CS50's Introduction to Artificial Intelligence with Python

Topics that are covering by that course

• graph search algorithms
• adversarial search
• knowledge representation
• logical inference
• probability theory
• Bayesian networks
• Markov models
• machine learning
• reinforcement learning
• neural networks
• natural language processing

Description of the Projects

• Project: Degrees
  • For two sets of CSV data files, where each file contains the same names, and the same structure, we’re interested in finding the shortest path between any two actors by choosing a sequence of movies that connects them. Therefore we are using the breadth-first search algorithm in degrees.py .
• Project: Tictactoe
  • Using the Minimax algorithm, implement an AI in tictactoe.py to play Tic-Tac-Toe optimally.
• Project: Knights
  • Determine how to represent "Knights and Knaves" puzzles using propositional logic, such that an AI running a model-checking algorithm could solve these puzzles for us. An implemenation of that problem can be found in logic.py together with puzzle.py.
• Project: Minesweeper
  • The goal in this project is to build an AI that can play Minesweeper. In doing so we use knowledge-based agents that make decisions by considering their knowledge base, and making inferences based on that knowledge. An implemenation of that problem can be found in minesweeper.py together with runner.py.
• Project: PageRank
  • In this project, we implementing the PageRank’s algorithm. In order to calculate PageRank we are using a combination of two approaches in pagerank.py: First by sampling pages from a Markov Chain random surfer. Second by iteratively applying the PageRank formula.
• Project: Heredity
  • In this project we use a Bayesian Network in heredity.py to make inferences about a population. Given information about people, who their parents are, and whether they have a particular observable trait (e.g. hearing loss) caused by a given gene, our AI will infer the probability distribution for each person’s genes, as well as the probability distribution for whether any person will exhibit the trait in question.
• Project: Crossword
  • In this project we write an AI to generate and solve crossword puzzles. Given the structure of a crossword puzzle (i.e., which squares of the grid are meant to be filled in with a letter), and a list of words to use, the problem becomes one of choosing which words should go in each vertical or horizontal sequence of squares. Therefore we model this sort of problem as a constraint satisfaction problem and use the AC3 algorithm in order to solve that problem. An implemenation of that problem can be found in generate.py together with crossword.py.
• Project: Shopping
  • In this project we write a nearest-neighbor classifier to predict whether online shopping customers will complete a purchas. An implemenation of that problem can be found in shopping.py.
• Project: Nim
  • In this project we write an AI that teaches itself to play Nim through reinforcement learning. In particular, we use the Q-learning algorithm for this project. The implementation of that problem can be found in play.py and nim.py .
• Project: Traffic
  • In this project, we use TensorFlow to build a convolutional neural network to classify road signs based on an image of those signs in traffic.py . To do so, we deploy a labeled dataset, which is in our case the German Traffic Sign Recognition Benchmark (GTSRB) dataset: a collection of images that have already been categorized by the road sign represented in them.
• Project: Parser
  • The goal in this project is to build an AI to parse sentences and extract noun phrases. Therefore we use Natural Language Toolkit (NLTK) together with a context-free grammar formalism to parse English sentences in parser.py.
• Project: Questions
  • The goal in this project is to design an AI in questions.py that can answer questions. Our AI will perform two tasks: document retrieval and passage retrieval. Our AI have access to a corpus of text documents. When presented with a query (a question in English asked by the user), document retrieval will first identify which document(s) are most relevant to the query. To find the most relevant documents, we use tf-idf to rank documents based both on term frequency for words in the query as well as inverse document frequency for words in the query. Once we’ve found the most relevant documents, we’ll use a combination of inverse document frequency and a query term density measure.

References

https://cs50.harvard.edu/ai/2020/