Time and Place: The midterm is 7-9pm on Wed July 24th in NVIDIA auditorium.


Logistics

The exam is open book, open computer, closed internet (you must be disconnected from the web). You will have 2 hours to complete the midterm. Partial credit will be given for partially correct answers and points will be commensurate with how long we expect a problem to take. We are going to stick to a transparent format for the exam. Your test will have five questions:

  1. Short Answer Questions (~5mins).
  2. Deterministic Search Problem (~20mins).
  3. Markov Decision Problem or Adversarial Search Problem (~20mins).
  4. Variable Based Models (~20mins).
  5. Temporal Models (~20mins).

One of the problems will involve writing python by hand. Do not worry about memorizing python. We will grade you on the correctness of your strategy more than on python semantics. The problems in the midterm will be similar in tenor and topic to your pset and programming homeworks, with the exception of the short answer questions. For the short answer questions, any material covered in lecture up to July 18th is fair game.


Practice Problems

Practice Midterm #1 [Download] [Solution]
Practice Midterm #2 [Download] [Solution]

You can always find more problems at the end of each chapter. Some words of advice (1) don't go into the real midterm without practicing. You will learn a lot about our format and our expectations. (2) when working on the practice problems, don't look at the solution until after you have finished. When taking the real midterm you will not have access to solutions :).


Topics Covered

In class so far we have covered a lot of topics. Here is a brief (hierarchical) list of what we have gone over. Hopefully you find it useful while studying.


AI
  - Intelligent Systems
  - AI Pipeline

Education Theory
  - Grit

Search Algorithms
  - Deterministic Search Problems (DSP)
    - DSP Formalization
      - Maze Pathfinding
      - Knight's Tour
      - DNA Alignment
    - BFS
    - DFS
    - Uniform Cost Search
    - A* Search
      - 8 Puzzle
    - Bellman-Ford
  - Consistent Heuristics
  - Search Algorithms Selection Criteria

  - Markov Decision Problems
    - MDP Formalization
      - Snowden
      - Pyramid Solitaire
      - Helicopter Flying Upside Down
    - Expectimax

  - Adversarial Search
    - Zero Sum Games
    - Adversarial Search Formalization
    - Min/Max
      - Chess
    - Alpha/Beta Pruning
    - State Heuristic Evaluation
    - Connect Four

Variable Based Models
  - Constraint Satisfaction Problems
    - CSP Formalization
      - Sudoku
      - Image Segmentation
    - Inferenece
      - Search
      - Improveed Search
      - Arc Conistency
      - Graph Structure
      - Genetic Algorithms
        - Evolving the mona lisa
      - Tree Decomposition
    - Arity
    - Weighted CSPs
    - CSPs vs DSPs

  - Probability
    - Drugs and Rockstars
    - Joint Distribution
    - Factorizing Joint Using Causality
    - Exact Inference
    - Conditional Probability Distributions
      - Conditional Probability Table
      - Gaussian Distribution
    - Independence
    - Conditional Independence
    - Bayes Theorem
    - Expectation

  - Bayesian Networks
    - BN Formalization
      - Diagnosis Networks
      - What does the NSA do with your data?
      - Predicting Misgrades
    - Independence in BN

  - Temporal Models as Bayesian Network
    - Markov Models
      - Predicting Weather
      - Time Slicing
      - Markov Assumption
      - Order and Markov Chains
    - Hidden Markov Models
      - HMM Formalization
        - Control Robotic Arm Using Mind
        - Roomba
      - Filtering using Exact Inference
        - Classroom Robot
      - Particle Filtering (Approximate Inference)
        - Elapse Time Step
        - Observed Evidence

  - Parameter Learning    
    - Stationarity Assumption
    - Learning in Fully Observed Scenarios
      - Learning Transitions
        - Tracking Self-Driving Car
    - Learning with Hidden Variables
      - Expectation Maximization


Good luck have fun