watsonx Data Scientist - Associate key terms

A studio environment for building, training, tuning, and deploying machine learning and foundation model applications, offering notebooks, AutoAI, prompt engineering tools, and managed inference services.

A hybrid, open data lakehouse platform that lets organizations store, query, and govern data across multiple sources using open formats and engines, supporting analytics and AI workloads.

A toolkit for managing AI risk and compliance across the model lifecycle, providing model inventory, factsheets, monitoring, and policy enforcement to support responsible AI practices.

An architecture combining data lake storage with data warehouse management features, enabling ACID transactions, schema enforcement, and efficient querying on open file formats.

An open table format for huge analytic datasets that supports schema evolution, time travel, and ACID transactions, commonly used in lakehouse architectures to manage data reliability.

A distributed SQL query engine for running interactive analytic queries against data sources of various sizes, often used in watsonx.data to query lakehouse tables without moving data.

A large-scale pretrained model, typically based on transformer architecture, that can be adapted to many downstream tasks through prompting, fine-tuning, or retrieval augmentation without training from scratch.

The practice of designing and refining input text to guide a foundation model toward accurate, relevant, and safe outputs, using techniques such as instructions, examples, and constraints.

An interactive tool in watsonx.ai for experimenting with foundation models, crafting prompts, adjusting parameters, and comparing outputs before integrating them into applications.

A prompting technique where several examples of the desired input-output pattern are included in the prompt to help the model infer the task and produce consistent responses.

A prompting method that encourages a model to articulate intermediate reasoning steps before giving a final answer, often improving performance on complex reasoning tasks.

A sampling parameter that controls randomness in model outputs; lower values make responses more deterministic and focused, while higher values increase diversity and creativity.

A decoding strategy that restricts token selection to the smallest set whose cumulative probability exceeds a threshold p, balancing coherence and diversity in generated text.

The process of further training a pretrained model on a smaller, task-specific dataset to adapt its behavior, often using parameter-efficient methods to reduce compute and data requirements.

A set of techniques that adapt a pretrained model by training only a small number of additional parameters, such as adapters or low-rank matrices, while keeping most original weights frozen.

Low-Rank Adaptation, a parameter-efficient fine-tuning method that inserts trainable low-rank matrices into model layers, enabling task adaptation with far fewer trainable parameters.

A fine-tuning approach where a model is trained on examples of instructions paired with desired responses, improving its ability to follow user directions and generalize to new tasks.

The process of making a trained model available for inference in a production environment, often through an online endpoint, batch job, or embedded service with appropriate scaling and monitoring.

A deployment mode that exposes a model as a real-time API endpoint for immediate predictions, suitable for interactive applications requiring low latency and high availability.

A deployment mode that runs predictions on large datasets at scheduled intervals or on demand, optimized for throughput rather than real-time response.

An automated machine learning capability in watsonx.ai that explores data, selects algorithms, engineers features, tunes hyperparameters, and builds candidate pipelines to accelerate model development.

A sequence of data processing and modeling steps, such as preprocessing, feature transformation, and estimator training, that can be trained and deployed as a single reusable artifact.

The automated search for the best configuration of model hyperparameters, such as learning rate or tree depth, using strategies like grid search, random search, or Bayesian optimization.

The process of creating, transforming, or selecting input variables from raw data to improve model performance, often including encoding, scaling, and derived feature creation.

A technique that combines a foundation model with an external knowledge retrieval system, supplying relevant documents as context to improve accuracy and reduce hallucination in generated answers.

A data store optimized for indexing and similarity search over high-dimensional vectors, commonly used in retrieval-augmented generation to find relevant documents by semantic similarity.

A dense numerical vector representation of text, images, or other data that captures semantic meaning, enabling similarity comparison and retrieval in machine learning applications.

An AI system that uses a foundation model to reason, plan, and take actions such as calling tools or APIs to accomplish a goal, often coordinating multiple steps autonomously.

A capability that lets a foundation model request execution of external functions or APIs by returning structured arguments, enabling the model to access data or perform actions beyond text generation.

The ongoing tracking of a deployed model's performance, data quality, and behavior over time to detect drift, degradation, or anomalies and trigger corrective action.

A decline in model performance caused by changes in the statistical properties of input data or the relationship between inputs and outputs after deployment.

The systematic assessment of a model's quality using metrics, validation data, and fairness or robustness checks to determine whether it meets requirements before and after deployment.

A structured document summarizing a model's purpose, training data, performance, limitations, and ethical considerations, used to support transparency and governance.

The governance process of identifying, assessing, mitigating, and monitoring risks associated with AI models throughout their lifecycle, including compliance, fairness, and operational risks.

A software development kit that provides Python libraries and classes for programmatically interacting with watsonx services, enabling automation of model training, deployment, and inference.

An interactive computing environment that combines code, text, and visualizations in a single document, commonly used in watsonx.ai for data exploration, model development, and experimentation.

A REST and Python interface for managing machine learning assets, including training jobs, deployments, and model storage, within the watsonx platform.