Glossary Terms

An approach that focuses on customer value, learning and reducing unnecessary work. Apply it to the whole workflow, keeping controls that are necessary for quality, safety and accountability.

A method for estimating a linear relationship between an outcome and input variables. Interpret results in light of assumptions and data quality; a fitted relationship does not establish causation.

A statistical model commonly used to estimate probabilities for a categorical outcome. Evaluate calibration, thresholds and error consequences in the intended context rather than treating the probability as certainty.

A recurrent neural networkA model made of connected computational units with adjustabl architecture designed to retain and update information across sequences. It can support sequential tasks, with suitability determined by evaluation against alternatives and the operating requirements.

The mathematical objective used to measure model error during training. A lower loss does not automatically mean better business results; the objective must reflect the task and relevant error consequences.

AI methods that learn patterns from data to support predictions or other outputs. Business value depends on representative information, sound evaluation and a workflow in which people can use the output appropriately.

A mathematical framework for sequential decisions involving states, actions, transitions and rewards. Its recommendations depend on how well those assumptions represent the environment and the outcomes that matter.

A statement of the organisation’s purpose, whom it serves and the value it seeks to provide. It helps frame business priorities before selecting tools, initiatives or investment opportunities.

A probabilistic classification approach that uses simplifying assumptions about relationships between features. It can be useful for suitable tasks, but predictions need validation against representative cases.

Methods for analysing, interpreting or generating human language using computers. Applications include classification, retrieval and summarisation; evaluate accuracy, context, information handling and human review for the intended task.