Date: May 15, 2024

Topic: Supervised and Unsupervised Learning

Recall

Supervised learning is a function approximation, and requires a paired input and output

Notes

Supervised Learning - Induction and Deduction

For supervised learning and function approximation, we make fundamental assumptions about the work

• Data consists of inputs and outputs
• Well-behaved function consistent with the data we have
• Able to generalize from the data

Induction

Going from specifics (data) to generalities (function approximation)

Deduction

Going from general rule to specific instances (ie reasoning)

Unsupervised learning allows us to find patterns in the data

Unsupervised Learning

We only have input, and need to derive the relationship between the inputs themselves

• We might want to divide groups based on how they look
• E.g., all four legged animals are dogs

Difference from Supervised Learning

Able to divide data up to how we want

• E.g., in a group of people, divide by ethnicity or gender or height etc

These techniques can be combined together as part of a pipeline

Combining Supervised and Unsupervised Learning

It is possible to combine both methods (e.g., density estimation)

<aside> 📌 SUMMARY: Machine learning usually consists of supervised, unsupervised and reinforcement learning algorithms. However, it is important that the data we get is of high quality in order to learn from them

</aside>