A/B Testing (Models)
- Definition
- A/B testing of models is a controlled experiment that splits live traffic between two model versions and compares their business outcomes under identical conditions before one version is adopted for everyone.
Showing 20 of 207 terms
Accuracy
- Definition
- Accuracy is the share of a classifier's predictions that were correct, counting all classes together, which makes it easy to understand and misleading whenever one class is rare.
Activation Function
- Definition
- An activation function is the nonlinear operation applied to each neuron's output in a neural network, which lets the network learn patterns that a straight line cannot describe.
Active Learning
- Definition
- Active learning is a training approach in which the model selects the unlabeled examples it is least sure about and asks a human to label those first.
Adaptable AI
- Definition
- Adaptable AI is an AI system that adjusts its behavior, strategy, or underlying models in response to changing conditions without being reprogrammed, using feedback from its own results. It depends on monitoring,…
Adversarial Example
- Definition
- An adversarial example is an input that has been deliberately altered, often in ways a person cannot notice, so that a machine learning model gets it wrong, frequently with high confidence.
Adversarial Machine Learning
- Definition
- Adversarial machine learning is the study of attacks that manipulate machine learning systems through their training data, inputs, or outputs, and of the defenses that make those systems harder to fool, corrupt,…
AI Model
- Definition
- An AI model is a mathematical function, learned from data, that turns inputs into predictions, classifications, or generated content. Its learned parameters are the stored result of training and the part that…
Algorithm
- Definition
- An algorithm is a precise, finite sequence of steps that turns an input into an output. In AI, algorithms both train models from data and use trained models to make predictions or…
Anomaly Detection
- Definition
- Anomaly detection is the use of statistics and machine learning to find records, events, or readings that differ markedly from normal patterns, such as a fraudulent payment or a machine that is…
Artificial Intelligence (AI)
- Definition
- Artificial intelligence (AI) is software that performs tasks which normally require human judgment, such as recognizing patterns, understanding language, and choosing actions. In business it ranges from fraud scoring to agents working…
Attention Mechanism
- Definition
- An attention mechanism lets a neural network decide how much weight to give each part of its input when producing each part of its output, and it is the central operation inside…
AUC (Area Under the Curve)
- Definition
- AUC, the area under the ROC curve, is a number from 0 to 1 that measures how well a classifier ranks positive cases above negative ones across all thresholds, where 0.5 means…
Autoencoder
- Definition
- An autoencoder is a neural network trained to compress its input into a compact representation and then reconstruct it, which makes it useful for anomaly detection, denoising, and dimensionality reduction.
Automated Machine Learning (AutoML)
- Definition
- Automated machine learning (AutoML) is software that automates the steps of building a model, including data preparation, algorithm selection, hyperparameter tuning, and evaluation.
Backpropagation
- Definition
- Backpropagation is the algorithm that computes how much each weight in a neural network contributed to the error, so gradient descent knows how to adjust it.
Baseline Model
- Definition
- A baseline model is a simple reference model, such as predicting the average or the most common outcome, used to judge whether a more complex model adds real value.
Batch
- Definition
- A batch is a group of data examples processed together in one step, either during training, where it drives a single parameter update, or during inference, where it improves throughput.
Batch Inference
- Definition
- Batch inference is the practice of running a model over a large set of records at once on a schedule, such as scoring every account overnight, instead of answering requests one at…
Batch Size
- Definition
- Batch size is the number of training examples a model processes before updating its parameters once, a hyperparameter that trades off memory use, training speed, and model quality.
Bayesian Network
- Definition
- A Bayesian network is a probabilistic model that represents variables as nodes in a directed graph and encodes how each depends on its causes, so evidence about some variables updates beliefs about…
Benchmark
- Definition
- A benchmark is a standardized set of tasks and a scoring method used to compare AI models against each other, giving a repeatable measure of capability on a defined problem.
BERT
- Definition
- BERT (Bidirectional Encoder Representations from Transformers) is a language model from Google that reads text in both directions at once and became the standard base for text classification, entity extraction, and search…
Bias (AI)
- Definition
- Bias in AI is a systematic error in a model's outputs. It can be statistical, a consistent gap between predictions and reality, or social, a pattern that treats groups of people unequally.
Bias Detection and Fairness Auditing
- Definition
- Bias detection and fairness auditing are the methods used to find out whether a model treats groups of people differently and to document whether those differences are justified.
Bias-Variance Tradeoff
- Definition
- The bias-variance tradeoff is the balance between a model that is too simple to capture the pattern and one so flexible that it chases noise in the training data.
