Глоссариум по искусственному интеллекту: 2500 терминов. Том 2 - страница 34

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 data