Artificial Intelligence
introduction
1 Overview
Machine learning automatically infers artificial consciousness knowledge representation
Speech, vision, motion recognition
artificial intelligence (Artificial Intelligence) , the English abbreviation is AI. It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a manner similar to human intelligence. Research in this field includes robotics, language recognition, image recognition, natural language processing, and expert systems . Artificial intelligence is the simulation of the information process of human consciousness and thinking. Artificial intelligence is not human intelligence, but it can think like humans and may even exceed human intelligence.
Types of artificial intelligence: Weak artificial intelligence:
Contains basic, role-based tasks in specific scenarios, such as chat robots such as Siri and chess-playing robots such as AlphaGo. Strong artificial intelligence:
There are two categories. One is human-like artificial intelligence, that is, machines think and reason just like human thinking.
The other type is non-human artificial intelligence, that is, machines have completely different perceptions and consciousness from humans, and use reasoning methods that are completely different from humans.
The Turing Test: 1950
Famous test to determine whether a machine can think
Problems:
unfairness of testing
Common knowledge questions that computers cannot answer correctly
Conclusion: Computers have a hard time passing tests
Serler’s Chinese Room Thoughts:
Proving that passing the Turing test does not prove that a computer has human intelligence
2. History
“Artificial Intelligence” was proposed in 1956
Samuel was invited by McCarthy to attend the Dartmouth Conference to present work on machine learning. The term "Artificial Intelligence' was first proposed
Samuel coined the term "machine learning", defining it as "the field of study that gives computers the ability to do so without explicit programming."
Machine learning is a path to achieve the goal of artificial intelligence. Feigenbaum proposed that KE (knowledge engineering) is a discipline that takes knowledge as the processing object and studies the knowledge representation, processing and application methods and development tools of knowledge systems.
The development history of A1 has changed from focusing on "reasoning" to focusing on "knowledge", and finally to focusing on "learning". Machine learning is an algorithm that can automatically "learn", that is, an algorithm that automatically analyzes and obtains patterns from data and uses the patterns to predict unknown data.
Artificial Intelligence > Machine Learning > Deep Learning (Deep Neural Network)
Reasoning (Connection School) -> Knowledge (Symbolism School) -> Learning
3. Research role
The reason why humans can solve problems is because humans have knowledge.
Expert system: It is to sort out the knowledge of experts in relevant fields and let the computer use this knowledge to solve problems in specialized fields. (proposed by the semiotic school)
Traditional knowledge engineering has achieved great success in application scenarios with clear rules, clear boundaries and closed applications.
Automatic theorem proving
The essence of theorem proof is to prove the eternal truth of conclusion O obtained from premise P. In 1958, Wang Hao proved all theorems (220) about propositional calculus and 85% of the 150 theorems in predicate calculus. In 1965, Robinson proposed the principle of reduction, making machine theorem proof a reality. Wu Wenjun, a famous Chinese mathematician and academician of the Chinese Academy of Sciences, transformed geometry into algebra and established a set of machine proof methods, known as "Wu's method"
Man-machine game
Open AI capabilities
AIOPS
4. Commonly used algorithms and learning libraries:
Machine learning
trains the model by inputting massive training data, so that the model can grasp the potential patterns contained in the data, and then accurately classify or predict the newly input data.
Data is big enough
Model: Algorithm with well-trained parameters
Reference materials: [Machine Learning] Classification summary of 14 common machine learning algorithms! -Tencent Cloud Developer Community-Tencent Cloud (tencent.com)
Machine Learning: Top Ten Machine Learning Algorithms - Zhihu (zhihu.com)
-
Supervised Learning Algorithms : During the supervised learning training process, a pattern (function/learning model) can be learned or established from the training data set, and new instances can be inferred based on this pattern. This algorithm requires specific input/output, and you first need to decide which data to use as an example.
Decision Tree Algorithm Decision Tree
Linear Regression Algorithm Linear Regression
Nearest Neighbor/k-Nearest Neighbors (KNN)
Naive Bayes Algorithm Naive Bayes -
Unsupervised learning algorithms (Unsupervised Algorithms) : This type of algorithm has no specific target output, and the algorithm divides the data set into different groups.
Dimensionality reduction and
clustering: automatically assign data to different categories -
Reinforcement learning algorithm (Reinforcement Algorithms) : Reinforcement learning has strong universality and is mainly trained based on decision-making. The algorithm trains itself based on the success or error of the output result (decision). The algorithm optimized through a large amount of experience training will be able to give better results. Prediction. Similar to organisms, under the stimulation of rewards or punishments given by the environment, they gradually form expectations for the stimulation and produce habitual behaviors that can obtain the maximum benefit. In the context of operations research and cybernetics, reinforcement learning is called "approximate dynamic programming" (ADP).
策略迭代 价值迭代 蒙特卡罗算法 时序差分算法
-
Python machine learning library—scikit
Scikit_learn: Machine learning python module
pandas: data processing tool
matplotlib: drawing tool
-
Deep learning library
-
TensorFlow: This name comes from the core component of the framework "Tensor", and its English meaning is "tensor". Tensor is an extension of the vector concept. Vector is a first-order tensor, while a scalar is a zero-order tensor. A matrix can be regarded as a second-order tensor.
-
Keras
is a high-level neural network API written in Python, which can run with TensorFlow, Theano or CNTK, MXNet as the backend.
Tensor is better for model optimization because the space is larger
Keras is more suitable for developers who want to quickly build, train and evaluate models using a plug-and-play framework. It is suitable for testing existing models.
The core of deep learning is feature learning
Recommended reading: "Machine Learning in Practice" "TensorFlow in Practice"