Glossary: Decoding AI Terms
The world of Artificial Intelligence is full of specific terms and jargon. This glossary, based on the section "Some terms to remember" from the book "AI, Beyond the Hype", helps you understand the most important concepts. Use the search bar to quickly find a term.
- Algorithm
- A set of rules or instructions that describe step by step how to perform a task or solve a problem.
- Artificial General Intelligence (AGI)
- A hypothetical form of AI that has the ability to understand, learn, and apply any intellectual task that a human can perform.
- Artificial Intelligence (AI)
- Systems that can perform tasks that normally require human intelligence, such as learning, problem-solving, and decision-making.
- Bias
- A phenomenon where AI systems exhibit systematic biases, often as a result of the data they are trained on, which can lead to unfair or discriminatory outcomes.
- Big Data
- Extremely large and complex datasets that traditional data processing software cannot handle. AI and Machine Learning are crucial for analyzing Big Data.
- Chatbot
- A software application designed to mimic human conversation, often used for customer service or information provision.
- Computer Vision
- A field of AI that enables computers to interpret and understand visual information (images, videos).
- Dataset
- A collection of data used to train and evaluate AI models.
- Deep Learning
- A specialized form of Machine Learning that uses neural networks with multiple layers to analyze complex patterns in large datasets.
- Edge AI
- AI applications that run directly on devices (such as smartphones or IoT devices) instead of in the cloud, resulting in faster processing and better privacy.
- Ethics
- The study and practice of developing technologies that are ethically responsible, taking into account privacy, transparency, and fairness.
- Explainable AI (XAI)
- A research area focused on developing AI systems that are transparent and understandable to humans, so decisions can be explained.
- Federated Learning
- A machine learning approach where models are trained on multiple devices or servers without sharing the data, improving privacy.
- Generative AI (GenAI)
- AI models that can create new content, such as text, images, audio, or video, based on the data they are trained on (e.g., ChatGPT, DALL-E).
- Governance
- The process of managing and regulating technologies to ensure they are used safely, fairly, and responsibly.
- Machine Learning (ML)
- A subset of AI where systems learn from data without being explicitly programmed, to recognize patterns and make predictions.
- MLOps (Machine Learning Operations)
- A set of practices aimed at reliably and efficiently deploying and maintaining machine learning models in production.
- Natural Language Processing (NLP)
- A branch of AI that enables computers to understand, interpret, and generate human language.
- Neural Network
- A computational model inspired by the structure of the human brain, consisting of interconnected nodes (neurons) that process information.
- Prompt Engineering
- The process of designing and optimizing input (prompts) for generative AI models to obtain the desired output.
- Quantum Computing
- A technology that uses the principles of quantum mechanics to perform calculations, potentially enabling revolutionary applications in AI and data processing.
- Reinforcement Learning
- A type of machine learning where an agent learns by interacting with its environment, receiving rewards for desired actions and penalties for undesired actions.
- Robotics
- A multidisciplinary field concerned with the design, construction, and application of robots. AI plays a major role in modern robotics.
- Technological Singularity
- A hypothetical point in the future where technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization, often associated with the rise of AGI.
- Transfer Learning
- A technique where a model trained on one task is reused for a related task, improving efficiency and performance.
- Use Case
- A specific situation or scenario in which an AI system can be applied to solve a particular problem or achieve a goal.
- Vibe Coding
- A new approach in software development that focuses on creating a positive and inspiring work environment, leading to better code and innovation.
- Virtual Assistant
- A more advanced form of chatbot that can perform a wider range of tasks and understands more contextual information (e.g., Siri, Google Assistant).
- A/B Testing
- A method to compare two versions of a model or system to determine which performs better.
- Activation Function
- A function in a neural network that determines whether a neuron is activated.
- Adversarial Example
- An input that is intentionally modified to mislead an AI model.
- Anomaly Detection
- Identifying deviations or unusual patterns in data.
- API (Application Programming Interface)
- A set of rules that allows software applications to communicate with each other.
- AutoML
- Automation of the process of building machine learning models.
- Backpropagation
- An algorithm for propagating errors backward in a neural network and adjusting weights.
- Batch Normalization
- A technique to improve the performance and stability of neural networks by normalizing inputs.
- Bayesian Network
- A probabilistic graphical model that represents relationships between variables.
- Black Box
- A system whose internal workings are not transparent or explainable.
- Boosting
- An ensemble learning method that combines weak models into a strong model.
- Bot
- An automated software program that performs tasks on the internet.
- Clustering
- Grouping similar data without predefined labels.
- Convolutional Neural Network (CNN)
- A type of neural network mainly used for image recognition.
- Cross-validation
- A method to evaluate model performance by testing on different subsets of data.
- Data Augmentation
- Artificially increasing a dataset by creating variations.
- Data Lake
- A central repository for structured and unstructured data.
- Data Mining
- Discovering patterns and knowledge from large amounts of data.
