Below is a glossary that includes 100 common “AI (Artificial Intelligence)” terms and phrases in alphabetical order. Enjoy!
| Term | Definition & Example |
| Accuracy | Percentage of correct predictions. Example: 92% accuracy. |
| Agent | AI entity performing tasks autonomously. Example: Task-planning agent. |
| AI Alignment | Ensuring AI goals match human values. Example: Safe AI systems. |
| AI Bias | Systematic unfairness in AI outcomes. Example: Biased hiring models. |
| Algorithm | A set of rules used to train models. Example: Decision tree algorithm. |
| Artificial General Intelligence (AGI) | Hypothetical AI with human-level intelligence. Example: Broad reasoning across tasks. |
| Artificial Intelligence (AI) | Systems that perform tasks requiring human-like intelligence. Example: Chatbots answering questions. |
| Artificial Neural Network (ANN) | A network of interconnected artificial neurons. Example: Credit scoring models. |
| Attention Mechanism | Focuses model on relevant input parts. Example: Language translation. |
| AUC | Area under ROC curve. Example: Model comparison. |
| AutoML | Automated model selection and tuning. Example: Auto-generated models. |
| Autonomous System | AI operating with minimal human input. Example: Self-driving cars. |
| Backpropagation | Method to update neural network weights. Example: Deep learning training. |
| Batch | Subset of data processed at once. Example: Batch size of 32. |
| Batch Inference | Predictions made in bulk. Example: Nightly scoring jobs. |
| Bias (Model Bias) | Error from oversimplified assumptions. Example: Linear model on non-linear data. |
| Bias–Variance Tradeoff | Balance between bias and variance. Example: Choosing model complexity. |
| Black Box Model | Model with opaque internal logic. Example: Deep neural networks. |
| Classification | Predicting categorical outcomes. Example: Email spam classification. |
| Clustering | Grouping similar data points. Example: Customer segmentation. |
| Computer Vision | AI for interpreting images and video. Example: Facial recognition. |
| Concept Drift | Changes in underlying relationships. Example: Fraud patterns evolving. |
| Confusion Matrix | Table evaluating classification results. Example: True positives vs false positives. |
| Data Augmentation | Expanding data via transformations. Example: Image rotation. |
| Data Drift | Changes in input data distribution. Example: New user demographics. |
| Data Leakage | Using future information in training. Example: Including test labels. |
| Decision Tree | Tree-based decision model. Example: Loan approval logic. |
| Deep Learning | ML using multi-layer neural networks. Example: Image recognition. |
| Dimensionality Reduction | Reducing number of features. Example: PCA for visualization. |
| Edge AI | AI running on local devices. Example: Smart cameras. |
| Embedding | Numerical representation of data. Example: Word embeddings. |
| Ensemble Model | Combining multiple models. Example: Random forest. |
| Epoch | One full pass through training data. Example: 50 training epochs. |
| Ethics in AI | Moral considerations in AI use. Example: Avoiding bias. |
| Explainable AI (XAI) | Making AI decisions understandable. Example: Feature importance charts. |
| F1 Score | Balance of precision and recall. Example: Imbalanced datasets. |
| Fairness | Equitable AI outcomes across groups. Example: Equal approval rates. |
| Feature | An input variable for a model. Example: Customer age. |
| Feature Engineering | Creating or transforming features to improve models. Example: Calculating customer tenure. |
| Federated Learning | Training models across decentralized data. Example: Mobile keyboard predictions. |
| Few-Shot Learning | Learning from few examples. Example: Custom classification with few samples. |
| Fine-Tuning | Further training a pre-trained model. Example: Custom chatbot training. |
| Generalization | Model’s ability to perform on new data. Example: Accurate predictions on unseen data. |
| Generative AI | AI that creates new content. Example: Text or image generation. |
| Gradient Boosting | Sequentially improving weak models. Example: XGBoost. |
| Gradient Descent | Optimization technique adjusting weights iteratively. Example: Training neural networks. |
| Hallucination | Model generates incorrect information. Example: False factual claims. |
| Hyperparameter | Configuration set before training. Example: Learning rate. |
| Inference | Using a trained model to predict. Example: Real-time recommendations. |
| K-Means | Clustering algorithm. Example: Market segmentation. |
