2023-HGU-ML Lecture 1. Introduction to AI and ML

Introduction to AI & ML

Came from Henry Choi’s Lecture and ChatGPT’s explanation

• Introduction to AI
  • Weak and Strong AI
    • strong AI: understanding Chinese
    • weak AI: simulating the ability to understand Chinese
  • Applied AI and General AI
  • Computationalism and Connectionism
    • Computationalism
      • thoughts are computation on symbols
        • Symbolic, interpretable
        • e.g) Turing Machine
    • Connectionism
      • Information is represented in neurons and networks
        • Low-level, black-box
        • Neural Networks
  • Turing Test
  • Implementation level and algorithmic level
    • Implmentation level: how the system is physically realized
      • AI and Human Intelligence are different
    • Algorithmic level: how the system does, what representation or process it uses
      • 3 level for intelligent system
        • Learning, Computational level(what the system does and why), Algorithmic level
      • Can achieve some intelligence on approximation
  • Superintelligence
    • bootstrapping from “child machine”, brain emulation, biological cognition, brain-computer interface, networks, and organizations
  • Deep Neural Networks
    • Hebbian learning, perceptron, multilayer perceptron, deep neural networks
  • Practical AI risks
    • affected by viruses
    • misused by people with bad intentions
    • biased AI
    • taking over roles
    • unable to reject AI’s decision
• Introduction to ML
  • What is learning
    • a computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at the tasks improves with the experiences
    • simple model (e.g., linear regression)
    • for supervised learning
      • Learning
      • Recognition
    • Complex models
  • Machine learning
    • Takes data and output → Makes program
    • source of knowledge is data
  • Workflow of machine learning
    • acquisition - data is gathered/collected from various sources
    • preparation - data is cleaned, preprocessed, and eventually becomes a dataset
    • analysis - data is evaluated to run and customize reports
    • modeling - data is patternized and generalized as models
    • visualization
    • deployment and maintenance
  • Components of ML
    • data: features, label, format
    • models: SVM, NN, K-means
    • objectives: cross-entropy, RMSE, likelihood
    • optimization - gradient descent, Newton, linear programming, convex optimization
  • Data
    • structured/unstructured
  • Categories
    • unsupervised learning
      • e.g., clustering, dimension reduction
    • supervised learning
      • e.g., speech/face recognition
    • semi-supervised learning
      • e.g., cancer detection
    • reinforcement learning
      • e.g., AlphaGo, self-driving car
  • Discriminative model and Generative model

    • Discriminative models: p(t

      x)

      • only for supervising

    • Generative model: p(t, x) or p(x

      t)

      • applicable to unlabeled data
      • focusing on modeling each class’ distribution
  • Pattern Recognition
    • measuring → preprocessing → dimensionality reduction → prediction → model selection -