Binary Classification
- Definition
- Binary classification is the machine learning task of sorting each input into one of exactly two classes, such as fraud or legitimate, and it is judged by how it balances the two…
Black Box Model
- Definition
- A black box model is a model whose inputs and outputs are visible but whose internal reasoning a person cannot follow, because of its complexity or because its vendor keeps the details…
Boosting
- Definition
- Boosting is an ensemble technique that trains a sequence of simple models, each concentrating on the examples its predecessors got wrong, and combines them into one accurate model.
Brute Force Search
- Definition
- Brute force search is a problem-solving method that tries every possible candidate solution and checks each one, guaranteeing the best answer at a cost in time that grows explosively with the size…
Calibration
- Definition
- Calibration is the degree to which a model's confidence scores match real outcomes, so that predictions made with a score of 0.8 turn out to be correct about eight times in ten.
Champion-Challenger Testing
- Definition
- Champion-challenger testing is a continuous method for improving a production model by running candidate models alongside the current one, comparing their results on the same cases, and promoting a challenger that proves…
Checkpoint
- Definition
- A checkpoint is a saved snapshot of a model's weights, and often its optimizer state, taken at a point during training so the run can be resumed, compared, or rolled back.
Churn Prediction
- Definition
- Churn prediction is the use of a model to estimate which customers are likely to leave, or sharply cut their business, within a set period, so a team can act before they…
Class (Machine Learning)
- Definition
- A class is one of the discrete categories a classification model assigns to each input, such as approved or declined for a loan application, or invoice, contract, and other for a document…
Class Balance
- Definition
- Class balance describes how evenly the categories a model predicts are represented in the training data; class imbalance, where one outcome is rare, distorts both training and accuracy.
Classic Machine Learning
- Definition
- Classic machine learning is the family of established statistical algorithms, such as linear and logistic regression, decision trees, support vector machines, and naive Bayes, that remain the best choice for most tabular…
Classification
- Definition
- Classification is the supervised learning task of assigning each input to one of a set of predefined categories, such as approve or decline, spam or legitimate, or one of several ticket queues.
Classifier
- Definition
- A classifier is a trained machine learning model that assigns an input to one of a set of categories, usually by producing a score for each category and selecting the highest.
Classifier Performance
- Definition
- Classifier performance is the set of measures that describe how well a classification model separates its classes, including accuracy, precision, recall, F1 score, and the area under the ROC curve.
Clustering
- Definition
- Clustering is an unsupervised machine learning technique that sorts records into groups so that members of each group are more similar to one another than to members of other groups.
Coefficient of Determination (R-Squared)
- Definition
- The coefficient of determination, or R-squared, measures how much of the variation in an outcome a regression model explains, where 1 means a perfect fit and 0 means no better than predicting…
Cohen’s Kappa
- Definition
- Cohen's kappa is a statistic that measures how much two raters agree on categorical labels beyond the agreement chance alone would produce. It is widely used to check the consistency of training…
Concept Drift
- Definition
- Concept drift is a change in the relationship between a model's inputs and the outcome it predicts, so that the same inputs now lead to a different result than they did in…
Confidence Score
- Definition
- A confidence score is a number a model attaches to each prediction to indicate how sure it is, usually on a scale from zero to one, so that uncertain outputs can be…
Confidence Threshold
- Definition
- A confidence threshold is the cutoff score above which a model's prediction is accepted automatically and below which it is rejected or sent to a person for review.
Confusion Matrix
- Definition
- A confusion matrix is a table that counts a classifier's correct and incorrect predictions for each class, showing which classes the model confuses with which.
Continual Learning
- Definition
- Continual learning is the ability of a model to keep learning from new data over time while retaining what it already knows, instead of being retrained from scratch or frozen after one…
Convolutional Filter
- Definition
- A convolutional filter is a small grid of weights that slides across an image to detect a local pattern such as an edge or texture. Stacks of learned filters form a convolutional…
Convolutional Neural Network (CNN)
- Definition
- A convolutional neural network is a deep learning model that scans images with small learned filters to detect edges, textures, and shapes, making it the standard architecture for visual recognition tasks.
Correlation
- Definition
- Correlation is a statistical measure of how closely two variables move together, usually expressed as a coefficient between minus one and plus one. It shows association but does not show that one…
Counterfactual Explanation
- Definition
- A counterfactual explanation states the smallest change to a model's inputs that would have produced a different outcome, for example what would have turned a declined application into an approved one.
