Glossary Terms

An innovation that can change market participation or established business models, often by addressing overlooked needs. Not every technological improvement is disruptive; assess the actual market mechanism and evidence.

A neural networkA model made of connected computational units with adjustabl training technique that temporarily omits selected units to reduce reliance on particular pathways. It is a form of regularisation, not a guarantee that a model will generalise.

An organisation’s ability to sense changes, seize opportunities and adapt resources or ways of working. AI adoption requires such adaptation, including leadership decisions, workforce capability and process redesign.

A model component that maps discrete inputs, such as words or identifiers, to numerical vectors. The learned relationships can support downstream tasks but require evaluation for relevance and possible bias.

Approaches that combine multiple models to produce a prediction or decision. Combining models may improve performance, but can increase complexity, cost and the effort required to explain or maintain the system.

One complete pass through a training dataset during model training. Additional epochs may improve fit or cause overfittingWhen a model learns the training data too closely and perfor; select training duration using appropriate validation rather than the count alone.

Optimisation methods that iteratively generate and select candidate solutions using processes inspired by evolution. Their usefulness depends on the objective, constraints and resources required to evaluate candidates.

An examination of market, customer, competitor and wider environmental conditions. Use reliable sources and stated assumptions to inform strategy, rather than treating a forecast as an established fact.

The harmonic mean of precisionFor a classification task, the proportion of predicted posit and recallFor a classification task, the proportion of actual positive for a classification task. It provides one view of performance, but should not replace analysis of individual error types, class balance and business consequences.

An assessment of whether an initiative can be delivered under practical conditions and constraints. Examine information, people, systems, time, cost and risk before making an implementation commitment.