How Do Machines Learn?

---

… and How Do We Humans Learn?

How do we learn?

How do we learn?

We assemble disordered
information into patterns
that can be more easily recognized

How do we learn?

We assemble disordered
information into patterns
that can be more easily recognized

How do we learn?

We group things
into categories based
on observed
characteristics

How do we learn?

We use prior knowledge to classify new observations into known categories

How do we learn?

We make predictions based on prior knowledge

We test those predictions

We update our understanding based on the outcome of the test

How do we learn?

We make predictions based on prior knowledge

We test those predictions

We update our understanding based on the outcome of the test

Machine Learning

We can use the same ideas to design methods and algorithms for machine learning

• Unsupervised machine learning
• Supervised machine learning

https://vas3k.com/blog/machine_learning/

Unsupervised Machine Learning

• Infer underlying structures within a set of data
• No prior knowledge needed (“unlabeled”)
• E.g., constellations, animal types
• E.g., Principal Component Analysis (PCA), clustering algorithms

Unsupervised Machine Learning

Unsupervised Machine Learning

Unsupervised Machine Learning

Supervised Machine Learning

Identify a function that, given input(s), best predicts an output

Requires data linking inputs with known outputs (“labeled”), e.g.:

• Predicting the tastiness of a sphere (regression)
• Discerning dogs from cats (classification)
• Catching a ball (classification: caught or not?)

E.g., regression analysis, classification algorithms

Supervised Machine Learning

Guess why Captcha is always asking you to select images of traffic lights and cars and motorcycles?

Supervised Machine Learning

Guess why Captcha is always asking you to select images of traffic lights and cars?

https://vas3k.com/blog/machine_learning/

https://vas3k.com/blog/machine_learning/