Cross-Validation
- Definition
- Cross-validation is a method for estimating how well a model will perform on new data by training and testing it several times on different splits of the same dataset. It is most…
Data Annotation
- Definition
- Data annotation is the work of marking raw data with the labels a model should learn, such as drawing boxes around objects in images or tagging clauses in contracts.
Data Drift
- Definition
- Data drift is a change in the makeup of the data a model receives in production, such as shifted averages or new categories, compared with the data it was trained on.
Data Labels
- Definition
- Data labels are the known answers attached to training examples, such as "fraud" on a transaction or "defect" on a photo, that a supervised learning model learns to reproduce.
Data Mining
- Definition
- Data mining is the process of discovering patterns, relationships, segments, and anomalies in large existing datasets using statistics and machine learning, usually to explain behavior rather than to serve live predictions.
Data Science
- Definition
- Data science is the discipline of drawing knowledge, explanations, and predictions from data by combining statistics, programming, machine learning, and an understanding of the business domain.
Data Visualization
- Definition
- Data visualization is the presentation of data as charts, maps, and other graphics so that patterns, trends, and outliers can be seen at a glance and acted on by people who do…
Data-Driven AI
- Definition
- Data-driven AI learns its behavior from patterns in historical data rather than from rules written by experts. Machine learning and deep learning are data-driven; its counterpart is knowledge-driven AI built on explicit…
Dataset
- Definition
- A dataset is a defined collection of records assembled for a purpose, such as training or testing a machine learning model, with a known source, scope, and structure.
Decision Tree
- Definition
- A decision tree is a model that predicts an outcome by asking a sequence of yes-or-no questions about a record, following the answers down branches until it reaches a leaf that holds…
Deep Learning
- Definition
- Deep learning is machine learning with neural networks that have many layers, which learn their own features from raw data such as images, audio, and text instead of relying on hand-built inputs.
Deep Learning Framework
- Definition
- A deep learning framework is a software library that supplies the building blocks for defining, training, and running neural networks, handling the underlying math and GPU acceleration so developers describe the model…
Deep Neural Network (DNN)
- Definition
- A deep neural network is a neural network with several hidden layers between input and output, which allows it to learn layered representations of complex data such as images, audio, and text.
Demand Forecasting
- Definition
- Demand forecasting is the estimation of how much of a product or service customers will want in a future period, by item, location, and time, so that purchasing, production, and staffing can…
Deterministic Algorithm
- Definition
- A deterministic algorithm is a procedure that always produces the same output from the same input by following the same steps. Tax calculations, pricing rules, and most traditional business software work this…
Differential Privacy
- Definition
- Differential privacy is a mathematical guarantee that a statistic or model changes almost imperceptibly whether or not any one person's record is included, which bounds what anyone can learn about that person…
Dimensionality Reduction
- Definition
- Dimensionality reduction is a set of techniques that cut the number of variables describing each record while keeping most of the useful information, which speeds up models, filters noise, and makes data…
Discriminative Model
- Definition
- A discriminative model is a machine learning model that learns the boundary between classes or the direct mapping from inputs to a label, without modeling how the input data itself is generated.
Domain Adaptation
- Definition
- Domain adaptation is the set of techniques for making a model trained on data from one setting perform well on data from a different but related setting.
Embeddings
- Definition
- Embeddings are numeric vectors that represent the meaning of text, images, or other data so that items with similar meaning sit close together and can be compared, searched, and clustered by a…
Ensemble Learning
- Definition
- Ensemble learning combines the predictions of several machine learning models into one answer, so that their separate errors partly cancel and the result is more accurate and more stable than any single…
Epoch
- Definition
- An epoch is one complete pass through the entire training dataset during model training, and the number of epochs is a hyperparameter that controls how long training runs.
Experiment Tracking
- Definition
- Experiment tracking is the practice of recording every model training run, including its code version, data, hyperparameters, metrics, and outputs, so results can be compared and reproduced later.
Explainable AI (XAI)
- Definition
- Explainable AI (XAI) is the set of methods that make an AI model's outputs understandable to people, by showing which inputs drove a result and how the result would change if they…
F1 Score
- Definition
- The F1 score is a single measure of classifier quality that combines precision and recall into their harmonic mean, ranging from zero to one, with one meaning every prediction and every real…
False Negatives
- Definition
- False negatives are cases a model predicts as negative that are actually positive, such as a failing bearing the maintenance model marked healthy. Statisticians call them Type II errors.
