.glossary-page { max-width: 900px; margin: 0 auto; font-family: -apple-system, BlinkMacSystemFont, ‚Segoe UI‘, Roboto, sans-serif; }
.glossary-page h1 { color: #1a1a2e; border-bottom: 3px solid #16213e; padding-bottom: 10px; }
.subtitle { color: #666; margin-bottom: 24px; }
.letter-nav { display: flex; flex-wrap: wrap; gap: 4px; margin-bottom: 24px; }
.letter-nav a { display: inline-block; padding: 6px 10px; background: #16213e; color: white; text-decoration: none; border-radius: 4px; font-weight: 600; font-size: 14px; }
.letter-nav a:hover { background: #0f3460; }
.acronym-section { margin-bottom: 32px; }
.acronym-section h2 { color: #16213e; font-size: 1.5em; border-left: 4px solid #16213e; padding-left: 12px; background: #f0f4ff; padding: 8px 12px; }
.acronym-grid { display: grid; grid-template-columns: 200px 1fr; gap: 8px 16px; margin-top: 12px; }
.acronym-code { font-weight: 700; color: #0f3460; font-family: ‚SF Mono‘, ‚Consolas‘, monospace; font-size: 1.05em; }
.acronym-def { color: #333; line-height: 1.6; }
.category-label { grid-column: 1 / -1; color: #666; font-size: 0.85em; text-transform: uppercase; letter-spacing: 1px; margin-top: 8px; border-bottom: 1px solid #e0e4f0; padding-bottom: 4px; }
AI Acronyms & Abbreviations — 150+ Terms
Reviewed: June 4, 2026
Quick reference for every AI/ML acronym you will encounter in research papers, documentation, and technical discussions.
A
AEAutoencoder — A neural network that learns to compress and reconstruct data.
AGIArtificial General Intelligence — AI with human-level cognitive abilities across all domains.
AIArtificial Intelligence — Systems that perform tasks typically requiring human intelligence.
AIGCAI-Generated Content — Any content (text, image, video) created by AI systems.
AMLAnti-Money Laundering (in finance) or Advanced Machine Learning (in research contexts).
APIApplication Programming Interface — A defined interface for software components to communicate.
ASICApplication-Specific Integrated Circuit — Custom chip designed for a particular use case, such as AI inference.
AGIArtificial General Intelligence — AI with human-level cognitive abilities across all domains.
AIArtificial Intelligence — Systems that perform tasks typically requiring human intelligence.
AIGCAI-Generated Content — Any content (text, image, video) created by AI systems.
AMLAnti-Money Laundering (in finance) or Advanced Machine Learning (in research contexts).
APIApplication Programming Interface — A defined interface for software components to communicate.
ASICApplication-Specific Integrated Circuit — Custom chip designed for a particular use case, such as AI inference.
B
BERTBidirectional Encoder Representations from Transformers — Google’s landmark NLP model architecture.
BLEUBilingual Evaluation Understudy — A metric for evaluating machine translation quality.
BNNBayesian Neural Network — A neural network that represents weights as probability distributions.
BLEUBilingual Evaluation Understudy — A metric for evaluating machine translation quality.
BNNBayesian Neural Network — A neural network that represents weights as probability distributions.
C
CAGRCompound Annual Growth Rate — Used in AI market analysis.
CICDContinuous Integration / Continuous Deployment — Automated software delivery pipeline.
CLIPContrastive Language-Image Pre-training — OpenAI’s vision-language model.
CNNConvolutional Neural Network — Architecture optimized for grid-like data (images).
CoTChain of Thought — Prompting technique that elicits step-by-step reasoning.
CUDACompute Unified Device Architecture — NVIDIA’s parallel computing platform.
CICDContinuous Integration / Continuous Deployment — Automated software delivery pipeline.
CLIPContrastive Language-Image Pre-training — OpenAI’s vision-language model.
CNNConvolutional Neural Network — Architecture optimized for grid-like data (images).
CoTChain of Thought — Prompting technique that elicits step-by-step reasoning.
CUDACompute Unified Device Architecture — NVIDIA’s parallel computing platform.
D
DANDo Anything Now — Jailbreak-style prompt pattern (safety concern).
DDPDistributed Data Parallel — Training paradigm splitting data across multiple GPUs.
DLDeep Learning — Machine learning using multi-layer neural networks.
DNNDeep Neural Network — A neural network with multiple hidden layers.
DPODirect Preference Optimization — Alignment technique for training LLMs from preferences.
DQNDeep Q-Network — Combines Q-learning with deep neural networks.
DDPDistributed Data Parallel — Training paradigm splitting data across multiple GPUs.
DLDeep Learning — Machine learning using multi-layer neural networks.
