// Knowledge Base

Artificial Intelligence
Lexicon A–Z

Expert reference for AI, Machine Learning, and Deep Learning terminology. From foundational concepts to cutting-edge research — for practitioners, researchers, and decision-makers.

500+
Defined Terms
26
Categories A–Z
2026
Continuously Updated
ABCD EFGH IJKL MNOP QRST UVWX YZ
// In-depth articles
Deep Dive
Transformer Models
How attention mechanisms revolutionized AI — from BERT to GPT-4 and beyond. Architecture, training, and real-world applications explained.
Deep Dive
Large Language Models (LLMs)
What makes a language model "large"? Scaling laws, emergent capabilities, RLHF, and the architecture behind modern AI assistants.
Deep Dive
Reinforcement Learning
From Q-learning to PPO and AlphaGo. How agents learn from rewards, and why RL is central to modern AI breakthroughs.
Deep Dive
AI Benchmarking
How to evaluate AI models rigorously — MLPerf, MMLU, HumanEval and beyond. What benchmarks actually measure and their limitations.
Deep Dive
AutoML
Automated Machine Learning — how systems select models, tune hyperparameters, and optimize pipelines without human intervention.
Deep Dive
Adaptive Computing
FPGAs, dynamic resource allocation, and real-time adaptation for AI workloads in edge and cloud environments.
A
Adaptive Computing
Technologies that dynamically adjust CPU, GPU, and memory allocation based on real-time AI workload demands.
AI (Artificial Intelligence)
A branch of computer science creating systems that simulate human reasoning, learning, and decision-making.
Algorithm
A finite set of well-defined instructions for solving a problem or performing a computation.
Artificial Neural Network (ANN)
A layered computational model inspired by biological neurons, capable of learning patterns from data.
Attention Mechanism
A technique allowing models to weight the relevance of different input parts when producing output — the core of Transformers.
Artificial General Intelligence (AGI)
A hypothetical AI with general cognitive capabilities matching or exceeding human intelligence across all domains.
Automated Machine Learning (AutoML)
Systems that automate model selection, feature engineering, and hyperparameter tuning.
Anomaly Detection
Identifying data points that deviate significantly from expected behavior — used in fraud detection, monitoring, and QA.
Adversarial Networks (GAN)
Two competing neural networks — a generator and discriminator — trained together to produce realistic synthetic data.
AI Ethics
The interdisciplinary study of moral implications in AI design, deployment, and governance.
AI Benchmarking
Systematic evaluation of AI model performance using standardized datasets, metrics, and comparison frameworks like MLPerf.
Adaptive Algorithms
Algorithms that modify their behavior based on changing input conditions or feedback from the environment.
B
Backpropagation
The algorithm for computing gradients in neural networks by propagating errors backward through layers to update weights.
Bayesian Networks
Directed acyclic graphs representing probabilistic relationships between variables, used for reasoning under uncertainty.
Batch Normalization
Normalizing layer inputs during training to stabilize and accelerate deep neural network optimization.
Bias in AI
Systematic errors in AI outputs caused by skewed training data or flawed model assumptions.
Boosting
An ensemble technique combining weak learners sequentially, each correcting the errors of the previous one (e.g. XGBoost).
Behavioral Cloning
Supervised imitation learning where a model replicates expert behavior directly from demonstration data.
C
Computer Vision
Enabling machines to interpret and understand visual information — images, video, and 3D scenes.
Convolutional Neural Networks (CNNs)
Neural networks using convolution operations to extract spatial features from images — the backbone of visual AI.
Cross-Validation
A model evaluation technique that partitions data into k folds to measure generalization performance.
Clustering
Unsupervised grouping of similar data points — includes k-means, DBSCAN, and hierarchical methods.
Causal Inference
Determining cause-and-effect relationships from observational data, going beyond correlation.
Continuous Learning
AI systems that update their knowledge incrementally without catastrophic forgetting of prior knowledge.
D
Deep Learning
Machine learning using multi-layered neural networks to model hierarchical representations of data.
Data Mining
Discovering hidden patterns, correlations, and anomalies in large datasets using statistical and ML techniques.
