ITCT101 - Lecture 2.1 - Understanding AI and ML PDF

ITCT101 Computer Technologies Module 2: AI, ML, and Data Science Understanding AI & ML Thanapon Noraset (Nor) [email protected] Agenda 1. AI, ML, DL, and DS 2. Machine Learning  AI Adoption Rate Source: McKinsey Global Survey, 2024 (Link)  AI Adoption Rate...

ITCT101 Computer Technologies Module 2: AI, ML, and Data Science Understanding AI & ML Thanapon Noraset (Nor) [email protected] Agenda 1. AI, ML, DL, and DS 2. Machine Learning  AI Adoption Rate Source: McKinsey Global Survey, 2024 (Link)  AI Adoption Rate AI used by the Experts Source: McKinsey Global Survey, 2024 (Link)  AI Adoption Rate AI used by the Experts AI for general public Source: McKinsey Global Survey, 2024 (Link) What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. John McCarthy and researchers coined the term Artificial Intelligence at Dartmouth College in 1956.  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Two categories: Artificial Narrow Intelligence Artificial General Intelligence  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Artificial Narrow Intelligence (Weak AI) Two categories: AI applications that do well on a specific task. Artificial Narrow Intelligence Artificial General Intelligence Image Source: AlphaGo - The Movie  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Artificial Narrow Intelligence (Weak AI) Two categories: AI applications that do well on a specific task. Artificial Narrow Intelligence AlphaGo is an AI that plays the Go Artificial General Intelligence game. AlphaGo won a Go grandmaster in 2016. Image Source: AlphaGo - The Movie  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Artificial Narrow Intelligence (Weak AI) Two categories: AI applications that do well on a specific task. Artificial Narrow Intelligence Artificial General Intelligence Movie Recommendation Diagnosis & Prescreening Image Restoration Spam mail detection & Synthesis  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Artificial General Intelligence (Strong AI) Two categories: AI applications that do well on many intellectual tasks that human can do. Artificial Narrow Intelligence Artificial General Intelligence  What is Artificial Intelligence? Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Artificial General Intelligence (Strong AI) Two categories: AI applications that do well on many intellectual tasks that human can do. Artificial Narrow Intelligence Not yet exist, but Artificial General Intelligence often portrayed in movies. Machine Learning 1801: Linear Regression (Gauss and Legendre) Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. 1936: Linear Classifier (Fisher) Machine Learning (1990s) Machines that learn to perform tasks from data. 1985: Reasoning with uncertainty: Bayesian networks (Pearl) 1995: Risk minimization: Support vector machines (Cortes and Vapnik) 2000s - present: Machine Learning Machine Learning Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Training Machine Learning (1990s) Machines that learn to perform tasks from data. Training Dataset Training Dataset Learning Usually, an input is Algorithm supervised by a label. Model Machine Learning Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Training Machine Learning (1990s) Machines that learn to perform tasks from data. Training Dataset Model Learning A model captures Algorithm relationship between input and label. Model Machine Learning Artificial Intelligence (1956) Machines that carry out tasks in an intelligent manner. Inferencing Machine Learning (1990s) Machines that learn to perform tasks from data. Input Prediction Model A model is used to predict an output for a new, often unseen, input. Prediction Deep Learning 1943: Artificial neural networks (McCulloch and Pitts) Artificial Intelligence (1956) 1986: Machines that carry out tasks in an intelligent manner. The rise of the backpropagation algorithm Machine Learning (1990s) (Rumelhardt, Hinton, Williams) Machines that learn to perform tasks from data. 2010s: The rise of Deep Learning Deep Learning (2010s) A subset of learning techniques based on Artificial Neural Networks. 2024: Nobel Prize in Physics and Chemistry Deep Learning Deep learning models scale well with the amount of data. Large models typically, artificial neural Performance networks. Also more expensive to operate Small models Amount of training data Data Science  Machine Learning Learning, Common Framework, Problem Types Definition of “Learning” in Machine 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 tasks in T, as measured by P, improves with experience E.” — Tom Mitchell Example Task: Assume pairs of numbers and we would like to predict the “right” number: Performance: Percentage of correct “guess” If a program predicts “(5, 6)”, then the performance is 0%. Example (Cont’d) Task: Assume pairs of numbers and we would like to predict the “right” number: Performance: Percentage of correct “guess” Experience: Data! If a program predicts “(5, 10)”, then the performance is then 100%. Machine Learning Process Learning Training Dataset Algorithm Model Testing Dataset Evaluation Performance Measure Machine Learning: Common Learning Types 1. Supervised Learning: ⭐ e.g., Regression, Classification 2. Unsupervised Learning: ⭐ e.g., Clustering, Dimensionality reduction, Anomaly detection 3. Semi-supervised Learning: e.g., Self-training, Co-training 4. Reinforcement Learning: e.g., Q-Learning, Policy Learning Supervised Learning Unsupervised Learning 1. Clustering 3. Anomaly Detection 2. Dimensionality Reduction Semi-supervised Learning Example Reinforcement Learning Example