Machine Learning

Machine Learning is fundamentally the science of getting computers to learn without being explicitly programmed.

In other words, the computer can learn things without a human coder explicitly programming them.

One of the biggest goals of machine learning is to achieve something called "AGI", or Artificial General Intelligence, which is basically human-like intelligence achieved by a computer. Some really interesting research is being done here in the field of "organoid intelligence," where they grow neurons on top of a microelectrode array and make them do various tasks, like play Pong, which it could learn to do in less than 5 minutes! If you're interested in this, totally check out the work done by Cortical Labs!

Now, there are 2 different types of machine learning: Supervised and Unsupervised.

Supervised Learning

• This is the more commonly used application and where the most advancements are being made right now
• The goal here is to learn input to output mappings, or to learn by being given the "right answers"
• For example, imagine giving a student a sheet with questions and answers, and then giving them a new one with only the questions(now they would probably just memorize the questions rather than actually learn, but that's a topic for later, called Overfitting).

There are two types of supervised learning: Regression and Classification.

Regression (Type 1)

• Predict a number from infinitely possible outputs
• For example, predicting the price of a house that's 2000 square feet (you could totally add more features)

Classification (Type 2)

• Predicts categories, a small number of possible outcomes
• For example, categorizing tumors as either benign or malignant, so putting them in a category

Basically, you give questions and answers, and train a machine to either predict a number or a category for different questions.

Unsupervised Learning

• The goal here is to find something interesting in unlabeled data
• The data comes with inputs x, but no output y's, so just the data, not the answers
• For example, imagine giving a student a hundred questions and having them find out what different subjects are being tested

There are three types of unsupervised learning: Clustering, Anomaly Detection, and Dimensionality Reduction.

Clustering (Type 1)

• Group similar data points together
• For example, group articles about pandas together and show them as a cluster

Anomaly Detection (Type 2)

• Find unusual data points in data
• For example, really useful for fraud detection in finance

Dimensionality Reduction (Type 3)

• Compress data to use fewer numbers
• For example, to compress datasets and work with less data, like PCA (we'll talk about this later!)

This is one of the core ideas behind ML, and really where all the roads diverge from.