Decision Trees
Tree-structured models making decisions by splitting data on feature values — interpretable and widely used.
Dimensionality Reduction
Reducing feature space (PCA, t-SNE, UMAP) while preserving essential data structure and variance.
Distributed Computing
Splitting computation across multiple nodes — essential for training large AI models at scale.
Digital Twin
A real-time virtual replica of a physical system, enhanced with AI for simulation and predictive analytics.
Dropout
A regularization technique randomly deactivating neurons during training to prevent overfitting.
Deep Reinforcement Learning
Combining deep neural networks with RL — used in game-playing AI (AlphaGo, OpenAI Five) and robotics.
E
Ensemble Learning
Combining multiple models (bagging, boosting, stacking) to reduce variance and improve prediction accuracy.
Edge AI
Deploying AI inference on edge devices (IoT, embedded systems) to minimize latency and bandwidth usage.
Explainable AI (XAI)
Methods for making AI decisions interpretable to humans — critical for regulated industries and trust.
Embedding
Dense vector representations of discrete objects (words, entities) in continuous space, capturing semantic relationships.
Evolutionary Algorithms
Optimization inspired by natural selection — genetic algorithms, evolution strategies, and differential evolution.
F
Federated Learning
Decentralized training where model updates — not raw data — are aggregated, preserving data privacy.
Feature Engineering
Creating, selecting, and transforming input variables to maximize model performance.
Fine-Tuning
Adapting a pre-trained model to a specific task using a smaller domain-specific dataset.
Few-Shot Learning
Training models to generalize from very few labeled examples — mimicking human rapid learning.
Fuzzy Logic
A multi-valued logic system handling approximate reasoning and partial truth values.
G
Generative Adversarial Networks (GANs)
Generator-discriminator pairs trained adversarially to produce realistic synthetic images, audio, or text.
Gradient Descent
Iterative optimization minimizing loss by moving model parameters in the direction of steepest descent.
Graph Neural Networks (GNNs)
Neural networks operating on graph-structured data — used in drug discovery, social networks, and knowledge graphs.
Generative AI
AI capable of creating novel content — text, images, code, audio — by learning data distributions.
Gradient Boosting
Builds models sequentially, each correcting predecessor errors — XGBoost and LightGBM are leading implementations.
Generative Pre-trained Transformers (GPT)
Large autoregressive language models pre-trained on massive text corpora and fine-tuned for various tasks.
H
Hyperparameter Tuning
Optimizing model configuration parameters (learning rate, batch size, layers) that are set before training begins.
Human-in-the-Loop (HITL)
AI systems incorporating human feedback during training or inference to improve quality and safety.
Hidden Markov Models (HMMs)
Statistical models for sequential data with hidden states — foundational in speech recognition.
Hybrid AI
Combining symbolic AI (rules, logic) with neural networks to leverage both interpretability and learning power.
I
Image Recognition
Classifying and identifying objects, scenes, or faces in digital images using deep learning models.
Inference
Using a trained model to generate predictions on new, unseen data — the deployment phase of AI.
Incremental Learning
Continuously updating models as new data arrives without retraining from scratch.
Imitation Learning
Training AI agents to mimic expert behavior from demonstrations — used in robotics and autonomous driving.
K
K-Means Clustering
Partitioning data into k clusters by minimizing intra-cluster variance — fast and widely applicable.
Knowledge Graphs
Graph-based structures representing entities and relationships — used by search engines and recommendation systems.
K-Nearest Neighbors (KNN)
Non-parametric classifier predicting labels based on majority class of k closest training examples.
L
Large Language Models (LLMs)
Transformer-based models with billions of parameters trained on vast text corpora — GPT-4, Claude, Gemini.
→ Deep dive article
Long Short-Term Memory (LSTM)
RNN variant using gating mechanisms to learn long-range dependencies in sequential data.
Loss Function
Measures prediction error during training — cross-entropy, MSE, and custom losses guide optimization.
Latent Space
A compressed, lower-dimensional representation of data learned by models like VAEs and diffusion models.
M
Machine Learning (ML)
A subset of AI enabling systems to learn and improve from data without explicit rule programming.
Multi-Task Learning
Training a single model on multiple related tasks simultaneously, improving generalization through shared representations.