False Positive Rate
- Definition
- The false positive rate is the share of actual negative cases that a model wrongly flags as positive, calculated as false positives divided by false positives plus true negatives. It is the…
False Positives
- Definition
- False positives are cases a model predicts as positive that are actually negative, such as a legitimate card purchase declined as fraud. Statisticians call them Type I errors.
Feature Engineering
- Definition
- Feature engineering is the work of turning raw data into input variables that help a machine learning model predict well, using domain knowledge to create, transform, and select the columns the model…
Feature Extraction
- Definition
- Feature extraction is the conversion of raw data such as images, audio, or text into a compact set of numeric values that capture its useful content, so that a model can learn…
Feature Importance
- Definition
- Feature importance is a measure of how much each input variable contributes to a model's predictions. Teams use it to check a model's logic, catch information leakage, and explain the model to…
Feature Store
- Definition
- A feature store is a data system that defines, stores, and serves the input variables machine learning models use, so the same feature definitions feed both model training and live predictions.
Features
- Definition
- Features are the input variables a machine learning model learns from, such as an account's age or an invoice's amount. Their quality usually matters more to accuracy than the choice of algorithm.
Federated Learning
- Definition
- Federated learning trains a shared machine learning model across several data holders without moving their raw data to a central location; only model updates travel between the participants and a coordinator.
Feedback Loop
- Definition
- A feedback loop is the path by which an AI system's results, such as user corrections or real-world outcomes, return as data that shapes its next version, whether by design or by…
Few-Shot Learning
- Definition
- Few-shot learning is a model's ability to learn a new task or category from only a handful of examples, with one-shot learning as the special case of a single example.
Fine-Tuning
- Definition
- Fine-tuning is the process of continuing to train a pre-trained model on a smaller, task-specific dataset so that it adapts to a company's language, formats, and decisions.
Fraud Detection
- Definition
- Fraud detection is the use of rules, statistics, and machine learning to spot transactions, claims, or account activity that are likely deceptive, so they can be blocked or reviewed before money is…
Gaussian Mixture Model (GMM)
- Definition
- A Gaussian mixture model (GMM) describes data as a blend of several bell-shaped distributions, one per group, and gives each record a probability of belonging to each group.
Generalization
- Definition
- Generalization is a model's ability to make accurate predictions on data it did not see during training, which is the only ability that matters once the model is deployed.
Generalized Linear Models (GLM)
- Definition
- Generalized linear models extend linear regression to outcomes that are not normally distributed, such as counts, yes-or-no events, and skewed amounts, by linking a weighted sum of inputs to the outcome through…
Generative Adversarial Network (GAN)
- Definition
- A generative adversarial network is a pair of neural networks trained against each other, one producing synthetic data and the other judging whether it is real, until the output becomes convincing.
Generative Model
- Definition
- A generative model is a machine learning model that learns the distribution of its training data well enough to produce new examples that resemble it, such as text, images, audio, or tabular…
Gradient Descent
- Definition
- Gradient descent is the optimization algorithm that trains most machine learning models by repeatedly adjusting parameters in the direction that reduces the loss function.
Gradient-Boosted Decision Trees (GBDT)
- Definition
- Gradient-boosted decision trees (GBDT) are an ensemble method that adds small decision trees one at a time, each fitted to the errors left by those before it, and are usually the most…
Graph Neural Network (GNN)
- Definition
- A graph neural network is a neural network that operates on data arranged as nodes and edges, passing information along connections so that each node's representation reflects its neighbors.
Grid Search
- Definition
- Grid search is a hyperparameter tuning method that trains and evaluates a model on every combination of values from a predefined list for each setting.
Ground Truth
- Definition
- Ground truth is the verified correct answer for a data example, used to train a model and to measure how often its predictions are right.
Holdout Data
- Definition
- Holdout data is a portion of a dataset deliberately kept out of model training, so the model can be evaluated on examples it has never seen before it goes into use.
Hyperparameters
- Definition
- Hyperparameters are the settings chosen before training begins, such as learning rate, batch size, and tree depth, that control how a model learns rather than what it learns.
ImageNet
- Definition
- ImageNet is a large labeled image dataset, with over a million photographs sorted into a thousand categories in its benchmark subset, on which modern deep learning for computer vision first proved itself.
Inference
- Definition
- Inference is the stage in which a trained model is used to produce predictions or generated output for new inputs, as distinct from training, which is where the model learns.
