Glosarium AI
Kamus lengkap Kecerdasan Buatan
Convolutional Neural Networks
Deep learning architecture specialized in processing images and spatial data. Uses convolutional layers to automatically extract hierarchical features.
Deep Reinforcement Learning
Combination of reinforcement learning with deep neural networks. Enables agents to learn optimal strategies in complex environments.
Natural Language Processing
Field of AI that enables machines to understand, interpret and generate human language. Includes sentiment analysis, translation and text generation.
Recommendation Systems
Algorithms that suggest relevant items to users based on their preferences and behaviors. Widely used in e-commerce, streaming, and social networks.
Computer Vision
Allows computers to interpret and understand the visual content of images and videos. Applications: object detection, facial recognition, medical analysis.
Supervised Machine Learning
Learning method where the model learns from labeled data to make predictions. Includes classification and regression.
Unsupervised Machine Learning
Techniques for exploring unlabeled data to discover hidden structures. Primarily clustering and dimensionality reduction.
Recurrent Neural Networks
Deep learning architecture designed to process sequential data. Internal memory allows it to capture temporal dependencies.
Transformers and Attention Architecture
Revolutionary architecture based on the attention mechanism for processing sequences. Foundation of modern language models like GPT and BERT.
Transfer Learning
Technique reusing pre-trained models on large data for specific tasks. Drastically reduces the need for data and training time.
Feature Engineering
Process of creating and selecting optimal variables for machine learning models. Crucial step directly impacting algorithm performance.
Cross-Validation and Model Evaluation
Statistical techniques to rigorously assess ML model performance. Essential to prevent overfitting and ensure generalization.
Big Data and Distributed Computing
Infrastructure and algorithms for processing massive volumes of data. Uses frameworks like Spark, Hadoop for parallel computing.
Exploratory Data Science
Initial analysis phase to discover patterns, anomalies, and relationships in data. Combines statistics and visualization.
Online Learning and Streaming
Adaptive learning methods for continuous real-time data. Models updated incrementally without full retraining.
Federated Learning
Distributed approach where training is done locally on devices without centralizing data. Preserves user privacy.
Interpretability and Explainability of AI
Set of techniques to understand and explain AI model decisions. Critical for trust and regulation of autonomous systems.
Multi-Agent Reinforcement Learning
Extension of RL where multiple agents learn simultaneously, often in competition or cooperation. Applications in gaming, robotics, and economics.
Retrieval-Augmented Generation (RAG)
Architecture combining document retrieval and text generation. Improves accuracy and reduces hallucinations of LLMs.
Large Language Models
Massive neural networks pre-trained on huge text corpora. Capable of advanced natural language understanding and generation.
Traitement du Signal et Séries Temporelles
Techniques spécialisées pour analyser des données séquentielles et temporelles. Applications en finance, IoT et prévisions météorologiques.
Meta-Learning
Learning to learn: models that discover how to quickly adapt to new tasks with few examples. Also called few-shot learning.
Anomaly detection
Identification of patterns or observations that deviate significantly from the normal. Crucial in security, finance, and predictive maintenance.
Graph Neural Networks
Deep learning architecture specialized in processing data structured as graphs. Applications in social networks, molecules, and recommendation systems.
MLOps and AI Industrialization
DevOps practices adapted to the lifecycle of ML models. Automation of deployment, monitoring, and updating of AI systems in production.
AutoML and ML Automation
Systems automating the complete process of creating ML models. Reduces required expertise and accelerates the development of AI solutions.
Edge AI and Embedded Artificial Intelligence
Deployment of AI models directly on edge devices. Latency reduction, privacy preservation and offline operation.
AI Ethics and Algorithmic Bias
Study of the moral and social implications of AI systems. Detection and mitigation of biases to ensure fairness and non-discrimination.
Security and Privacy-Preserving ML
Techniques protecting models and data against adversarial attacks. Includes homomorphic encryption and differential privacy.
Apprentissage par Renforcement Classique
Ensemble des méthodes fondamentales d'apprentissage par renforcement incluant Q-learning, SARSA, et les méthodes de programmation dynamique pour la prise de décision séquentielle.
Arbres de Décision et Méthodes d'Ensemble
Techniques basées sur les structures arborescentes comme Random Forest, Gradient Boosting, et XGBoost pour la classification et la régression robustes.
Machines à Vecteurs de Support
Algorithmes d'apprentissage supervisé utilisant des hyperplans pour la classification maximisant la marge entre les classes, avec extensions aux noyaux non-linéaires.
Modèles Génératifs Avancés
Ensemble des techniques de génération de données incluant GANs, VAEs, modèles de diffusion, et auto-encodeurs pour la création synthétique de contenu.
Intelligence Artificielle Symbolique
Approche de l'IA basée sur la manipulation de symboles et règles logiques, incluant les systèmes experts et le raisonnement déductif.
Algorithmes Évolutionnaires
Méthodes d'optimisation inspirées de l'évolution naturelle incluant algorithmes génétiques, stratégies d'évolution, et programmation génétique.
Apprentissage Semi-Supervisé
Techniques combinant données étiquetées et non étiquetées pour améliorer les performances des modèles lorsque les données étiquetées sont rares.
Apprentissage par Contraste
Paradigme d'apprentissage auto-supervisé basé sur la comparaison de paires d'exemples pour apprendre des représentations discriminatives.
Réseaux Bayésiens
Modèles graphiques probabilistes représentant les dépendances conditionnelles entre variables pour l'inférence et la prise de décision sous incertitude.
Réduction de Dimensionnalité
Ensemble des techniques (ACP, t-SNE, UMAP) pour réduire la complexité des données tout en préservant l'information pertinente.
Active Learning
Strategies where the model intelligently selects samples to label in order to optimize learning with a limited annotation budget.
Change Detection
Techniques for identifying transitions in data distributions and continuously adapting models to new contexts.
Self-Supervised Learning
Paradigm that automatically creates labels from unlabeled data to pre-train models on proxy tasks.
Collective Intelligence
Approaches inspired by the collective behavior of social insects for optimization and distributed problem solving.
Spiking Neural Networks
Neuromorphic models mimicking the temporal communication of biological neurons for more efficient and bio-inspired computing.
Incremental Learning
Ability of models to continuously learn new data without forgetting previously acquired knowledge.
Model Quantization
Neural network compression techniques reducing weight precision to optimize memory and computation.
Causal Learning
Field studying cause-and-effect relationships in data to improve model generalization and robustness.
Adversarial Attacks and Defense
Study of AI model vulnerabilities to malicious perturbations and development of protection techniques.
IA Quantique
Intersection de l'informatique quantique et de l'IA exploitant les phénomènes quantiques pour accélérer les algorithmes d'apprentissage.
Imitation Learning
Techniques where an agent learns by imitating expert demonstrations without requiring explicit rewards.