What is Machine Learning?

• Field of study that gives computers the ability to learn without being explicitly programmed.
  (Arthur Samuel, 1959)
• Machine learning algorithms:
  • Supervised Learning
    • Supervised Learning is used most in real-world applications.
  • Unsupervised Learning
  • Recommender systems
  • Reinforcement learning

Supervised Learning

What is it?

• X -> Y, input to output mappings.
• Give examples to learn from.
• Learns from being given "right answers"
• Later only given X input and tries to predict outputs.
• Example:
  • input -> output
  • email -> spam? = spam filtering,
  • ad, user info -> click? = online advertising.
  • image, radar info -> position of other cars = self-driving car
  • image of phone -> defect? = visual inspection (manufacture)
• Types of supervised learning algorithms are regression and classification.

Regression Algorithm

• predicting a number from infinitely many possible numbers.
• Example: Housing price prediction
  
• What is the price of 750 feet^2?
  • draw a straight line (regression) that fits the data and pinpoint for 750 feet^2 what the price would be.
  • or even better draw a curve.
• The data set (right answers) are given, and the learning algorithm is asked to predict a house price.

Classification

• Predict categories that can be non-numeric or numeric (cat or dog, benign or malignant, (0, 1, or 2)).
• Only a limited number of outputs (category/class), as compared to infinitely many outputs from Regression.
• Example: Breast cancer detection
  • Trying to devise a diagnostic tool for detecting malignant vs benign tumors.
  • Only a limited number of outputs: 0 for benign, 1 for malignant.
    
  • Can be drawn with 2 axes, or just one.
  • attempts to guess malignant vs benign with tumor size.
    
  • More than one input can be used for classification.
  • Attempts to draw the boundary line that separates malignant and benign and tries predicting the result.

Unsupervised Learning

• Finding something intersting in unlabeled data. (All by yourself hehe)
• Data only comes with inputs x, but not output labels y.
  • Algorithm has to find structure in the data.
    
• In the same breast cancer example, the algorithm should be able to find that there are clusters in the data set.
• This is a particular type of unsupervised learning called clustering algorithm.
  • Clustering - Google news: Grouping articles related with pandas & twins & zoo together.
    
  • The algorithm looks for the clusters by itself: in other words, the algorithm figures out that there are some articles with panda & twins & zoos together.
• Anomaly detection algorithm:
  • find unusual data points.
• Dimensionality reduction algorithm
  • compress data using fewer numbers.

Optional Lab: Brief python and Jupyter Notebooks

variable = "something"
print(f"f strings allow you to embed variables {varaible}")