Information Leakage
- Definition
- Information leakage, also called data leakage, occurs when training data contains information that will not be available at prediction time, which inflates test results and fails in production.
Interpretability
- Definition
- Interpretability is the degree to which a person can understand how a model turns inputs into outputs by examining the model itself, such as its rules, weights, or structure.
Jupyter Notebook
- Definition
- A Jupyter notebook is an interactive document that combines executable code cells, their outputs, charts, and explanatory text, and it is the standard workspace for data exploration and model prototyping.
k-Nearest Neighbors (k-NN)
- Definition
- The k-nearest neighbors (k-NN) algorithm predicts the label or value of a new record from the k most similar records with known outcomes, using their majority vote or their average.
Knowledge Distillation
- Definition
- Knowledge distillation is a technique that trains a smaller student model to reproduce the behavior of a larger teacher model, so the result runs faster and cheaper while keeping most of the…
Learning Rate
- Definition
- The learning rate is the hyperparameter that sets how large a step gradient descent takes when it updates a model's parameters after each batch of training examples.
LIME (Local Interpretable Model-Agnostic Explanations)
- Definition
- LIME (Local Interpretable Model-Agnostic Explanations) is a technique that explains one prediction from any machine learning model by testing small variations of the input and fitting a simple model to the results.
Linear Regression
- Definition
- Linear regression is a statistical method that predicts a numeric outcome as a weighted sum of input variables, fitting the weights so the predictions come as close as possible to observed values.
Logistic Regression
- Definition
- Logistic regression is a statistical model that predicts the probability of a yes-or-no outcome from input variables, and it remains the most widely used and most explainable classification method in business.
Long Short-Term Memory (LSTM)
- Definition
- Long short-term memory is a type of recurrent neural network with gates that control what information is kept, updated, or forgotten, allowing it to learn patterns that span long sequences.
LoRA (Low-Rank Adaptation)
- Definition
- LoRA (Low-Rank Adaptation) is a fine-tuning method that freezes a model's original weights and trains small added matrices instead, producing a compact adapter that changes the model's behavior.
Loss Function
- Definition
- A loss function is the formula that scores how far a model's predictions are from the correct answers during training, giving the optimizer a single number to minimize.
Low-Dimensional Representation
- Definition
- A low-dimensional representation describes a complex item, such as a document, image, customer, or product, with a short list of numbers that keeps the properties that matter for a task.
Machine Learning (ML)
- Definition
- Machine learning is the branch of artificial intelligence in which software learns patterns from data and improves at a task through experience instead of following rules written by a programmer.
Machine Learning Hardware Requirements
- Definition
- Machine learning hardware requirements are the processors, memory, storage, and network capacity a project needs, which depend on whether the work is classic machine learning, deep learning training, or language model inference.
Machine Learning Pipeline
- Definition
- A machine learning pipeline is an automated, repeatable sequence of steps that turns raw data into a trained, evaluated, and deployable model, so that retraining runs the same way every time.
Mean Absolute Error (MAE)
- Definition
- Mean absolute error (MAE) is the average size of a model's prediction errors, ignoring direction, expressed in the same units as the value being predicted, such as tickets, days, or units sold.
Mean Absolute Percent Error (MAPE)
- Definition
- Mean absolute percent error (MAPE) is the average size of a forecast's errors expressed as a percentage of the actual values, which allows accuracy to be compared across items of very different…
Misclassification Rate
- Definition
- The misclassification rate is the share of all predictions a classifier gets wrong, calculated as false positives plus false negatives divided by all cases. It equals one minus accuracy.
MLOps
- Definition
- MLOps (machine learning operations) is the set of practices and tools for moving machine learning models into production and keeping them accurate, monitored, and governed over their working life.
Model Architecture
- Definition
- Model architecture is the structural design of a model: the types of layers it contains, how they connect, and how data flows from input to output, independent of the learned weights.
Model Deployment
- Definition
- Model deployment is the process of moving a trained model out of the lab and into a running system where it receives real requests, returns predictions, and is monitored, secured, and updated…
Model Drift
- Definition
- Model drift is the gradual or sudden decline in a deployed model's accuracy because the data it sees in production, or the patterns behind that data, no longer match what it learned…
Model Evaluation
- Definition
- Model evaluation is the measurement of how well a model performs against defined criteria, using held-out data, task-specific tests, and safety checks, before and after it is put into service.
Model Inputs and Outputs
- Definition
- Model inputs and outputs are the data a model receives and the results it returns, defined by type, format, and meaning, forming the contract between the model and the systems around it.
