OCI AI Foundations key terms

A branch of AI in which systems learn patterns from data instead of following hand-written rules, then apply those learned patterns to make predictions or decisions on new, unseen inputs.

A training approach that uses labeled examples, where each input is paired with the correct answer, so the model learns a mapping from inputs to outputs that generalizes to new data.

Training on data that carries no labels, letting the algorithm discover structure on its own, such as groups of similar records, hidden patterns, or compact representations of the data.

Training an agent that acts in an environment and receives rewards or penalties, gradually learning a policy that maximizes cumulative reward over time.

A supervised task that assigns an input to one of a fixed set of discrete categories, such as spam versus legitimate mail, or which of several document types a file is.

A modeling failure in which a model memorizes training data, including its noise, so it scores well on examples it has seen but poorly on new data.

The stage in which a trained model is used to produce predictions or generated output for new input, as opposed to the training stage that built the model.

A model composed of layers of connected units, each computing a weighted sum of its inputs plus a nonlinear activation, which together can learn complex patterns from data.

A nonlinear function applied to a unit's weighted sum of inputs, such as ReLU or sigmoid, that lets a network learn relationships beyond simple linear combinations.

The algorithm that computes how much each weight contributed to the model's error by sending gradients backward through the layers, enabling the weights to be updated.

An optimization method that repeatedly nudges model parameters in the direction that reduces the loss, stepping toward a minimum using the gradient of that loss.

A neural network architecture that slides learned filters over local regions of data, well suited to images because it captures spatial patterns such as edges, textures, and shapes.

A neural network architecture built on self-attention that processes an entire sequence in parallel rather than step by step, and underpins most modern large language models.

A technique that lets a model weigh how relevant each part of its input is to each output position, so it can concentrate on the most useful information.

A dense numeric vector representing text, images, or other items so that semantically similar items land close together in vector space, enabling search and comparison.

AI systems that create new content such as text, images, audio, or code by learning statistical patterns from large amounts of training data and sampling from them.

A very large transformer-based model trained on massive text corpora to predict and produce language, capable of summarization, translation, question answering, and conversation.

The process of splitting text into smaller units called tokens, typically words or word pieces, which are the actual units a language model reads and generates.

The practice of designing and refining the instructions, examples, and context given to a generative model so it returns accurate, relevant, and correctly formatted output.

A prompting technique that places a small number of worked examples inside the prompt, letting the model infer the desired pattern or output format without any retraining.

A prompting approach that asks a model to reason through intermediate steps before stating a final answer, which often improves accuracy on multi-step problems.

A pattern that fetches relevant documents from an external knowledge source and inserts them into the prompt, grounding the model's answer in current, private, or verifiable content.

A confident but incorrect or fabricated statement produced by a generative model, often occurring when the model lacks the knowledge or grounding needed to answer the question.

Further training of an already pretrained model on a smaller, task-specific dataset so its behavior adapts to a particular domain, style, or use case.

A sampling setting that controls output randomness; lower values make responses more focused and repeatable, while higher values increase variety and creative variation.

The maximum amount of text, measured in tokens, that a model can consider at one time, covering both the prompt provided and the response generated.

A managed Oracle Cloud service offering pretrained natural language APIs for tasks such as sentiment analysis, entity recognition, key-phrase extraction, and text classification.

A managed Oracle Cloud service for image analysis, providing pretrained and custom models for object detection, image classification, and text extraction from photos and scans.

A managed Oracle Cloud service that converts speech into text and text into speech, supporting transcription of audio with timestamps and batch or real-time processing.

A managed Oracle Cloud service that extracts text, tables, and key fields from scanned or digital documents, classifies document types, and handles forms such as invoices and receipts.

A managed Oracle Cloud service that trains models on historical data to flag unusual data points or behavior, supporting use cases such as fraud, fault, and outlier detection.

A managed Oracle Cloud time-series service that projects future values from historical data, automatically handling trends and seasonality so users need little machine learning expertise.

A managed Oracle Cloud platform for building, training, and deploying machine learning models, offering notebook sessions, reproducible environments, a model catalog, and pipelines.

Publishing a trained model as a hosted endpoint or stored artifact so that applications can send input and receive predictions in a production environment.

A managed Oracle Cloud service that provides access to pretrained large language models for generation, summarization, and chat, along with customization and retrieval-augmented generation features.

Reserved compute capacity in OCI Generative AI that hosts models exclusively for one customer, giving predictable performance and isolation, typically used for fine-tuned or high-volume workloads.

The practice of designing, building, and operating AI systems that are fair, transparent, safe, privacy-preserving, and accountable to the people and society they affect.

Systematic error or unfair skew arising from data, model design, or how a system is used, causing consistently distorted outcomes for particular groups or situations.

The degree to which people can understand and justify why a model produced a given output, supported by techniques such as feature importance or local explanations.

A design pattern in which people review, approve, or correct AI outputs at key points, keeping human judgment in decisions that carry meaningful consequences.