Machine Learning Use Cases

Keith VanderLinden
Calvin University

Machine Learning

Machine learning is an approach to learn complex patterns from existing data and use these patterns to make predictions on unseen data.

The next slides talk through this definition with respect to the example.

References: - DMLS Chapter 1

Learning Complex Patterns

The patterns should be sufficiently complicated and changeable that they can’t be pre-specified. Distinguish:

• Traditional Programming
  inputs + algorithms = outputs

• Machine learning
  inputs + outputs = patterns

The former paradigm works better for computable/consistent phenomena; the latter for hard-to-compute/changeable phenomena.

This has been a real paradigm shift. We spent a half-century trying to program ChatGPT; we trained an ML system to learn it around a week’s computation.

References: - The Peter Norvig video on this issue. - DMLS Figure 1-2.

Finding Existing Data

There must be data for learning that’s:

• Appropriate - Supervised leaning requires input/output pairs.
• Voluminous - ML systems need lots of learning examples.
• Balanced - Skewed data produces skewed results.
• Available - The data must be (ethically) available.

Predicting Unseen Data

Predictions must be:

• Possible - The past and future must have similar distributions.
• Valuable - We actually need to care about what’s being predicted.
• Non-Mission-Critical - We need to be able to live with incorrect predictions.