Artificial Intelligence and Machine Learning for 2025 MeSH (in progress)

Term	Scope Note		Term	Scope Note
Adaptive Algorithms	Algorithms designed to solve a problem by modifying their behaviors when facing with an unseen task with limited data during training.		Genetic Algorithms	Heuristic search algorithms which direct the random searches to the area of better performance or fitness function by computationally creating the conditions inspired by NATURAL SELECTION. They are commonly used to generate solutions for optimization problems.
Artificial Intelligence	Theory and development of COMPUTER SYSTEMS which perform tasks that normally require human intelligence. Such tasks may include speech recognition, LEARNING; VISUAL PERCEPTION; MATHEMATICAL COMPUTING; reasoning, PROBLEM SOLVING, DECISION-MAKING, and translation of language.		Graph Neural Networks	Neural networks designed to process signals supported on graphs.
Autoencoder	An unsupervised learning technique where the machine learns patterns in unlabeled data sets, makes compressed representations and reconstructs the input as output at minimum loss of features from the input. Autoencoders are useful in denoising images, DIMENSIONALITY REDUCTION and feature extraction.		Hidden Markov Models	Finite learnable stochastic models often used in MACHINE LEARNING consisting of two processes: a MARKOV CHAIN and an event observed and analyzed in a state unknown or hidden to the observer. Hidden Markov Models are often used in BIOINFORMATICS, time series analysis and speech and handwriting recognition.
Autonomous Robots	Machines that are engineered to learn about their environment and make decisions to complete tasks with little to no human intervention. These robots may be designed to complete tasks such as patient intervention, remote monitoring, inventory transportation, among many others.		Hopfield Neural Networks	Fully interconnected, single layer neural network, which can be used for associative memory and optimization tasks.
Backtracking Algorithms	A general algorithmic technique that considers solutions one option at a time backtracking to get the desired solution. This allows searching every possible option recursively.		Intelligent Systems	Technological systems that perform automated tasks within specific environments enabled by artificial intelligence.
Boosting Machine Learning Algorithms	A form of ensemble learning methods which combines weak learners to create strong ones, which reduces training errors and bias in the training models. There are different types of boosting such as gradient boosting, XGBoost, and Adaboost.		Long Short Term Memory	A form of recurrent computer neural networks which allows for important information to be retained and less important information to be erased. These kinds of networks are used for tasks such as language translation, anomaly detection, speech and handwriting recognition, and time series prediction.
Cellular Neural Networks, Computer	A type of neural network that consists of a large number of processing elements called cells arranged in a grid-like structure that interact with each other and neighboring cells through a set of predefined rules to perform complex computations.		Machine Learning Algorithms	A class of algorithms which enables a computer to learn as it goes, thereby accomplishing its goal without being explicitly programmed to do so.
Chatbot	Computer program that supplies services to users via conversation in natural language, acting as virtual assistants within social networks or web applications. (From https://ieeexplore.ieee.org/document/8404430)		Multilayer Perceptrons	A feedforward neural network with multiple layers, an input layer, at least three hidden layers, and an output layer, generally used for pattern recognition and classification.
Classification Algorithms	A class of algorithms that predicts which category an observation belongs based on training data.		Multiple-Instance Learning Algorithms	Weakly supervised learning algorithms where training set is arranged in a labeled bag of instances, and a single class label is assigned to a bag.
Clustering Algorithms	Algorithms designed to find groups of similar data points (called clusters) in the dataset. Clustering is often used for feature engineering or pattern discovery in unsupervised machine learning.		Parallel Algorithms	Algorithms that can execute several instructions simultaneously on different processing devices and then combine all the individual outputs to produce the final result.
Compression Algorithms	A class of algorithms which both reduces and restores the size of files to increase their portability. Compression algorithms can be lossy or lossless.		Particle Swarm Optimization	A heuristic computational algorithm designed to find optimal solutions in a random search space. Particle swarm optimization measures swarm intelligence which is a computation measurement inspired by behaviors of self-organized swarming animals.
Convolutional Neural Networks	Artificial neural networks designed to automatically learn features through the use of sequence of layers which transform one volume to the next. Convolutional neural networks initially produce a feature map by use of convolution layers consisting of a set of learnable filters convolved with input, i.e., having smaller widths and heights and the same depth as that of input volume.		Pattern Analysis, Machine	The process of recognizing and extracting meaningful information from patterns in data. Pattern analysis in machine learning is often used for predictions and forecasting.
Detection Algorithms	Algorithms designed to find or localize the objects in images and assign the objects (affordances).		Prediction Algorithms	A class of algorithms which are designed to forecast unseen data. Prediction Algorithms can discover patterns, perform classification tasks, and identify trends.
Dimensionality Reduction	A machine learning process used to reduce the number of attributes or features of data (called dimensionality) thereby simplifying understanding of the dataset without a loss in data integrity.		Prediction Methods, Machine	Methods of making prediction about a new measurement based on calculations of previous data.
Ensemble Learning	A method in machine learning where several base learners are combined to create an optimal learning model. This can be used for a variety of models including forecasting, classification, or function approximation.		Predictive Learning Models	Machine learning models which are trained to analyze historical data to find patterns and trends, allowing it to predict future outcomes.
Extreme Learning Machines	A feedforward neural network designed with a single layer hidden node where the weights between inputs and hidden nodes are randomly assigned. Extreme learning machines can be trained fast and used for the data clustering, grouping, regression, prediction, and forecasting.		Radial Basis Function Networks	A three-layer neural network made of an input layer, hidden layer, and output layer. Radial basis function networks are used for function approximation, interpolation, classification, and time series prediction.
Federated Learning	A method of training machine learning models whereby multiple devices train their own copy of the artificial intelligence model using decentralized local data. These locally trained model results are then aggregated by a central server to update the global model without accessing local data.		Recurrent Neural Networks	A type of neural network where the output from the previous step is fed as input to the current step with the help of a hidden layer which tracks data sequences. Recurrent neural network is often used when the data is sequential, and weights need to be treated same across all the layers such as in time series data and speech recognition. Variation of recurrent neural network include bidirectional neural network and long short-term memory.
Feedforward Neural Networks	A network where information is only passed in one direction. This is the opposite of a recurrent neural network, in which information is processed in a cycle.		Reinforcement Machine Learning	A machine learning training method in which an agent in an uncertain environment is rewarded for optimal behaviors and punished for undesired behaviors, allowing the agent to learn from trial-and-error.
Generative Adversarial Networks	An unsupervised learning network which is composed of two separate networks, a generator which produces content and a discriminator which works to identify real content from the generated content. Over time, generative adversarial networks create data that increasingly resembles the input data.		Representation Machine Learning	A class of machine learning approaches that allow a system to discover the representations required for feature detection or classification from raw data. The requirement for manual feature engineering is reduced by allowing a machine to learn the features and apply them to a given activity.
Generative Artificial Intelligence	An artificial intelligence (AI) technique that generates synthetic artifacts by analyzing training examples; learning their patterns and distribution; and then creating realistic facsimiles. It uses generative modeling and advances in deep learning to produce diverse content at scale by utilizing existing media such as text, graphics, audio, and video. (From Association for Computing Machinery, https://doi.org/10.1145/3626110)		Transfer Machine Learning	Process of training a new learning network by beginning with previously trained for a related problem, leading to less time and energy required for training the new learning network.

Last Reviewed: December 20, 2023