Glossary Terms

Creating or transforming input variables to help a model learn useful patterns. Select features based on task relevance and check for leakage, sensitivity and availability at the point of use.

Deriving a representation of relevant information from raw data, such as image or text features. Assess whether the representation preserves what the task needs and performs consistently in real conditions.

A process in which results inform subsequent decisions or system behaviour. Design feedback carefully: it can support improvement, but can also reinforce errors or bias when unchecked outputs become future inputs.

A narrowing of information exposure caused by personalised filtering or recommendations. For business decisions, deliberately examine alternative sources and perspectives rather than treating personalised results as a complete view.

A method for examining factors that support or resist a proposed change. Use it to plan practical interventions, making assumptions explicit and reviewing them with the people affected.

Effort, delay or difficulty that makes a task or interaction harder. Examine its cause and business impact before removing it, because some checks provide necessary quality control or accountability.

An approach representing degrees of membership or truth rather than only binary categories. It can help model imprecise rules, but those rules and their consequences still require validation.

Comparing the current situation with a defined desired state to identify what needs to change. Describe the evidence, causes and priorities behind each gap rather than treating every difference as equally urgent.

A model approach in which a generator and discriminator are trained in opposition to produce realistic samples. Generated content still requires checks for quality, appropriateness and the intended use.

An optimisation method that adjusts parameters in a direction intended to reduce an objective function. Successful optimisation does not establish that the objective captures the business outcome or relevant risks.