resolution images to a lower-resolution format. Training on a disproportionately low percentage of over-represented class examples in order to improve model training on under-represented classes. For example, in a class-imbalanced dataset, models tend to learn a lot about the majority class and not enough about the minority class. Downsampling helps balance the amount of training on the majority and minority classes426.

Driver is computer software that allows other software (the operating system) to access the hardware of a device427.

Drone – unmanned aerial vehicle (unmanned aerial system)428.

Dropout regularization is a form of regularization useful in training neural networks. Dropout regularization works by removing a random selection of a fixed number of the units in a network layer for a single gradient step. The more units dropped out, the stronger the regularization429.

Dynamic epistemic logic (DEL) is a logical framework dealing with knowledge and information change. Typically, DEL focuses on situations involving multiple agents and studies how their knowledge changes when events occur430.

Dynamic model is a model that is trained online in a continuously updating fashion. That is, data is continuously entering the model431,432.

«E»

Eager execution is a TensorFlow programming environment in which operations run immediately. By contrast, operations called in graph execution don’t run until they are explicitly evaluated. Eager execution is an imperative interface, much like the code in most programming languages. Eager execution programs are generally far easier to debug than graph execution programs433.

Eager learning is a learning method in which the system tries to construct a general, input-independent target function during training of the system, as opposed to lazy learning, where generalization beyond the training data is delayed until a query is made to the system434.

Early stopping is a method for regularization that involves ending model training before training loss finishes decreasing. In early stopping, you end model training when the loss on a validation dataset starts to increase, that is, when generalization performance worsens435.

Earth mover’s distance (EMD) is a measure of the relative similarity between two documents. The lower the value, the more similar the documents436.

Ebert test is a test which gauges whether a computer-based synthesized voice can tell a joke with sufficient skill to cause people to laugh. It was proposed by film critic Roger Ebert at the 2011 TED conference as a challenge to software developers to have a computerized voice master the inflections, delivery, timing, and intonations of a speaking human. The test is similar to the Turing test proposed by Alan Turing in 1950 as a way to gauge a computer’s ability to exhibit intelligent behavior by generating performance indistinguishable from a human being437.

Echo state network (ESN) is a recurrent neural network with a sparsely connected hidden layer (with typically 1% connectivity). The connectivity and weights of hidden neurons are fixed and randomly assigned. The weights of output neurons can be learned so that the network can (re) produce specific temporal patterns. The main interest of this network is that although its behaviour is non-linear, the only weights that are modified during training are for the synapses that connect the hidden neurons to output neurons. Thus, the error function is quadratic with respect to the parameter vector and can be differentiated easily to a linear system438.

Ecosystem of the digital economy is a partnership of organizations that ensures the constant interaction of their technological platforms, applied Internet services, analytical systems, information systems of state authorities of the Russian Federation, organizations and citizens439.

Edge computing is a subspecies of distributed computing in which information processing takes place in close proximity to the place where the data was received and will be consumed (for example, using phones and other consumer devices)440.

Electronic circuit is a product, a combination of individual electronic components, such as resistors, capacitors, diodes, transistors and integrated circuits, interconnected to perform any task or a circuit with conventional signs441,442.

Electronic Data Interchange (EDI) is a series of standards and conventions for the transfer of structured digital information between organizations, based on certain regulations and formats of transmitted messages443.

Electronic government (e-Government) is a package of technologies and a set of related organizational measures, regulatory and legal support for organizing digital interaction between public authorities of various branches of government, citizens, organizations and other economic entities444.

Electronic industry is a set of organizations that perform scientific, technological and other work in the field of development, production, maintenance of operation, as well as providing services related to electronic and microelectronic products, respectively445.

Electronic Medical Record (EMR) is electronic health record, is the systematized collection of patient and population electronically stored health information in a digital format. These records can be shared across different healthcare settings446.

Electronic state is a way of implementing the information aspects of state activity based on the use of IT systems, as well as a new type of state based on the use of this technology. In the Russian Federation, activities to create an «electronic state» are carried out within the framework of the federal target program «Electronic Russia»447,448.

Eli5 environment is a Python environment that is used to debug and visualize machine learning models. By default, it supports several machine learning frameworks – Scikit-learn, XGBoost, LightGBM, CatBoost, lightning, Keras and so on. Eli5 also provides LIME and Permutation Importance models to test machine learning pipelines as black boxes449.

ELIZA effect is a term used to discuss progressive artificial intelligence. It is the idea that people may falsely attach meanings of symbols or words that they ascribe to artificial intelligence in technologies450.

Embedding (Word Embedding) is one instance of some mathematical structure contained within another instance, such as a group that is a subgroup451.

Embedding space – the d-dimensional vector space that features from a higher-dimensional vector space are mapped to. Ideally, the embedding space contains a structure that yields meaningful mathematical results; for example, in an ideal embedding space, addition and subtraction of embeddings can solve word analogy tasks. The dot product of two embeddings is a measure of their similarity452.

Embeddings is a categorical feature represented as a continuous-valued feature. Typically, an embedding is a translation of a high-dimensional vector into a low-dimensional space453.

Embodied agent (also interface agent) is an intelligent agent that interacts with the environment through a physical body within that environment. Agents that are represented graphically with a body, for example a human or a cartoon animal, are also called embodied agents, although they have only virtual, not physical, embodiment454.

Embodied cognitive science is an interdisciplinary field of research, the aim of which is to explain the mechanisms underlying intelligent behavior. It comprises three main methodologies: 1) the modeling of psychological and biological systems in a holistic manner that considers the mind and body as a single entity, 2) the formation of a common set of general principles of intelligent behavior, and 3) the experimental use of robotic agents in controlled environments455.

Empirical risk minimization (ERM) – choosing the function that minimizes loss on the training set. Contrast with structural risk minimization456,457.

Encoder in general, is any system that converts from a raw, sparse, or external representation into a more processed, denser, or more internal representation. Encoders are often a component of a larger model, where they are frequently paired with a decoder. Some Transformers pair encoders with decoders, though other Transformers use only the encoder or only the decoder. Some systems use the encoder’s output as the input to a classification or regression network. In sequence-to-sequence tasks, an encoder takes an input sequence and returns an internal state (a vector). Then, the decoder uses that internal state to predict the next sequence. Refer to Transformer for the definition of an encoder in the Transformer architecture458.