FOUNDATIONS OF MACHINE LEARNING ​

Introduction ​

Neural networks and and machine learning.

What is Artificial Intelligence? ​

ChatGPT cannot pass the turing test.

To pass the test the Ai needs these skills:

• Natural language processing (to interact with humans)
• Knowledge representation (to memorize things)
• Automated reasoning (to use the knowledge to answer questions)
• Machine learning (to adapt to new circumstances and detect patterns)
• Perception (Computer vision and speech recognition)
• Robotics (to manipulate objects and move around)

AI: an interdisciplinary edeavor:

Cognitive science, Neurobiology, Philosophy, Sociology, Perception & Learning, Linguitics, Robotics

Geoffrey Hinton
He is the father of deep learning
He can solve very difficult problems with ai

Two approaches to make a computer do what you want:

• Intelligent design (classic algorithms. i.e. Djikstra)
• Machine Learning (AI, the modern way of solving problems)

Two modes of thinking:

• System 1: Fast, conscious, automatic, everyday decisions, error prone
• System 2: Slow, unconscious, effortful, complex decisions, reliable

First successes:

• 1943 McCulloch and Pitts propose a model for an artificial neuron and analyze its properties
• 1949 Donald Hebb proposes a learning mechanims in the brain
• 1950-53 Shannon and Turing work (independently) on chess-playing programs
• 1951 Minsky and Edmonds develop the first "neural" computer
• 1952+ Samuel develops a checker playing game
• 1956 Newell e Simon develop the "Logix Theorist"
• 1957 First attempts at automatic translation (during Cold war)
• 1958 McCarthy invents LIPS (programming language)
• 1956 DARTMOUTH SUMMER RESEARCH PROJECT ON ARTIFICIAL INTELLIGENCE
  • The birth of AI (2 months in summer of research financed by the goverment)
• 1961 Newell and Simon develop General Problem Solver (GPS)
• 1963+ Minsky and students study problems on micro-worlds (es. ANALOGY SHRDLU)
• 1962 Rosenblatt develops the Perceptron, a neural net that learn from examples (classification problem: recognize male vs female). This approach was bad and died:

Failures:

• 1966 Financing to "automatic translation" is project in the USA is canceled
• 1969 Minsky and Papert publish "Perceptrons" where they show that the Rosenblatt model cannot sove some very simple problems
• 1971-72 Cook and Karp develop the computational complexity theory, showing that a lot of problems are "intractable" (NP-complete)

The expert systems are boom:

• 1969 Feigenbaum (Satnford) develop DENDRAL, an ES for making predictions on molecular structures
• MYCIN, an ES with some 450 rules for the diagnosis of infectious diseases
• 1979 PROSPECTOR, an Expert System (ES) for mineral exploration