| Knowledge Graph | Graph-based representation of knowledge. Example: Search engines. |
| Label | The correct output for supervised learning. Example: “Fraud” or “Not Fraud”. |
| Large Language Model (LLM) | AI trained on massive text corpora. Example: ChatGPT. |
| Loss Function | Measures model error during training. Example: Mean squared error. |
| Machine Learning (ML) | AI that learns patterns from data without explicit programming. Example: Spam email detection. |
| MLOps | Practices for managing ML lifecycle. Example: CI/CD for models. |
| Model | A trained mathematical representation of patterns. Example: Logistic regression model. |
| Model Deployment | Making a model available for use. Example: API-based predictions. |
| Model Drift | Model performance degradation over time. Example: Changing customer behavior. |
| Model Interpretability | Ability to understand model behavior. Example: Decision tree visualization. |
| Model Versioning | Tracking model changes. Example: v1 vs v2 models. |
| Monitoring | Tracking model performance in production. Example: Accuracy alerts. |
| Multimodal AI | AI handling multiple data types. Example: Text + image models. |
| Naive Bayes | Probabilistic classification algorithm. Example: Spam filtering. |
| Natural Language Processing (NLP) | AI for understanding human language. Example: Sentiment analysis. |
| Neural Network | Model inspired by the human brain’s structure. Example: Handwritten digit recognition. |
| Optimization | Process of minimizing loss. Example: Gradient descent. |
| Overfitting | Model learns noise instead of patterns. Example: Perfect training accuracy, poor test accuracy. |
| Pipeline | Automated ML workflow. Example: Training-to-deployment flow. |
| Precision | Correct positive predictions rate. Example: Fraud detection precision. |
| Pretrained Model | Model trained on general data. Example: GPT models. |
| Principal Component Analysis (PCA) | Technique for dimensionality reduction. Example: Compressing high-dimensional data. |
| Privacy | Protecting personal data. Example: Anonymizing training data. |
| Prompt | Input instruction for generative models. Example: “Summarize this text.” |
| Prompt Engineering | Crafting effective prompts. Example: Improving LLM responses. |
| Random Forest | Ensemble of decision trees. Example: Classification tasks. |
| Real-Time Inference | Immediate predictions on live data. Example: Fraud detection. |
| Recall | Ability to find all positives. Example: Cancer detection. |
| Regression | Predicting numeric values. Example: Sales forecasting. |
| Reinforcement Learning | Learning through rewards and penalties. Example: Game-playing AI. |
| Reproducibility | Ability to recreate results. Example: Fixed random seeds. |
| Robotics | AI applied to physical machines. Example: Warehouse robots. |
| ROC Curve | Performance visualization for classifiers. Example: Threshold analysis. |
| Semi-Supervised Learning | Mix of labeled and unlabeled data. Example: Image classification with limited labels. |
| Speech Recognition | Converting speech to text. Example: Voice assistants. |
| Supervised Learning | Learning using labeled data. Example: Predicting house prices from known values. |
| Support Vector Machine (SVM) | Algorithm separating data with margins. Example: Text classification. |
| Synthetic Data | Artificially generated data. Example: Privacy-safe training. |
| Test Data | Data used to evaluate model performance. Example: Held-out validation dataset. |
| Threshold | Cutoff for classification decisions. Example: Probability > 0.7. |
| Token | Smallest unit of text processed by models. Example: Words or subwords. |
| Training Data | Data used to teach a model. Example: Historical sales records. |
| Transfer Learning | Reusing knowledge from another task. Example: Image model reused for medical scans. |
| Transformer | Neural architecture for sequence data. Example: Language translation models. |
| Underfitting | Model too simple to capture patterns. Example: High error on all datasets. |
| Unsupervised Learning | Learning from unlabeled data. Example: Customer clustering. |
| Validation Data | Data used to tune model parameters. Example: Hyperparameter selection. |
| Variance | Error from sensitivity to data fluctuations. Example: Highly complex model. |
| XGBoost | Optimized gradient boosting algorithm. Example: Kaggle competitions. |
| Zero-Shot Learning | Performing tasks without examples. Example: Classifying unseen labels. |
Please share your suggestions for any terms that should be added.
The Data Community 