- Data Pipeline
- A series of processes for collecting, processing, and storing data.
- Decision Tree
- A model that makes decisions based on a tree structure.
- Dimensionality Reduction
- Reducing the number of variables in a dataset.
- Dropout
- A regularization technique where neurons are randomly deactivated during training.
- Early Stopping
- Stopping the training of a model before it overfits.
- Embedding
- A representation of data in a lower-dimensional space.
- Ensemble Learning
- Combining multiple models to improve performance.
- Epoch
- One complete pass of the training data through a model.
- Feature Engineering
- Creating new input variables based on existing data.
- Feature Extraction
- Automatically identifying relevant features from data.
- Fuzzy Logic
- A logic system that works with degrees of true/false instead of binary values.
- GAN (Generative Adversarial Network)
- A neural network consisting of a generator and a discriminator competing against each other.
- Gradient Descent
- An optimization algorithm to minimize a models error.
- Grid Search
- A method to find the best hyperparameters for a model.
- Heuristic
- A practical approach to problem-solving, often based on experience.
- Hyperparameter
- A parameter whose value is set before the learning process of a model.
- Inference
- Applying a trained model to new data to make predictions.
- K-means
- A popular clustering algorithm.
- K-nearest neighbors (KNN)
- An algorithm that makes predictions based on the closest examples in the data.
- Label
- The target value or category assigned to a data point.
- Latent Space
- An abstract space in which data is represented by a model.
- Leaky ReLU
- A variant of the ReLU activation function that allows small negative values.
- Learning Rate
- The speed at which a model adjusts its parameters during training.
- Logistic Regression
- A statistical model used for binary classification problems.
- Loss Function
- A function that measures the difference between a models predictions and the actual values.
- LSTM (Long Short-Term Memory)
- A type of recurrent neural network suitable for sequential data.
- Markov Chain
- A stochastic model that describes the probability of a sequence of events.
- Meta-Learning
- Learning how to learn; a model that learns from other learning algorithms.
- Mini-batch
- A small subset of the training data used for one update of the model.
- Model Drift
- The phenomenon where a models performance degrades over time due to changing data.
- Natural Language Generation (NLG)
- Automatically generating text by an AI system.
- Normalization
- Scaling data to a standard range.
- Object Detection
- Identifying and locating objects in images or video.
- One-hot Encoding
- A technique to convert categorical data into binary vectors.
- Optimizer
- An algorithm that adjusts a models parameters to improve performance.
- Outlier
- A data point that significantly deviates from other observations.
- Overfitting
- When a model performs too well on training data but poorly on new data.
- PCA (Principal Component Analysis)
- A technique for dimensionality reduction.
- Perceptron
- The simplest type of neural network, consisting of a single layer.
- Precision
- The percentage of relevant results among all retrieved results.
- Preprocessing
- Preparing and cleaning data before it is used for modeling.
- Random Forest
- An ensemble learning method that uses multiple decision trees.
- Recall
- The percentage of relevant results that were actually retrieved.
- Regression
- A type of model that predicts continuous values.
- Regularization
- Techniques to prevent overfitting by limiting the complexity of a model.
- ReLU (Rectified Linear Unit)
- A commonly used activation function in neural networks.
- ResNet
- A type of neural network with so-called residual connections.
- ROC Curve
- A graph that visualizes the performance of a classification model.
- Scaler
- A method to scale data to a certain range.
- Self-supervised Learning
- A learning method where the model generates labels from the data itself.
- Semi-supervised Learning
- A learning method that uses both labeled and unlabeled data.
- Shapley Value
- A game theory method to measure the contribution of each feature to a prediction.
- Sigmoid
- An activation function that produces an S-shaped curve.
- SMOTE
- A technique to balance datasets with skewed class distributions.
- Softmax
- A function that converts a vector into probabilities.
- Stochastic Gradient Descent (SGD)
- A variant of gradient descent where each update is made based on a random subset of the data.
- Supervised Learning
- A learning method where the model learns from labeled data.
- Support Vector Machine (SVM)
- A powerful algorithm for classification and regression tasks.
- Swarm Intelligence
- The collective behavior of decentralized, self-organizing systems.
- Tensor
- A data structure used in deep learning, similar to a matrix.
- Tokenization
- Splitting text into smaller units (tokens) for processing by a model.
- Transfer Learning
- Reusing a trained model for a new, related task.
- Turing Test
- A test to determine whether a machine exhibits intelligent behavior indistinguishable from that of a human.
- Underfitting
- When a model fails to learn sufficiently from the data and performs poorly.
- Unsupervised Learning
- A learning method where the model discovers patterns in unlabeled data.
- Validation Set
- A dataset used to evaluate model performance during training.
- Variance
- The extent to which a models predictions vary with different training sets.
- Weight
- A parameter in a neural network that determines how important an input is.
- Word Embedding
- A technique to represent words as vectors in a continuous space.
- Zero-shot Learning
- A learning method where a model performs tasks for which it was not explicitly trained.