Meta-Learning
Learning to learn — models that adapt rapidly to new tasks with minimal data by leveraging prior experience.
Model Compression
Reducing model size via pruning, quantization, or knowledge distillation for efficient deployment.
N
Natural Language Processing (NLP)
AI for understanding, interpreting, and generating human language — from parsing to generation.
Neural Architecture Search (NAS)
Automating the design of neural network architectures using RL or evolutionary methods.
Neural Networks
Layered computational graphs inspired by the brain — the foundation of modern deep learning.
O
Object Detection
Locating and classifying multiple objects within images — YOLO, Faster R-CNN, and DETR are leading models.
Overfitting
When a model learns training data too specifically and fails to generalize to new examples.
Online Learning
Updating model parameters incrementally as new data streams in, without storing all past data.
P
Principal Component Analysis (PCA)
Linear dimensionality reduction projecting data onto orthogonal axes of maximum variance.
Prompt Engineering
Crafting inputs to LLMs to elicit desired outputs — a critical skill for applied generative AI.
Pre-trained Models
Models trained on large datasets that can be fine-tuned for downstream tasks, reducing compute requirements.
Policy Gradient Methods
RL algorithms that optimize policies directly using gradient ascent on expected reward.
Q
Quantum Machine Learning
Applying quantum computing principles to accelerate ML algorithms — an emerging research frontier.
Q-Learning
A model-free RL algorithm learning optimal action-value functions through environment interaction.
Quantization
Reducing model weight precision (FP32 → INT8) for faster inference with minimal accuracy loss.
R
Reinforcement Learning (RL)
Learning optimal behavior through reward signals from environment interaction — foundational to AlphaGo and ChatGPT RLHF.
→ Deep dive article
Random Forests
Ensemble of decision trees trained on random data subsets — robust, interpretable, and widely used.
Recurrent Neural Networks (RNNs)
Sequential models with hidden state memory — largely superseded by Transformers for most tasks.
RAG (Retrieval-Augmented Generation)
Combining LLM generation with external knowledge retrieval to reduce hallucinations and improve factuality.
S
Supervised Learning
Training on labeled input-output pairs to learn a mapping function — the most common ML paradigm.
Support Vector Machines (SVMs)
Finding optimal hyperplanes to separate classes with maximum margin — strong for small, structured datasets.
Self-Supervised Learning
Generating training labels from unlabeled data structure itself — the key to pre-training large models.
Sentiment Analysis
Detecting and classifying emotional tone in text — positive, negative, or nuanced opinion mining.
T
Transformer Models
Self-attention-based architecture that replaced RNNs — the foundation of all modern LLMs and vision models.
→ Deep dive article
Transfer Learning
Adapting a model trained on one task to a different but related task, drastically reducing data needs.
Time Series Analysis
Modeling time-ordered data to identify trends, seasonality, and predict future values.
Text Generation
AI systems producing coherent, contextually appropriate text — from summaries to code to creative writing.
U
Unsupervised Learning
Discovering patterns in unlabeled data — clustering, dimensionality reduction, and generative modeling.
Uncertainty Quantification
Measuring and communicating model confidence — critical in safety-critical AI applications.
V
Variational Autoencoders (VAEs)
Generative models learning a probabilistic latent space — used for image synthesis and anomaly detection.
Vector Embeddings
Dense numerical representations of data in continuous space — the backbone of semantic search and RAG systems.
W
Word Embeddings
Vector representations of words capturing semantic meaning — Word2Vec, GloVe, and contextual embeddings.
Weak Supervision
Training on noisily or incompletely labeled data using heuristics or programmatic labeling functions.
X
XAI (Explainable AI)
Techniques (SHAP, LIME, attention maps) for making AI decisions interpretable and auditable.
XGBoost
Optimized gradient boosting framework — consistently top performer in structured data competitions.
Y
YOLO (You Only Look Once)
Real-time object detection processing entire images in a single forward pass — extremely fast and accurate.
Z
Zero-Shot Learning
Recognizing new classes at inference time without any labeled training examples for those classes.
Z-Score Normalization
Standardizing features by subtracting mean and dividing by standard deviation — ensures equal feature contribution.