Fine-Tuning

The dataset used to train a machine learning model, comprising examples from which the model learns patterns, relationships, and decision boundaries. High-quality, proprietary training data is a significant competitive advantage and intangible asset, particularly in regulated industries where data scarcity creates barriers to entry.

A machine learning technique where a model trained on one task is repurposed as the starting point for a different but related task, significantly reducing the data and compute required for training. Transfer learning accelerates AI development timelines and reduces costs, making AI adoption more accessible to SMEs.

A mathematical model trained on data to identify patterns and make predictions without being explicitly programmed for each task. Machine learning models underpin many AI-driven business applications, from demand forecasting to fraud detection, and their development costs are increasingly recognised as intangible assets under IAS 38 when they meet the identifiability and future economic benefit criteria.

Proprietary datasets, analytics capabilities, and data infrastructure that provide competitive advantage. Data assets include customer behavioural data, market intelligence, training datasets for AI models, and proprietary databases that improve decision-making or product quality.

A set of practices combining machine learning, DevOps, and data engineering to standardise and streamline the end-to-end lifecycle of machine learning models, from development through deployment to monitoring. MLOps encompasses version control for models and data, automated testing, continuous integration and deployment, and model performance monitoring in production.

A technical architecture that enhances large language model outputs by retrieving relevant information from an external knowledge base before generating a response, grounding the model's output in verified, up-to-date, and domain-specific data. RAG reduces hallucination risk, enables LLMs to access proprietary or recent information not in their training data, and provides citation capabilities.

The degradation in a machine learning model's predictive accuracy over time as the statistical properties of the input data diverge from the training data distribution. Model drift requires ongoing monitoring and periodic retraining to maintain performance, and is a key operational risk in production AI systems.

An automated sequence of data processing steps that extracts, transforms, and loads data from source systems into target systems for analysis, reporting, or machine learning model training. Well-architected data pipelines are critical infrastructure assets that enable data-driven decision-making and AI deployment, and their reliability directly impacts downstream business processes.