table of Contents
AI MOOC Peking University
Part I. Basics :Chapter 1. Introduction
§1.1 Overview of Artificial Intelligence
How to create a computer and computer software have the discipline to make it intelligent behavior;
AI was born 1956 Dartmouth conference
Turing test Turing (1950) Computing Machinery and Intelligence Computing Machinery and Intelligence
It is to provide a satisfactory operational definition of intelligence
Visual Turing test Donald - Germaine (2014)
Operator using a auxiliary equipment, generating binary sequences of random questions based on given image
- Computer vision systems are currently testing the accuracy of tasks, including object detection, image segmentation and positioning, but there are still gaps and human behavior. (Humans have the ability to understand images)
- Computer graphics capability assessment
- Pattern recognition and understanding of the relationship between graphic
- There is an association between the subject and the tester only answer yes, no
§1.2 Foundations of Artificial Intelligence
Mathematics:
- Logic: What is the correct conclusion in the form of rules?
- 1847, George Bool: propositional logic;
- 1879, Gottlob Frege: first-order logic, extends the Boolean logic, an increase of objects and relationships;
- 1983, Alfred Tarski: alleged theory, explain how to speak logic objects associated with an object;
Computing: What is computable?
Alan Turing 1912-1954: attempts to accurately describe those functions are computable
Cobham Edmonds 1960s: Concepts tractability calculated
Steven Cook, Richard Karp 1972: proposed the theory of NP-completeness;
P: polynomial time polynomial
NP: Non-deterministic Polynomial time uncertainty in polynomial time
NP-complete: NP and NP-hard intersection.
Probability: How to be inferred from the uncertain information?
- Gerolamo Cardano: to build the concept of probability, game events may result;
- James Bernoulli: the introduction of a new statistical methods;
- Thomas Bayes: Bayesian proposed rule, known as the most modern methods of reasoning under uncertainty.
Neuroscience
How the brain processes information?
The brain is superior to rational decision-making, but did not like the modular software, forecasting and simulation are critical decisions!
Cognitive Psychology
How human thought and action?
Cognitive behavioral and mental processes behind the behavior sciences, to study how the human brain receives input of the external world, the treatment effect (cognitive psychology)
How to form in the brain as well as cross-cutting disciplines transcription process (cognitive science)
The brain is regarded as information processing equipment, is the study of mental processes of the subject.
- Note mechanisms Attention
- First, focus on a subset of information useful to the state of perception
- Language use
- Language acquisition, assembly language form, the tone of the language used, or other relevant art.
- memory
- Memory: three subsets: the process of memory, semantic memory, episodic memory.
- Perceive
- Physical perception (vision), cognitive processes
- Metacognition: cognition about cognition, thinking about thinking, knowledge about knowledge
- Regulation on cognitive knowledge, cognition
- Problem Solving
- creativity
- Think
Control theory and cybernetics
How the machine can operate under its own control
Control theory: interdisciplinary engineering and mathematics: the behavior of dynamic systems handling input, and how to adjust behavior through feedback.
