AI and ML Concepts, Technologies, and Tools
Domain 1: AI and ML Concepts, Technologies, and Tools
This domain tests whether you can classify an AI problem correctly and choose technologies that fit it. The exam uses precise definitions: AI is the broad field of machines performing tasks associated with human intelligence; machine learning (ML) is a subset that learns patterns from data rather than following only explicit rules; deep learning (DL) is a subset of ML built on multilayer neural networks. You must also recognize the main learning paradigms by the feedback available.
Core concepts
- Supervised learning uses labeled examples. Classification predicts discrete labels; regression predicts continuous values.
- Unsupervised learning works without labels: clustering groups similar records, dimensionality reduction compresses features, and association finds co-occurrence rules.
- Reinforcement learning trains an agent through rewards and penalties as it acts in an environment.
- Generative AI models data distributions to create new text, images, audio, or code; training may be self-supervised or reinforced with human feedback.
- Structured data fits rows and columns; unstructured data includes text, images, audio, and video; semi-structured data includes JSON, XML, and logs.
Technologies and tools
Match the tool class to the job: data-frame and SQL tools for preparation; ML libraries for algorithms; notebooks for exploration; experiment tracking and model registries for reproducibility; vector databases for embeddings; pipeline orchestration for repeatable workflows. Selection depends on data volume, latency, explainability, cost, and governance, not on trendiness.
Worked example
A utility wants to predict transformer failures within 30 days from hourly sensor readings and maintenance logs. The target is a binary failure flag, so the task is supervised classification, not regression, because the output is discrete. The data is time-series tabular data, so features should include rolling averages, trends, and time since maintenance. A gradient-boosted tree is a sensible baseline; a sequence model is justified only with enough history. The stack needs a data pipeline, experiment tracking, and a model registry so results are reproducible.
Traps
Treating all AI as ML; calling clustering classification; assuming reinforcement learning is supervised learning with delayed labels; assuming deep learning always beats simpler models on small tabular data; confusing a library with a full platform; and missing labels hidden in the problem statement.