Model Maintenance
- Definition
- Model maintenance is the ongoing work that keeps a deployed model accurate, secure, and compliant, including retraining, recalibration, software updates, revalidation, and eventual retirement.
Model Monitoring
- Definition
- Model monitoring is the ongoing tracking of a deployed model's inputs, outputs, accuracy, and operational health, with alerts that tell its owners when performance degrades or the data changes.
Model Prototyping
- Definition
- Model prototyping is the fast, early stage of a machine learning project in which a team builds rough models on real data to learn whether the problem can be solved before investing…
Model Registry
- Definition
- A model registry is a central catalog of trained models that records each version's lineage, evaluation results, approvals, and lifecycle stage, so teams know exactly which model is running where.
Model Robustness
- Definition
- Model robustness is a model's ability to keep producing correct results when its inputs are noisy, unusual, shifted away from the training data, or deliberately manipulated by an attacker.
Model Size
- Definition
- Model size is the scale of a model, usually expressed as its parameter count, which determines how much memory it needs, how fast it runs, and what hardware can serve it.
Model Training
- Definition
- Model training is the process of adjusting a machine learning model's parameters to fit example data, so the model learns patterns it can apply to new cases. Using the trained model afterward…
Model Tuning
- Definition
- Model tuning is the iterative process of adjusting a model's hyperparameters, features, and training setup to improve its performance on validation data.
Model Validation
- Definition
- Model validation is the independent review of a model's design, data, performance, and limitations to confirm it is fit for its intended use before and during deployment.
Model Weights
- Definition
- Model weights are the learned numerical parameters of a trained model, saved as files that can be loaded to run the model, shared under a license, or fine-tuned further.
Naive Bayes
- Definition
- Naive Bayes is a family of fast probabilistic classifiers that apply Bayes' theorem and assume every feature is independent of the others given the class, which makes them a common baseline for…
Neural Architecture Search
- Definition
- Neural architecture search is an automated method that explores many possible neural network designs and selects the one that performs best under given accuracy and compute constraints.
Neural Network
- Definition
- A neural network is a machine learning model made of layers of simple computing units, connected by adjustable weights, that learns to map inputs to outputs by tuning those weights on examples.
Noise (Data)
- Definition
- Noise is random or irrelevant variation in data that carries no useful signal, such as sensor jitter, typos, and inconsistent labels, which a model can mistake for real patterns.
Normalization
- Definition
- Normalization rescales numeric features to a common range or distribution so that no single feature dominates a model because of the units it happens to be measured in.
Null Error Rate
- Definition
- The null error rate is the share of predictions that would be wrong if a model always predicted the most common class. A model's error rate must fall below it before its…
ONNX (Open Neural Network Exchange)
- Definition
- ONNX (Open Neural Network Exchange) is an open file format for trained machine learning models that lets a model built in one framework run in a different runtime, programming language, or device.
Outsourced Labeling
- Definition
- Outsourced labeling is paying an external vendor or crowd to annotate images and other training data. It scales labeling quickly but raises quality control and confidentiality issues that need contractual and technical…
Overfitting
- Definition
- Overfitting happens when a model learns the training data so closely, including its noise, that it performs well on that data and poorly on new data.
Parameter-Efficient Fine-Tuning (PEFT)
- Definition
- Parameter-efficient fine-tuning (PEFT) is a family of methods that adapt a large pre-trained model by training only a small number of added or selected parameters instead of the whole network.
Parameters (Model)
- Definition
- Model parameters are the numerical values, mainly weights and biases, that a model learns during training, and their count is the standard measure of a model's size.
Pre-Trained Model
- Definition
- A pre-trained model is a model that has already been trained on a large general dataset and is distributed for others to use directly or adapt to their own tasks.
Pre-Training
- Definition
- Pre-training is the first and largest phase of building a model, in which it learns general patterns from a huge unlabeled dataset before any task-specific training.
Precision
- Definition
- Precision is the share of a model's positive predictions that are actually correct, calculated as true positives divided by all predicted positives. It is also called positive predictive value.
Prediction
- Definition
- A prediction is the output a trained model produces for a new input, such as a category, a number, a ranked list, or a probability, before the true outcome is known.
Predictive AI
- Definition
- Predictive AI uses historical data to forecast what will happen or to classify what something is, producing a number, a category, or a probability rather than new content. It is the counterpart…
Predictive Maintenance
- Definition
- Predictive maintenance uses sensor data and machine learning to forecast when equipment is likely to fail, so repairs can be scheduled shortly before failure instead of on a fixed calendar or after…
Predictive Model
- Definition
- A predictive model is a trained statistical or machine learning model that estimates an unknown outcome, such as a future event or a hidden attribute, from known inputs.