DNNDeep Neural Network — A neural network with multiple hidden layers.
DPODirect Preference Optimization — Alignment technique for training LLMs from preferences.
DQNDeep Q-Network — Combines Q-learning with deep neural networks.
E
ELMoEmbeddings from Language Models — Contextual word representation model.
ELUExponential Linear Unit — Activation function for neural networks.
EMExpectation-Maximization — Iterative algorithm for finding maximum likelihood estimates.
EU AI ActEuropean Union Artificial Intelligence Act — Comprehensive AI regulation framework.
ELUExponential Linear Unit — Activation function for neural networks.
EMExpectation-Maximization — Iterative algorithm for finding maximum likelihood estimates.
EU AI ActEuropean Union Artificial Intelligence Act — Comprehensive AI regulation framework.
F
FIDFrechet Inception Distance — Metric for evaluating quality of generated images.
FLFederated Learning — Training models across decentralized data sources.
FPGAField-Programmable Gate Array — Reconfigurable hardware used for AI acceleration.
FSDPFully Sharded Data Parallel — Memory-efficient distributed training paradigm.
FLFederated Learning — Training models across decentralized data sources.
FPGAField-Programmable Gate Array — Reconfigurable hardware used for AI acceleration.
FSDPFully Sharded Data Parallel — Memory-efficient distributed training paradigm.
G
GANGenerative Adversarial Network — Two-network system (generator + discriminator) for generating realistic data.
GGMLGPT-Generated Machine Learning — Tensor library for CPU-based LLM inference.
GGUFGPT-Generated Unified Format — File format for quantized LLM weights.
GPTGenerative Pre-trained Transformer — OpenAI’s foundational language model series.
GPUGraphics Processing Unit — Parallel processor essential for AI training and inference.
GRPOGroup Relative Policy Optimization — RL technique for LLM alignment (used in DeepSeek R1).
GGMLGPT-Generated Machine Learning — Tensor library for CPU-based LLM inference.
GGUFGPT-Generated Unified Format — File format for quantized LLM weights.
GPTGenerative Pre-trained Transformer — OpenAI’s foundational language model series.
GPUGraphics Processing Unit — Parallel processor essential for AI training and inference.
GRPOGroup Relative Policy Optimization — RL technique for LLM alignment (used in DeepSeek R1).
H
HIPAAHealth Insurance Portability and Accountability Act — US healthcare data protection law.
HPCHigh-Performance Computing — Computing systems with extremely high processing capacity.
HPOHyperparameter Optimization — Process of finding optimal model hyperparameters.
HPCHigh-Performance Computing — Computing systems with extremely high processing capacity.
HPOHyperparameter Optimization — Process of finding optimal model hyperparameters.
I
ICLIn-Context Learning — LLMs learning from examples in the prompt without weight updates.
IoTInternet of Things — Network of connected physical devices generating data for AI.
IPUIntelligence Processing Unit — Graphcore’s custom AI accelerator chip.
IoTInternet of Things — Network of connected physical devices generating data for AI.
IPUIntelligence Processing Unit — Graphcore’s custom AI accelerator chip.
K
KL DivergenceKullback-Leibler Divergence — Measures how one probability distribution differs from another.
KNNK-Nearest Neighbors — Simple classification algorithm based on proximity in feature space.
KNNK-Nearest Neighbors — Simple classification algorithm based on proximity in feature space.
L
LIMALess Is More for Alignment — Paper showing high-quality small datasets can align LLMs.
LLMLarge Language Model — AI model trained on vast text for language understanding and generation.
LLMOpsLLM Operations — Operational practices specific to deploying and managing LLMs in production.
LoRALow-Rank Adaptation — Parameter-efficient fine-tuning method using low-rank matrices.
LSTMLong Short-Term Memory — Recurrent neural network architecture for sequential data.
LLMLarge Language Model — AI model trained on vast text for language understanding and generation.
LLMOpsLLM Operations — Operational practices specific to deploying and managing LLMs in production.
LoRALow-Rank Adaptation — Parameter-efficient fine-tuning method using low-rank matrices.
LSTMLong Short-Term Memory — Recurrent neural network architecture for sequential data.
M
MCPModel Context Protocol — Open standard for connecting AI systems to external tools and data.
MLMachine Learning — AI systems that learn from data rather than following explicit rules.
MLLMMultimodal Large Language Model — LLM that processes multiple input types (text, images, audio).
MLOpsMachine Learning Operations — Practices for deploying and maintaining ML models in production.
MoEMixture of Experts — Architecture routing inputs to different expert sub-networks.
MSEMean Squared Error — Common loss function for regression tasks.
MLMachine Learning — AI systems that learn from data rather than following explicit rules.
