Machine Learning (ML)

Machine learning (ML) is the branch of artificial intelligence where systems learn patterns from data rather than being explicitly programmed with hand-coded rules. The canonical definition is Tom Mitchell's (1997): a program learns from experience E with respect to task T and performance measure P if its performance on T, as measured by P, improves with E. Every ML system — from a linear regression forecasting sales to a trillion-parameter LLM writing code — is an instantiation of that pattern: define the task, define the performance measure, provide the experience, and let an optimisation algorithm produce the program.

The field divides into three classical paradigms. **Supervised learning** learns from labelled input-output pairs — classification, regression, sequence labelling. **Unsupervised learning** finds structure in unlabelled data — clustering, dimensionality reduction, density estimation. **Reinforcement learning** learns by interacting with an environment and receiving rewards — games, robotics, recommendation, RLHF for LLMs. Modern foundation models add a fourth paradigm of sorts — **self-supervised pretraining plus adaptation** — that combines the scalability of unsupervised data with supervision derived from the data itself.

For APAC mid-market enterprises, the strategic question is almost never "should we use machine learning" — by 2026 the answer is obvious yes for nearly any workload involving perception, prediction, classification, or generation. The real questions are: **what tier of solution** (hosted API, self-hosted open model, fine-tuned model, from-scratch training), **what build-vs-buy balance** (custom ML vs packaged SaaS with ML inside), and **what MLOps maturity** is needed to operate the thing safely once shipped. The last of these is consistently underestimated in project planning.

The non-obvious truth that older ML engineers learned the hard way: **the bulk of an ML project's effort is not in modelling**. Data cleaning, feature engineering, evaluation design, deployment infrastructure, monitoring, retraining pipelines — these consume 70-90% of engineering hours across the ML project lifecycle. The modelling step that gets written up in papers is the small bright part of the iceberg. Plan, hire, and budget accordingly.