“50 Essential AI Terms You Need to Know: A Comprehensive 2025 Glossary”
Introduction
Artificial Intelligence (AI) is evolving at lightning speed, and its own vocabulary can feel like a moving target. Whether you’re a developer, a business leader, or simply curious, mastering the core terminology will help you navigate this rapidly changing field. Below is a curated list of 50 key AI terms—each with a concise definition—to keep you up to speed in 2025.
-
Artificial Intelligence (AI)
The simulation of human intelligence processes by machines, especially computer systems. -
Machine Learning (ML)
A subset of AI focused on algorithms that improve automatically through experience and data. -
Deep Learning
A family of ML techniques based on artificial neural networks with many layers (“deep” networks). -
Neural Network
A computational model inspired by the human brain’s network of neurons, used in deep learning. -
Supervised Learning
ML approach where models are trained on labeled data (inputs paired with correct outputs). -
Unsupervised Learning
ML approach where models infer structure from unlabeled data (no preassigned categories). -
Reinforcement Learning (RL)
A learning paradigm where an agent learns to make decisions by receiving rewards or penalties. -
Natural Language Processing (NLP)
Techniques that enable computers to understand, interpret, and generate human language. -
Computer Vision
AI methods for interpreting and making decisions based on visual inputs (images, video). -
Generative AI
Models that can generate new content—text, images, audio—based on learned patterns. -
Transformer
A neural network architecture particularly effective for sequence modeling and NLP tasks (e.g., GPT). -
Attention Mechanism
A component of models (notably Transformers) that weights the importance of different input elements. -
Large Language Model (LLM)
A massive-scale NLP model (e.g., GPT-4, LLaMA) trained on extensive text corpora to perform varied language tasks. -
Tokenization
The process of breaking text into discrete units (“tokens”) such as words or subwords for model input. -
Embedding
A dense numeric representation of discrete data (words, images) in a continuous vector space. -
Overfitting
When a model learns training data too well—including noise—resulting in poor generalization to new data. -
Underfitting
When a model is too simple to capture underlying patterns, yielding high errors on both training and test data. -
Hyperparameters
Settings (e.g., learning rate, network depth) configured before training that govern model behavior. -
Gradient Descent
An optimization algorithm used to minimize a model’s loss function by iteratively adjusting parameters. -
Loss Function
A mathematical measure of how far a model’s predictions are from the actual target values. -
Backpropagation
The method of computing gradients of the loss function with respect to model parameters for update. -
Epoch
One full pass through the entire training dataset during the model training process. -
Batch Size
Number of training samples processed before the model’s internal parameters are updated. -
Transfer Learning
Reusing a pretrained model on a new, related task to reduce training time and data requirements. -
Fine-Tuning
The process of taking a pretrained model and training it further on a specific downstream dataset. -
Zero-Shot Learning
The ability of a model to perform a task it was never explicitly trained on, based on prompts or descriptions. -
One-Shot & Few-Shot Learning
Learning tasks where the model is given only one (or a few) labeled examples to adapt to new tasks. -
Prompt Engineering
Crafting inputs (prompts) to guide a generative model toward desired outputs. -
Recurrent Neural Network (RNN)
A class of neural networks designed for sequence data, with feedback loops to maintain state. -
Long Short-Term Memory (LSTM)
A type of RNN designed to overcome long-range dependency issues via gated memory units. -
Convolutional Neural Network (CNN)
Neural network architecture optimized for processing grid-like data (e.g., images) using convolutional filters. -
Generative Adversarial Network (GAN)
A framework where two neural networks (generator and discriminator) compete, yielding highly realistic synthetic data. -
Autoencoder
A neural network trained to compress and then reconstruct its input, useful for dimensionality reduction. -
Bayesian Learning
A probabilistic approach that updates beliefs (model parameters) as more evidence (data) becomes available. -
Markov Decision Process (MDP)
A mathematical framework for modeling decision-making in RL, defined by states, actions, rewards, and transitions. -
Q-Learning
A model-free RL algorithm that learns the value of actions in each state to maximize cumulative reward. -
Policy Gradient
An RL method that directly optimizes the policy (action-selection strategy) via gradient ascent on expected reward. -
Edge AI
Running AI inference locally on devices (“the edge”) instead of centralized servers to reduce latency and privacy risk. -
Federated Learning
A decentralized ML approach where models are trained across multiple devices while keeping data localized. -
Explainable AI (XAI)
Techniques that make AI decision-making transparent and interpretable to humans. -
Model Drift
The phenomenon where a model’s performance degrades over time due to changing data distributions. -
Ethical AI
The study and practice of developing AI systems that adhere to fairness, accountability, transparency, and safety principles. -
Bias & Fairness
Efforts to identify and mitigate systematic errors that disadvantage particular groups in AI predictions. -
Data Augmentation
Generating modified versions of existing data (e.g., rotated images) to increase training diversity. -
Synthetic Data
Artificially generated data that mimics real-world data, used to augment or replace scarce datasets. -
Computer Ethics
The field that examines moral and societal implications of AI and computing technologies. -
AI Governance
Frameworks and policies guiding the responsible development, deployment, and monitoring of AI systems. -
MLOps
Practices and tools for automating and scaling the end-to-end ML lifecycle (development, deployment, monitoring). -
AI as a Service (AIaaS)
Cloud-based platforms offering ready-to-use AI tools (e.g., AWS SageMaker, Azure Cognitive Services). -
Quantum Machine Learning
Emerging techniques that leverage quantum computing to potentially accelerate ML algorithms.
Save this glossary for quick reference, share it with your team, and bookmark it to stay on top of AI’s fast-moving landscape!
댓글 없음:
댓글 쓰기