Applied Machine Learning

webstie

Types of Machine Learning

• Supervised learning
  • Input: Examples of inputs and desired outputs
  • Output: Model that predicts output given a new input
• Unsupervised learning
  • Input: Examples of some data (no “outputs”)
  • Output: Representation of structure in the data
• Reinforcement learning
  • Input: Sequence of interactions with an environment
  • Output: Policy that performs a desired task

---

Grading Scheme

• Homeworks (5x): 20%

• Project (team of 3): 25%

  • we will not have presentations for final projects. Historically, we have provided the option of a video presentation of your project for extra credit, but we will not have everyone present to each other.

• Midterms exams: 40%

  • midterm 1, 33.5 / 50.0
  • midterm 2

• In-class quizzes: 15%

  • Approx. weekly short quiz: 5-10 mins at the beginning. Best 50% of scores out of ≈10 quizzes will count.
  • Quizzes will be done on paper
  • Based on last semester’s format, it is not open-book and typically takes up the first five minutes of the lecture, with several questions covering the current or previous week’s topics.
    • 2/5
    • 5/5
    • 3/3
    • 0/?

• Bonus points up to 5% of course grade for extra-good citizenship, e.g.

  • answering lots of questions on Ed Forum
  • bug / typo reporting (more info on this soon)
    • first report bug get 1% for total grade

---

• A+: 95+
• A: 90-95
• A-: 85-90
• B+: 80-85
• B: 75-80
• B-: 70-75
• Lower passing grades: 50-70
• May be curved up

Policy: Human and AI Assistance

Report which AI agent, how you used it, for which problems

Machine Learning

What is Machine Learning?

Algorithms that

• improve their performance 𝑃
• at task 𝑇
• with experience 𝐸

A well-defined machine learning task is given by $(P,T,E)$

Prediction

• 𝑇 = predict Arctic sea ice extent
• • 𝑃 = prediction error (e.g., absolute difference)
• • 𝐸 = historical data

Game Playing

• 𝑇 = playing Chess
• • 𝑃 = win rate against opponents
• • 𝐸 = playing games against itself

Machine Learning Workflow

Types of Learning

• Supervised learning
  • • Input: Examples of inputs and desired outputs
  • • Output: Model that predicts output given a new input
• • Unsupervised learning
  • • Input: Examples of some data (no “outputs”)
  • • Output: Representation of structure in the data
• • Reinforcement learning
  • • Input: Sequence of interactions with an environment
  • • Output: Policy that performs a desired task

Supervised Learning

Given 𝑥1, 𝑦1 , … , 𝑥𝑛, 𝑦𝑛 , learn a function that predicts 𝑦 given 𝑥

• • Regression: Labels 𝑦 are real-valued
• • Classification: Labels 𝑦 are categories
• Inputs 𝑥 can be multi-dimensional

Unsupervised Learning

Given 𝑥1, … , 𝑥𝑛 (no labels), output hidden structure in 𝑥’s

• E.g., clustering

Reinforcement Learning

• Learn how to perform a task from

interactions with the environment

• Examples:

• • Playing chess (interact with the game)
• • Robot grasping an object (interact with the object/real world)
• • Optimize inventory allocations (interact with the inventory system)