The 212AY Library
AI
Glossary.
200+ artificial intelligence terms explained in plain language — from machine learning and LLMs to AI agents, RAG and prompt engineering. Every definition comes with a concrete example, in English, French and Arabic.
200 terms
Convolutional Neural Network (CNN)
A convolutional neural network is a type of neural network specialized for processing grid-like data such as images, using filters that slide across the input to detect local patterns like edges, textures, and shapes. CNNs revolutionized image recognition: they power facial recognition on phones, quality inspection in factories, and medical image analysis. A Moroccan hospital can use a CNN to screen chest X-rays for tuberculosis, achieving speed and accuracy that complement radiologists rather than replacing them.
Curriculum Learning
Curriculum learning is a training strategy where a model receives easy examples first, gradually increasing difficulty, mirroring how humans learn. Just as students master arithmetic before calculus, models trained on simple cases before complex ones often learn faster and generalize better. This approach reduces training time and improves final performance for tasks with naturally ordered difficulty levels.
Data Augmentation
Data augmentation is the practice of artificially expanding a training dataset by creating modified copies of existing data: rotating or flipping images, paraphrasing text, adding noise to audio, or synthesizing new examples. It helps models generalize better and reduces overfitting, especially when collecting real data is expensive or impractical. A Moroccan hospital with few rare-disease X-rays can augment its dataset by applying rotations and brightness changes to existing scans.
Data Cleaning
Data cleaning is detecting and correcting errors, inconsistencies, and inaccuracies in a dataset before analysis or model training. Tasks include handling missing values, removing duplicates, fixing formatting errors, and resolving outliers. Poor data quality is the leading cause of failed AI projects — a model trained on dirty data produces unreliable results. Automated cleaning pipelines combined with human review are the standard approach.
Data Labeling
Data labeling is assigning meaningful tags or categories to raw data — images, text, audio, or video — so supervised learning models can learn from it. A label might mark an image as containing a cat, tag sentiment as positive, or identify an object's bounding box. Labeling is often the most time-consuming step in an AI project, which is why companies use labeling platforms, crowdsourcing, and AI-assisted annotation to scale.
Data Pipeline
A data pipeline is a series of automated steps that collect, transform, validate, and deliver data from source to destination — a database, dashboard, or ML model. Pipelines handle extraction, cleaning, feature engineering, and loading. Well-built pipelines ensure data is fresh, consistent, and trustworthy. A Moroccan e-commerce company might pipeline data from its website, payment gateway, and logistics into a unified analytics platform.
Dataset
A dataset is a structured collection of data used to train, validate, or test AI models. Quality and diversity of the dataset determine how well a model will perform in the real world: biased data produces biased models. Datasets range from a few hundred rows for a prototype to billions of images for training foundation models. Popular open datasets include ImageNet for vision, Common Crawl for language, and various government open-data portals used by Moroccan data scientists.
Decision Tree
A decision tree is a model that makes predictions by learning a series of if-then rules from data, splitting the data at each step on the feature that best separates the classes. It resembles a flowchart: start at the root, answer a question, follow the branch, repeat. Decision trees are highly interpretable — you can visualize every decision path — making them popular in regulated industries like banking. A Moroccan bank might use one to decide loan approval based on income, age, and employment history.
Deepfake
A deepfake is synthetic media — video, audio, or images — generated by AI to depict someone saying or doing something they never did. The technology uses generative models to swap faces, clone voices, or create fictional scenes. Deepfakes pose serious risks: political manipulation, fraud, and erosion of trust in digital media. Detection tools and digital watermarking are active areas of research to combat them.