MLLMMultimodal Large Language Model — LLM that processes multiple input types (text, images, audio).
MLOpsMachine Learning Operations — Practices for deploying and maintaining ML models in production.
MoEMixture of Experts — Architecture routing inputs to different expert sub-networks.
MSEMean Squared Error — Common loss function for regression tasks.
N
NERNamed Entity Recognition — Identifying entities (people, places, organizations) in text.
NLGNatural Language Generation — AI systems that produce human-readable text.
NLPNatural Language Processing — AI field focused on human language understanding.
NISTNational Institute of Standards and Technology — US body behind the AI Risk Management Framework.
NPUNeural Processing Unit — Processor specialized for neural network inference.
NVDIANVIDIA — Leading GPU manufacturer for AI (included for completeness).
NLGNatural Language Generation — AI systems that produce human-readable text.
NLPNatural Language Processing — AI field focused on human language understanding.
NISTNational Institute of Standards and Technology — US body behind the AI Risk Management Framework.
NPUNeural Processing Unit — Processor specialized for neural network inference.
NVDIANVIDIA — Leading GPU manufacturer for AI (included for completeness).
O
OCROptical Character Recognition — Converting images of text into machine-readable text.
OSSOpen Source Software — Software with publicly available source code.
OSSOpen Source Software — Software with publicly available source code.
P
PEFTParameter-Efficient Fine-Tuning — Methods that adapt large models with minimal parameter updates.
PPOProximal Policy Optimization — Popular reinforcement learning algorithm.
PRPull Request (in software development) or Precision-Recall (in ML evaluation).
PPOProximal Policy Optimization — Popular reinforcement learning algorithm.
PRPull Request (in software development) or Precision-Recall (in ML evaluation).
Q
QLoRAQuantized LoRA — Combines 4-bit quantization with LoRA for efficient fine-tuning.
Q-learningA model-free reinforcement learning algorithm learning action values.
Q-learningA model-free reinforcement learning algorithm learning action values.
R
RAGRetrieval-Augmented Generation — Enhancing LLM outputs with retrieved external information.
ReLURectified Linear Unit — Most common activation function: max(0, x).
RLReinforcement Learning — Learning optimal behavior through environmental rewards.
RLHFReinforcement Learning from Human Feedback — Training LLMs using human preferences as rewards.
RMReward Model — A model trained to predict human preferences for RLHF.
RNNRecurrent Neural Network — Architecture for sequential data with cyclic connections.
ROUGERecall-Oriented Understudy for Gisting Evaluation — Metric for evaluating summaries.
ReLURectified Linear Unit — Most common activation function: max(0, x).
RLReinforcement Learning — Learning optimal behavior through environmental rewards.
RLHFReinforcement Learning from Human Feedback — Training LLMs using human preferences as rewards.
RMReward Model — A model trained to predict human preferences for RLHF.
RNNRecurrent Neural Network — Architecture for sequential data with cyclic connections.
ROUGERecall-Oriented Understudy for Gisting Evaluation — Metric for evaluating summaries.
S
SAMSegment Anything Model — Meta’s zero-shot image segmentation model.
SGDStochastic Gradient Descent — Optimization algorithm for training neural networks.
SFTSupervised Fine-Tuning — Training method using labeled examples to adapt pre-trained models.
SNAPSupplemental Nutrition Assistance Program (in AI fairness contexts).
SQLStructured Query Language — Language for managing relational databases.
SVMSupport Vector Machine — Classification algorithm finding optimal decision boundaries.
SGDStochastic Gradient Descent — Optimization algorithm for training neural networks.
SFTSupervised Fine-Tuning — Training method using labeled examples to adapt pre-trained models.
SNAPSupplemental Nutrition Assistance Program (in AI fairness contexts).
SQLStructured Query Language — Language for managing relational databases.
SVMSupport Vector Machine — Classification algorithm finding optimal decision boundaries.
T
TPUTensor Processing Unit — Google’s custom chip for machine learning workloads.
TTSText-to-Speech — Converting written text into spoken audio.
TTSText-to-Speech — Converting written text into spoken audio.
V
VAEVariational Autoencoder — Generative model that learns a latent representation of data.
ViTVision Transformer — Applying transformer architecture to image recognition tasks.
VRAMVideo RAM — High-bandwidth memory on GPUs for storing model weights and activations.
ViTVision Transformer — Applying transformer architecture to image recognition tasks.
VRAMVideo RAM — High-bandwidth memory on GPUs for storing model weights and activations.
X
XAIExplainable AI — AI systems whose decisions can be understood and interpreted by humans.
Z
ZSLZero-Shot Learning — Classifying inputs from categories never seen during training.