Cybernetics: Exploring the regulation system, their structure, constraints, and possibilities, defined as the scientific study of animal and machine control and communications, the new century is interpreted as "any system with control technology"
§1.3 History of Artificial
- Born 1950-1956 Artificial Intelligence
- Turing Test: machine intelligence metrics
- Dartmouth conference: official birth of artificial intelligence research
- 1956-1974 Golden Age
- 1958 Herbert Simon and Allen Newell first program ai: logic theorists (LT);
- 1958 John McCarthy invented the Lisp language;
- 1960 Masterman semantic network designed for machine translation;
- 1963 Leonard Uhr and Charles Vossler published a paper about pattern recognition, describe the first machine learning programs;
- 1965 Fergenbaum Dendral expert system of the invention, the molecular structure of organic compounds estimation software;
- 1974 Shortliffe demonstrated MYCIN program, a very practical method for medical diagnosis based on rules;
- The first winter 1974-1980
- 1966 machine translation fails
- 1970 abandonment of connectionism
- 197-1975 DARPA understanding of the research project at Carnegie Mellon University speech dismayed;
- 1973, by the Lighthill: impact of "Artificial Intelligence A General Survey" report, AI research substantial reduction of the UK;
- 1973-1974, DARPA, cut funding for general research;
- 1980-1987 boom
- American Association of Artificial Intelligence 1980 AAAI, held the first national conference at Stanford University;
- 1982 Japan launched the fifth generation computer systems FGCS project for knowledge processing;
- 1980-1987 decision tree model is the invention, and exit in the form of software, the model having a visual, easy-described characteristics;
- 1980-1987 Multilayer artificial neural network ANN the invention, having a sufficient number of measurements to hide, a ANN can express any function, thus breaking the perceived limitations;
- The second winter 1987-1993
- 1987 Lisp machine market collapse;
- 1988 US Government Strategic Computing Association canceled the new ai funds;
- 1993 Expert System slow slide into the trough;
- 1990s Japan's fifth generation computer project failed to achieve its initial target, quietly exit;
- 1993-Present Breakthrough
- 1997 IBM Deep Blue chess system to defeat the defending champion;
- 2005 Stanford autonomous robotic vehicle Stanley, won the DARPA Challenge unmanned vehicles;
- 2006 Geoffrey Hinton and Ruslan Salakhutdinov depth study published papers in Science;
- 2011 Watson on Jeopardy! On game victory over two-time champion won the grand prize of $ 1m
- 2011 Google started deep learning project, as Google Brain As one of Google X project, down sixteen thousand computers connected into a cluster dedicated to mimic the human brain, by ten million digital pictures of study, we have successfully learned to identify a cat;
- 2012 Apple introduced Siri, a personal assistant and knowledge navigation software;
- 2012 Microsoft Research chief Rick Rashid Hall demonstrated a real-time English - Chinese generic translation system, the translation is not only accurate, but also keep the speakers accent and intonation;
- 2014.4 Microsoft launched Cortana;
- 2014.6 Microsoft launched Microsoft wheatgrass;
- 2015.9.8 Baidu launch of the secret, secretary of the search services;
- 2014.6 Eugene Goostman activity is 33% of the judges that it is the human, by a Turing test;
- 2014.8 IBM published a humanoid brain work TrueNorth chip;
- 2015.2 Google DeepMind company on Natrue published by Deep Q-Network depth of intensive study to achieve human-level control;
- 2015.12 DeepMind company AlphaGo defeated European champion Go Fan Hui; depth learning software for the first time beat human Go professional player;
- 2016.3 AlphaGo in Seoul, South Korea play against nine players win,
§1.4 The State of The Art
| Humanly | Rationally | |
|---|---|---|
| Acting | Acting humanly | Acting rationally |
| Thinking | Thinking humanly | Thinking rationally |
- Weak AI (ANI): unconscious AI, focusing on a specific task (only for a specific problem);
- Strong Artificial Intelligence (AGI): a smart deal for any questions, research the main objective;
- Super Artificial Intelligence (ASI): an imaginary agent.
Application of Artificial Intelligence
- Computer Vision Computer vision
- Image Processing
- VR、AR、MR
- Pattern Recognition
- Intelligent diagnosis
- Game theory and strategic planning
- ai game
- machine translation
- Natural Language Processing
- Nonlinear Control Robotics
- \(\cdots\)
Appreciation excellent paper:
- "A method for non-linear dimensionality reduction global geometric framework"
- "A neural network to reduce the data dimension"
- "Conducted by quickly finding and found that the density peak clustering"
Artificial Intelligence research areas:
- AI
- SEARCHING
- PROBLEMS SPACE
- REASONING
- KNOLEDGE
- PLANNING
- RULES
- LEARNING
- DATA
- APPLYING
- COMMUNICATING
- NLP
- MACHINE TRANSLATE
- PERCEIVING
- VISION
- SPEECH
- SENSING
- ACTING
- ROBOT
- COMMUNICATING
- SEARCHING
Part I. Basics: Chapter 2.Intelligent Agent
§2.1 Ctbernetics and Brain Simulation
§2.2 Symbolic vs. Sub-symbolic
Symbol mode
Expert system: reasoning by the rules, work rules;
Subsymbolic mode
The researchers are convinced that the symbol system can never imitate the whole process of human cognition, especially perception.