The least squares regression tree is a commonly used regression tree algorithm. In order to minimize the square error, it is necessary to traverse the values of each feature in turn and calculate the error of each possible split point at the moment. Finally, the point with the smallest split error is selected and the input space is split into two parts. The above steps are repeated recursively until the split is completed. This method splits […]

Latent Dirichlet Allocation (LDA) is a topic model that can express the topic of each document in a document set in the form of probability distribution. It is also an unsupervised learning algorithm that does not require a manually annotated training set for training. It only requires a document set and the number of specified topics K. In addition, for each topic, […]

The minimum description length is also called the minimum description length principle, which means that among multiple data storage hypotheses, the hypothesis that can produce the most data compression effect is the best. The minimum description length principle is a result of the formalization of Occam's razor, introduced by Jorma Rissanen in 1978. Its main point is: in any […]

Meta-learning is a subfield of machine learning, where automatic learning algorithms are applied to the metadata of machine learning experiments. The core of meta-learning is the "learning (training)" process, that is, studying how to make neurons make good use of old knowledge so that they can adjust themselves according to new tasks. The diagram is as follows: The initial parameters of the neural network (blue ■ […]

Max pooling is a commonly used pooling operation that reduces the amount of data by maximizing the value. Its usual operation is to divide the input image into several rectangular regions and output the maximum value for each subdomain. Currently, in addition to max pooling, average pooling is also commonly used. It reduces the complex calculations from the upper hidden layer and can be unaffected by the target […]

The maximum margin refers to the situation in the classification task where the training data interval is the largest. The hyperplane that satisfies the maximum margin can classify the data into positive and negative categories, and the decision confidence is the highest at this time. In the support vector machine, there are hard margin maximization and soft margin maximization, both of which are strategies for finding the maximum margin.

Markov random field is also called Markov network. It is a model described by an undirected graph, which contains a set of nodes. Each node corresponds to a single variable or a set of variables. The links between nodes are undirected. Markov random field describes that among multiple data points, there is no clear relationship between two points in front and back, or in the direction. Although there is a mutual relationship between the two points, […]

Automatic summarization is the process of shortening text documents using software in order to create summaries that contain the main points of the original document. It is currently part of the field of machine learning and data mining, and its purpose is to find subsets of data that contain relevant "information". There are currently two methods for automatic summarization: extraction and abstraction. Extraction is based on words, phrases in the original text […]

Automatic speech recognition technology is a technology that converts human speech into text. Due to the diversity and complexity of speech signals, the current speech recognition system can only achieve satisfactory performance under certain restrictions (it can only be applied to certain specific occasions). Automatic speech recognition definition The goal of automatic speech recognition technology is to enable computers to "[…]

An autoencoder is an artificial neural network used for efficient encoding in unsupervised learning, usually for dimensionality reduction. It is also a data compression algorithm in which the compression and decompression functions of the data are lossy and automatically learned from samples, but in most cases where autoencoders are mentioned, the compression and decompression functions are learned through a neural network […]

The human attention mechanism is based on intuition. It is a means for humans to use limited resources to quickly filter out high-value information from a large amount of information. The attention mechanism in deep learning draws on the human attention thinking mode and is widely used in various scenarios such as natural language processing, image classification, speech recognition, etc., and has achieved remarkable results.

AUC is defined as the area under the ROC curve and the coordinate axis. Since the ROC curve is above y=x, the value range of AUC is between 0.5 and 1. AUC can be used as an indicator of the quality of the model when comparing different classification models. Its main significance lies in AU […]

The Hessian matrix, also known as the Hessian matrix or the Hessian matrix, is a block matrix consisting of the second-order partial derivatives of multivariate real-valued functions. This is a generalization of the second-order derivative to multivariate functions and is closely related to the extreme values of the function.

Alpha-beta pruning is a search algorithm used to reduce the number of nodes in the Minimax search tree.

Adaptive resonance theory (ART for short) refers to a theoretical model that states that when there is interaction between a neural network and the environment, the encoding of environmental information will spontaneously occur in the neural network, and the network can self-organize to generate the encoding of environmental knowledge.

Application-specific integrated circuits, or ASICs for short, are integrated circuits with special specifications that are customized according to different product requirements; on the contrary, non-customized ones are application-specific standard product (ASSP) integrated circuits.

An intelligent agent refers to a software or hardware entity that can act autonomously. It has been translated as "agent", "agent", "intelligent subject", etc.

The original sampling method is a basic sampling method for directed graph models, which refers to generating samples from the joint distribution represented by the model, also known as the ancestral sampling method.

Anomaly detection is to find objects that are different from most objects, in fact, it is to find outliers. Anomaly detection is sometimes also called deviation detection. Abnormal objects are relatively rare.

The learning rule is a concept in neural network models that represents how the weights in the network are adjusted over time. This is generally viewed as a long-term dynamical rule.

The actor-critic algorithm is a reinforcement learning algorithm that combines a policy network and a value function. It uses the reward and punishment information of the results to calculate the probability of taking various actions under different states. It is also called the AC algorithm.

The task of the acoustic model is to calculate P(O|W), which is the probability of generating a speech waveform for the model. The acoustic model is one of the most important parts of the speech recognition system. It accounts for most of the computational overhead of speech recognition and determines the performance of the speech recognition system.

The adaptive bitrate algorithm is a video transmission technology that automatically adjusts the streaming media bitrate. The adjustment factors mainly depend on the network conditions or client delay.

The Tensor Processing Unit (TPU) is a special-purpose integrated circuit developed specifically for machine learning.

