OCI AI Foundations flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Supervised learning

A model is trained on labeled input-output pairs so it can predict the correct label or value for new, unseen inputs.

Unsupervised learning

Training on unlabeled data to discover structure such as clusters, associations, or outliers; clustering and anomaly detection are typical examples.

Overfitting

The model memorizes training noise, producing low training error but poor performance on new data; countered with more data, regularization, or early stopping.

F1 score

The harmonic mean of precision and recall, ranging from 0 to 1; useful when classes are imbalanced and both false positives and false negatives matter.

Recall

Recall equals true positives divided by true positives plus false negatives, measuring the share of actual positive cases the model correctly identified.

Bias-variance tradeoff

High bias means the model is too simple and underfits; high variance means it is overly sensitive to the training sample and overfits.

Backpropagation

An algorithm that applies the chain rule to compute the gradient of the loss with respect to every weight, enabling gradient-descent parameter updates.

ReLU activation

A function returning max(0, x); it is cheap to compute and reduces the vanishing-gradient problem in deep hidden layers.

Convolutional neural network

A network using shared convolutional filters and pooling to learn local spatial features; the standard architecture for image classification and object detection.

Epoch

One complete pass of the entire training dataset through the model; training usually runs for many epochs, each split into smaller batches.

Self-attention

A transformer mechanism in which every token weighs its relationship to all other tokens in the sequence, enabling parallel processing and long-range context.

Token

A subword text unit produced by the model's tokenizer; prompts and completions are limited by context windows and billed by token counts.

Temperature

A sampling parameter controlling randomness: values near zero give focused, repeatable output, while higher values increase diversity and creativity.

Few-shot prompting

Including a small number of worked examples inside the prompt so the model infers the desired task, output format, and style.

Chain-of-thought prompting

Asking the model to show intermediate reasoning steps before its final answer, which improves accuracy on multi-step math and logic tasks.

Retrieval-augmented generation

Retrieving relevant documents from a knowledge base or vector store and inserting them into the prompt so answers are grounded in current, verifiable facts.

Hallucination

Fluent but factually incorrect or unsupported model output; mitigated by grounding with retrieved sources, lower temperature, and human review.

OCI Language

A pretrained NLP service offering sentiment analysis, key phrase extraction, named entity recognition, text classification, translation, and PII detection through APIs.

OCI Vision

An AI service performing image classification and object detection, plus document AI features such as OCR text detection, key-value extraction, and table extraction.

OCI Speech

A service that converts audio to text using batch or real-time transcription, with custom vocabulary support and speaker diarization.

OCI Document Understanding

A service that extracts text, key-value pairs, and tables from scanned documents and classifies document types using pretrained or custom-trained models.

OCI Anomaly Detection

An unsupervised service trained on normal data to flag unusual records in multivariate time series, requiring no labeled anomaly examples.

OCI Forecasting

A time-series service that produces future values with confidence intervals; training data must span at least twice the forecast horizon.

OCI Data Science

A managed platform with JupyterLab notebooks, conda environments, jobs, pipelines, a model catalog, and deployment endpoints for building custom machine learning models.

Accelerated Data Science SDK

Oracle's Python library, ADS, that simplifies data exploration, model training and evaluation, and deployment from notebooks or scripts.

Model deployment

Publishing a trained model artifact as an HTTP endpoint on OCI Data Science so applications can request real-time predictions, optionally with autoscaling.

OCI Data Flow

A managed Apache Spark service for large-scale data processing and feature preparation that feeds machine learning pipelines and data lakes.

OCI GPU shapes

Compute shapes with attached NVIDIA GPUs used to train and fine-tune deep learning models; large clusters add RDMA networking for fast inter-node communication.

Algorithmic bias

Systematic unfair outcomes caused by skewed or unrepresentative training data or flawed objectives, affecting some groups disproportionately.

Explainability

The degree to which a model's predictions can be understood by humans, supported by techniques such as feature importance, SHAP, and LIME.

Least privilege

Granting users and services only the IAM permissions their task requires, limiting the impact if credentials are compromised.

Human-in-the-loop

A design where a person reviews or approves AI output before it drives high-stakes decisions such as lending, hiring, or medical triage.