Prevalence
- Definition
- Prevalence is the share of cases in a population that actually belong to the positive class, such as the fraction of circuit boards that are defective. It shapes how precision and accuracy…
Probabilistic Algorithm
- Definition
- A probabilistic algorithm uses random choices or probability estimates in its steps, so the same input can produce different outputs, or an output expressed as a likelihood rather than a certainty.
Problem Tractability
- Definition
- Problem tractability is the judgment of whether a business problem can realistically be solved with the data, methods, and computing resources available, within the time and accuracy the business needs.
Production Environment
- Definition
- A production environment is the live set of servers, software, and data where an AI system serves real users and real business processes, as distinct from the development and test environments used…
Programming Languages for Machine Learning
- Definition
- The programming languages for machine learning are the languages used to build, train, and deploy models; Python dominates, while R, SQL, Scala, C++, Rust, and Julia each serve particular stages or environments.
PyTorch
- Definition
- PyTorch is an open-source deep learning framework, started at Meta and now governed by the PyTorch Foundation, that has become the most widely used tool for building and training neural networks, including…
Random Forest
- Definition
- A random forest is an ensemble of many decision trees, each trained on a random sample of the data and a random subset of features, whose predictions are combined by voting or…
Real-Time Inference
- Definition
- Real-time inference is serving a model's prediction on demand, within milliseconds to a few seconds of the request, so that the result can be used inside a live transaction or conversation.
Recall (Sensitivity)
- Definition
- Recall is the share of all actual positive cases that a model correctly identifies, calculated as true positives divided by true positives plus false negatives. It is also called sensitivity or the…
Recommendation System
- Definition
- A recommendation system is software that predicts which products, content, or actions a person is most likely to want and ranks them, using past behavior, item attributes, and context.
Recurrent Neural Network (RNN)
- Definition
- A recurrent neural network is a neural network that processes a sequence one step at a time while carrying a hidden state forward, so earlier inputs influence how later ones are interpreted.
Regression
- Definition
- Regression is the family of supervised learning methods that predict a continuous numeric value, such as a price, a quantity, or a duration, from a set of input variables.
Regression Performance
- Definition
- Regression performance is the set of measures that describe how close a model's numeric predictions are to actual values, including mean absolute error, root mean square error, and R-squared.
Regularization
- Definition
- Regularization is any technique that penalizes complexity during model training so the model fits general patterns in the data instead of memorizing individual records.
Reinforcement Learning
- Definition
- Reinforcement learning is machine learning in which an agent learns by acting in an environment and receiving rewards or penalties, gradually finding a policy that earns the most reward over time.
Reinforcement Learning from Human Feedback (RLHF)
- Definition
- Reinforcement learning from human feedback (RLHF) is a training method that teaches a language model to produce answers people prefer by learning from human rankings of its outputs.
Reporting Bias
- Definition
- Reporting bias occurs when the frequency of events, outcomes, or opinions in a dataset differs from their frequency in reality because people record some things more readily than others.
Ridge Regression
- Definition
- Ridge regression is linear regression with a penalty on large coefficients, which stabilizes the model when input variables are highly correlated or numerous and reduces overfitting.
ROC Curve
- Definition
- A receiver operating characteristic (ROC) curve plots a classifier's true positive rate against its false positive rate at every decision threshold, showing the trade-off between catching real cases and raising false alarms.
Root Mean Square Error (RMSE)
- Definition
- Root mean square error (RMSE) is the square root of the average squared difference between predicted and actual values. Because errors are squared, it penalizes large misses more heavily than mean absolute…
Selection Bias
- Definition
- Selection bias arises when the data used to train or evaluate a model is not representative of the population the model will serve, because of how the records were chosen.
Self-Supervised Learning
- Definition
- Self-supervised learning trains a model on unlabeled data by hiding part of each example and teaching the model to predict it, which is how large language models and other foundation models are…
Semi-Supervised Learning
- Definition
- Semi-supervised learning trains a model on a small set of labeled examples together with a much larger set of unlabeled data, cutting the cost of labeling while improving accuracy.
