A five-session introduction to how AI/ML systems are actually built — the concepts behind them, the data that sets their ceiling, the lifecycle around them, and the methods that explain what they decided.

Learning Goals:

  • Place machine learning and deep learning within AI, tell the five types of learning apart by what each one needs from your data, and recognize the main computer vision and NLP tasks by what they take in and return
  • Judge a dataset before trusting a model: where training data comes from, how the label unit changes across tables, text, images, audio, video, time series, satellite tiles and graphs, and why annotator agreement and the split key decide what your numbers mean
  • Follow a project through the eight lifecycle stages and two loops that turn at different speeds, and set the success criterion, the baseline, and the deployment exposure level before the model exists
  • Run a classical ML pipeline end to end — problem definition, exploratory analysis, cleaning, feature engineering, splitting, algorithm choice, evaluation, hyperparameter tuning, and deployment with monitoring — and know why classical methods still win on tabular data
  • Explain a single prediction using LIME, SHAP, and DiCE, and tell intrinsic from post-hoc and local from global explanations when choosing a method