Technology

Optical fiber is a thin glass or plastic waveguide that transmits light for telecommunications, sensing, illumination, and other optical applications.

Overfitting occurs when a model learns sample-specific patterns that impair its performance on previously unseen data.

A p–n junction is the boundary between p-type and n-type semiconductor regions, whose internal electric field enables rectification, carrier injection, and light–electricity conversion.

Papermaking is the craft of dispersing plant fibers in water, filtering them onto a screen as a thin sheet, then pressing and drying it. It originated in Han China and is one of the Four Great Inventions.

Papyrus is a plant-based writing material developed in ancient Egypt and widely used throughout the ancient Mediterranean.

Parallel computing uses multiple processing resources simultaneously to execute parts of a computational workload, improving speed or enabling larger problems.

Parchment is a writing and binding material made from animal skin that is cleaned, scraped, and dried under tension rather than tanned.

A perceptron is a linear classification model and learning algorithm that adjusts weighted inputs to distinguish between classes.

Polymerase chain reaction is a laboratory technique that selectively amplifies DNA through repeated cycles of strand separation, primer binding, and enzymatic synthesis.

Positional encoding supplies neural networks with information about the order or spatial location of elements in an input.

Printing is the mass reproduction of text and images by transferring ink from a prepared surface onto paper or other materials. It began with East Asian woodblocks.

A programming language is a formal system for expressing computations through rules that define program structure and behavior.

Prompt engineering is the design, testing, and refinement of inputs that guide generative artificial intelligence models toward specified outputs.

Q-learning is a model-free reinforcement learning algorithm that estimates optimal action values from experience to learn decisions maximizing expected cumulative reward.

A computer that processes quantum information to perform computations using superposition, entanglement, and interference.

Quantum error correction protects quantum information by encoding it redundantly and identifying errors without measuring the stored logical state.

A quantum gate is a reversible operation on qubits, forming a basic building block of quantum circuits.

A qubit is a two-level quantum information unit whose states can exhibit superposition and entanglement.

Radar uses radio waves to detect objects and measure their distance, direction, motion, or physical characteristics.

A railway is a guided transport system in which trains carry passengers or freight along tracks, supported by infrastructure, vehicles, signalling, and operating procedures.

A machine-learning method that combines randomized decision trees to make classification or regression predictions.

A recurrent neural network processes sequential data by repeatedly updating an internal state that carries information between sequence positions.

Recycling converts discarded materials into usable raw materials and products through collection, sorting, and reprocessing.

Reinforcement learning is a machine-learning approach in which agents learn decision-making policies by optimizing cumulative rewards from interaction or recorded experience.

Reinforcement learning from human feedback trains models using human judgments as a source of reward, often to improve instruction following and other desired behaviors.

Remote sensing acquires information about objects and environments from a distance, using measured signals to map their properties and changes.

Representation learning discovers useful features from data, enabling machine-learning systems to organize information and transfer it across tasks.

Retrieval-augmented generation combines information retrieval with generative models to produce responses informed by external sources.

Robotics is the interdisciplinary study and engineering of machines that sense, act, and perform tasks in the physical world.

Self-attention is a neural-network mechanism that contextualizes elements of an input by weighting and combining information from other elements of the same input.

A machine-learning approach that derives training targets from the data itself, enabling learning without manually supplied task labels.

Semantic segmentation assigns a category label to each image pixel, producing a detailed map of objects, materials, and scene regions.

A machine-learning framework that combines labeled and unlabeled examples to learn predictive models when task-specific annotations are limited.

Shared memory is storage accessible to multiple execution agents, enabling communication through common data rather than exclusively through explicit messages.

Signal processing is the analysis and transformation of signals to extract information, improve representations, or support communication and decision-making.

Software documentation describes a software system’s requirements, design, interfaces, operation, and use for users, developers, and maintainers.

A solar cell is a semiconductor device that converts light directly into electrical energy through the photovoltaic effect.

Speech recognition is the computational conversion of spoken language into text, using acoustic analysis and statistical or neural models.

A steam engine is a heat engine that uses steam to do mechanical work. It powered mines, factories, railways and ships during industrialization.

A steam turbine converts energy in pressurized steam into rotating mechanical power for electricity generation, industrial machinery, and marine propulsion.

A steamship is a vessel propelled by steam-powered machinery, a technology that transformed inland navigation and ocean transport from the nineteenth century.

A superconducting quantum interference device (SQUID) detects minute changes in magnetic flux through quantum interference in a superconducting circuit.

A superconducting qubit stores quantum information in selected energy states of a low-temperature superconducting electrical circuit containing Josephson junctions.

Supervised learning trains predictive models from examples that pair inputs with known target outputs.

A support vector machine is a learning model that uses margin-based optimization and optional kernels for classification, regression, and related tasks.

Symbolic artificial intelligence represents knowledge explicitly and uses rules, logic, and search to derive conclusions or select actions.

Synchronization coordinates concurrent computations by controlling access to shared resources, ordering operations, and establishing when changes become visible.

Telecommunications is the transmission and reception of information over distance through electrical, radio, optical, and related electromagnetic systems.

A family of learning methods that updates predictions using rewards and differences between successive estimates of future outcomes.

A Tensor Processing Unit is a Google-designed processor specialized for accelerating neural-network training and inference through high-throughput matrix computation.

A test set is data reserved for evaluating a trained machine-learning model independently of the decisions used to develop it.

Training data consists of examples used to fit machine-learning models, shaping their learned patterns, capabilities, and limitations.

A machine-learning approach that reuses knowledge from one task or domain to improve learning in another.

A neural-network architecture that uses attention to process sequences, supporting language understanding, text generation, and image recognition.

A transistor is a semiconductor device that controls electrical current, serving as an amplifier or switch in electronic circuits.

A Turing machine is an abstract computational model used to define algorithms, establish limits of computability, and analyze computational resources.

Underfitting occurs when a learned model fails to capture important patterns in its training data, limiting predictive performance on both familiar and unseen examples.

A machine-learning paradigm that discovers patterns, representations, or probability distributions in data without externally supplied target labels.

A validation set is data reserved for evaluating and selecting machine-learning models during development, distinct from training data and final test data.

The vanishing gradient problem occurs when derivatives shrink during backpropagation, weakening learning signals across many neural-network layers or time steps.