Shadow Mode
- Definition
- Shadow mode runs a new model on live production inputs alongside the current system, records its outputs for comparison, and takes no action on them, so the new model can be judged…
Shapley Values (SHAP)
- Definition
- Shapley values are a game theory method that splits a single model prediction among its input features, showing how much each one pushed the result up or down. SHAP is the most…
Siamese Network
- Definition
- A Siamese network is a neural network that runs two inputs through identical twin branches and compares the results, learning to judge whether the inputs match rather than what class they belong…
Specificity
- Definition
- Specificity is the share of actual negative cases that a model correctly identifies as negative, calculated as true negatives divided by true negatives plus false positives. It is the complement of the…
Stochastic Optimization
- Definition
- Stochastic optimization is optimization in which some inputs, such as demand, prices, or lead times, are uncertain and modeled as probability distributions, so the chosen plan performs well across many possible outcomes.
Structure Learning (Graph Learning)
- Definition
- Structure learning is the process of discovering the structure of relationships among variables from data, such as which edges belong in a Bayesian network or how entities in a graph connect.
Sub-Symbolic (Connectionist) AI
- Definition
- Sub-symbolic AI, also called connectionist AI, stores what it knows as numeric weights across many simple units, as in neural networks, rather than as readable rules. Deep learning and large language models…
Supervised Learning
- Definition
- Supervised learning is machine learning in which a model is trained on examples that carry the correct answer, so it can predict that answer for new, unlabeled records.
Support Vector Machine (SVM)
- Definition
- A support vector machine (SVM) is a supervised learning algorithm that separates classes with the boundary that leaves the widest possible margin between them. It suits small and medium-sized datasets with many…
Synthetic Data
- Definition
- Synthetic data is artificially generated data that reproduces the statistical properties of real data without containing real records, used to train and test models when real data is scarce, sensitive, or unbalanced.
Target Function
- Definition
- The target function is the unknown true relationship between a model's inputs and the outcome it predicts, which training tries to approximate from labeled examples.
TensorFlow
- Definition
- TensorFlow is an open-source deep learning framework developed by Google for building, training, and deploying neural networks, known for its production deployment tooling and its support for mobile and edge devices.
Test Data
- Definition
- Test data is a set of labeled examples withheld from model training and tuning and used once, at the end, to estimate how the finished model will perform on new cases.
Text Classification
- Definition
- Text classification is the task of assigning a piece of text to one or more predefined categories, such as routing a support email to billing, shipping, or returns.
Time Series
- Definition
- A time series is a sequence of measurements of one quantity recorded at successive points in time, such as hourly machine temperature or daily sales, where the order of the observations carries…
Time Series Forecasting
- Definition
- Time series forecasting is the use of historical, time-ordered data to predict future values of a quantity, such as next month's demand for a product or tomorrow's inbound call volume.
Training Data
- Definition
- Training data is the set of examples a machine learning model learns from during training. Its coverage, label quality, and history largely determine how the model behaves on new cases.
Transfer Learning
- Definition
- Transfer learning reuses a model trained on one large task as the starting point for a related task, so the new task needs far less data and compute.
Transformer
- Definition
- A transformer is a neural network architecture that uses attention to weigh the relationship between every element in a sequence at once, and it is the foundation of modern large language models.
Tree-Based Models
- Definition
- Tree-based models are machine learning models built from decision trees, either a single tree or an ensemble of many, such as a random forest or gradient-boosted trees, and are the usual first…
True Negatives
- Definition
- True negatives are cases a model correctly predicts as negative, such as a routine email that a phishing filter correctly lets through. They are usually the largest cell in a confusion matrix.
True Positives
- Definition
- True positives are cases a model correctly predicts as positive, such as a shipment flagged as likely to arrive late that does arrive late. They are the numerator of both precision and…
Underfitting
- Definition
- Underfitting happens when a model is too simple to capture the real patterns in the data, so it performs poorly on both the training data and new data.
Unsupervised Learning
- Definition
- Unsupervised learning is machine learning on data without labeled answers, where the model finds structure on its own, such as natural groupings, unusual records, or a compact summary of many variables.
Validation Data
- Definition
- Validation data is a portion of labeled data held out of training and used during development to compare candidate models, tune their settings, and decide when to stop training.
XGBoost
- Definition
- XGBoost is an open-source software library for training gradient-boosted decision trees, valued for its accuracy on tabular data, its speed on ordinary hardware, and its support across common programming languages.
Zero-Shot Learning
- Definition
- Zero-shot learning is a model's ability to perform a task or recognize a category it was never explicitly trained on, using only a description of what is wanted.
No terms match your search.
Try a broader keyword or another category.