Technologies
Each entry explains a technology, the problems it solves and its connections.
Filters: Mathematics and MethodologyClear
Artificial intelligence
Named in the 1955 Dartmouth proposal; the research project took place in 1956 · matureInformation · Mathematics and MethodologyA set of reusable methods and system capabilities for learning, reasoning, perception, and goal-directed action through computational systems.
Artificial neural networks
The 1943 McCulloch–Pitts model of a logical neuron is an early reference point · matureInformation · Mathematics and MethodologyInterconnected artificial computational units use their connection structure and parameters to represent transformations from inputs to outputs.
Backpropagation
Earlier work on differentiation and training preceded it; a 1986 paper advanced representation learning in multilayer networks · matureInformation · Mathematics and MethodologyA method for computing parameter gradients in differentiable networks by applying the chain rule to propagate derivatives from the output toward earlier layers.
Bayesian networks
Developed into a systematic approach in the 1980s; a 1986 paper and a 1988 book are representative works · matureInformation · Mathematics and MethodologyA method that represents dependencies between variables using a directed acyclic graph and conditional probabilities, and updates beliefs after observing evidence.
Convolutional neural networks
Research began earlier; the 1998 document-recognition paper serves as this entry’s engineering reference point · matureInformation · Mathematics and MethodologyNeural networks that use local connections and weight sharing to process spatial or sequential structure.
Deep learning
Research on training deep models in 2006 and large-scale vision applications in 2012 are important development milestones · matureInformation · Mathematics and MethodologyA machine-learning approach that progressively extracts and combines features through multiple layers of trainable representations.
Deep reinforcement learning
The 2015 DQN paper is a representative reference point for decision-making from pixel inputs; research began earlier · matureInformation · Mathematics and MethodologyReinforcement learning methods that use deep networks to represent values, policies, or models of the environment.
Diffusion generative models
2015: the diffusion-based generative approach; 2020: representative DDPM image-generation results · matureInformation · Mathematics and MethodologyGenerative models that recover data structure from noise by learning a stepwise denoising or reverse-diffusion process.
Electromagnetic induction
Discovered in 1831; self-excited generators developed in the late 19th century · matureEnergy · Information · Mathematics and MethodologyThe physical principle by which changing magnetic flux induces an electromotive force in a conductor, linking mechanical motion with electric current.
Expert systems
Knowledge-based systems such as DENDRAL developed in the 1960s · matureInformation · Mathematics and MethodologySystems that organize domain knowledge and reasoning processes to provide analysis or advice within a limited problem area.
Heuristic search
Earlier explorations existed; the 1968 A* paper serves as this entry’s verified reference point · matureInformation · Mathematics and MethodologyA method that uses problem-specific estimates to guide the search order and reduce the number of candidate states explored.
Reinforcement learning from human feedback
2017: research on rewards learned from preferences; 2022: application to language-model instruction following · matureInformation · Mathematics and MethodologyA method that learns a reward signal from human evaluations or preferences, then uses reinforcement learning to adjust model behavior.
Language-model agents
Represented by the 2022 ReAct preprint and 2023 conference paper; the concept of agents predates them · emergingInformation · Mathematics and MethodologyTask-execution systems that combine language models with tool use, environmental feedback, and multistep decision-making.
Large language models
Language models and pretraining existed earlier; the 2020 GPT-3 paper is a representative milestone in scaling · matureInformation · Mathematics and MethodologyModels trained on large-scale language data that can predict, generate and process language sequences.
Machine learning
Samuel's 1959 paper on learning in a board game is an early representative example, not the sole origin · matureInformation · Mathematics and MethodologyMethods that adjust a model using data or experience from interaction to improve a system's performance on a target task.
Perceptron
Rosenblatt's 1958 paper; this entry focuses on the single-layer classification model · matureInformation · Mathematics and MethodologyAn early neural learning model that classifies inputs using a weighted sum and a threshold, and adjusts weights using misclassified examples.
Reinforcement learning
Its historical roots are earlier; the 1992 Q-learning paper is the representative milestone used in this entry · matureInformation · Mathematics and MethodologyA method for optimizing long-term cumulative return by interacting with an environment, receiving rewards and improving an action policy.
Retrieval-augmented generation
The paper naming RAG appeared in 2020; research combining retrieval and generation predates it · matureInformation · Mathematics and MethodologyA method that introduces content retrieved from external materials during generation to supplement knowledge stored in the model's parameters.
Support vector machines
The 1995 support-vector networks paper is an important reference point for the mature method · matureInformation · Mathematics and MethodologyA supervised learning method that constructs decision boundaries using a margin criterion and can handle nonlinear data through kernel methods.
Symbolic artificial intelligence
Took shape gradually in the 1950s; a 1976 treatment of symbols and search offers a representative synthesis · matureInformation · Mathematics and MethodologyAn approach to artificial intelligence that represents knowledge and solves problems using explicit symbols, rules and reasoning operations.
Transformer
The architecture was proposed in Attention Is All You Need in 2017 · matureInformation · Mathematics and MethodologyA neural network architecture that organizes sequence information processing through attention mechanisms and feed-forward transformations.