The least squares regression tree is a commonly used regression tree algorithm. In order to minimize the square error, it is necessary to traverse the values of each feature in turn and calculate the error of each possible split point at the moment. Finally, the point with the smallest split error is selected and the input space is split into two parts. The above steps are repeated recursively until the split is completed. This method splits […]

Latent Dirichlet Allocation (LDA) is a topic model that can express the topic of each document in a document set in the form of probability distribution. It is also an unsupervised learning algorithm that does not require a manually annotated training set for training. It only requires a document set and the number of specified topics K. In addition, for each topic, […]

The minimum description length is also called the minimum description length principle, which means that among multiple data storage hypotheses, the hypothesis that can produce the most data compression effect is the best. The minimum description length principle is a result of the formalization of Occam's razor, introduced by Jorma Rissanen in 1978. Its main point is: in any […]

Meta-learning is a subfield of machine learning, where automatic learning algorithms are applied to the metadata of machine learning experiments. The core of meta-learning is the "learning (training)" process, that is, studying how to make neurons make good use of old knowledge so that they can adjust themselves according to new tasks. The diagram is as follows: The initial parameters of the neural network (blue ■ […]

Max pooling is a commonly used pooling operation that reduces the amount of data by maximizing the value. Its usual operation is to divide the input image into several rectangular regions and output the maximum value for each subdomain. Currently, in addition to max pooling, average pooling is also commonly used. It reduces the complex calculations from the upper hidden layer and can be unaffected by the target […]

The maximum margin refers to the situation in the classification task where the training data interval is the largest. The hyperplane that satisfies the maximum margin can classify the data into positive and negative categories, and the decision confidence is the highest at this time. In the support vector machine, there are hard margin maximization and soft margin maximization, both of which are strategies for finding the maximum margin.

Markov random field is also called Markov network. It is a model described by an undirected graph, which contains a set of nodes. Each node corresponds to a single variable or a set of variables. The links between nodes are undirected. Markov random field describes that among multiple data points, there is no clear relationship between two points in front and back, or in the direction. Although there is a mutual relationship between the two points, […]

Automatic summarization is the process of shortening text documents using software in order to create summaries that contain the main points of the original document. It is currently part of the field of machine learning and data mining, and its purpose is to find subsets of data that contain relevant "information". There are currently two methods for automatic summarization: extraction and abstraction. Extraction is based on words, phrases in the original text […]

Automatic speech recognition technology is a technology that converts human speech into text. Due to the diversity and complexity of speech signals, the current speech recognition system can only achieve satisfactory performance under certain restrictions (it can only be applied to certain specific occasions). Automatic speech recognition definition The goal of automatic speech recognition technology is to enable computers to "[…]

An autoencoder is an artificial neural network used for efficient encoding in unsupervised learning, usually for dimensionality reduction. It is also a data compression algorithm in which the compression and decompression functions of the data are lossy and automatically learned from samples, but in most cases where autoencoders are mentioned, the compression and decompression functions are learned through a neural network […]

The human attention mechanism is based on intuition. It is a means for humans to use limited resources to quickly filter out high-value information from a large amount of information. The attention mechanism in deep learning draws on the human attention thinking mode and is widely used in various scenarios such as natural language processing, image classification, speech recognition, etc., and has achieved remarkable results.

AUC is defined as the area under the ROC curve and the coordinate axis. Since the ROC curve is above y=x, the value range of AUC is between 0.5 and 1. AUC can be used as an indicator of the quality of the model when comparing different classification models. Its main significance lies in AU […]

The Hessian matrix, also known as the Hessian matrix or the Hessian matrix, is a block matrix consisting of the second-order partial derivatives of multivariate real-valued functions. This is a generalization of the second-order derivative to multivariate functions and is closely related to the extreme values of the function.

Alpha-beta pruning is a search algorithm used to reduce the number of nodes in the Minimax search tree.

Adaptive resonance theory (ART for short) refers to a theoretical model that states that when there is interaction between a neural network and the environment, the encoding of environmental information will spontaneously occur in the neural network, and the network can self-organize to generate the encoding of environmental knowledge.

Application-specific integrated circuits, or ASICs for short, are integrated circuits with special specifications that are customized according to different product requirements; on the contrary, non-customized ones are application-specific standard product (ASSP) integrated circuits.

An intelligent agent refers to a software or hardware entity that can act autonomously. It has been translated as "agent", "agent", "intelligent subject", etc.

The original sampling method is a basic sampling method for directed graph models, which refers to generating samples from the joint distribution represented by the model, also known as the ancestral sampling method.

Anomaly detection is to find objects that are different from most objects, in fact, it is to find outliers. Anomaly detection is sometimes also called deviation detection. Abnormal objects are relatively rare.

The learning rule is a concept in neural network models that represents how the weights in the network are adjusted over time. This is generally viewed as a long-term dynamical rule.

The actor-critic algorithm is a reinforcement learning algorithm that combines a policy network and a value function. It uses the reward and punishment information of the results to calculate the probability of taking various actions under different states. It is also called the AC algorithm.

The task of the acoustic model is to calculate P(O|W), which is the probability of generating a speech waveform for the model. The acoustic model is one of the most important parts of the speech recognition system. It accounts for most of the computational overhead of speech recognition and determines the performance of the speech recognition system.

The adaptive bitrate algorithm is a video transmission technology that automatically adjusts the streaming media bitrate. The adjustment factors mainly depend on the network conditions or client delay.

The Tensor Processing Unit (TPU) is a special-purpose integrated circuit developed specifically for machine learning.