Features
Features:

Product Tour >

Edraw AI >

Paid Plans:

Individuals >

Business >

Eduaction >
Resources
Blog

History

How-tos & Tips

Discovery

Biography

Business Analysis

Examples

AI concept Map

Free AI Mind Map Generator

Onenote Mind Map

Bcg Matrix Examples

Nike Marketing Strategy

Unilever SWOT Analysis

Make Mind Maps in Google Docs

Guide

FAQs

What's New

Resource Center
Templates
All Templates

Brain Storming Templates

Strategy and Planning Templates

Project Management Templates

Product Management Templates

Human Resources Templates

Agile Workflow Templates

Marketing Templates

Education Templates

Fun and Games Templates

User Gallery
Download
Pricing
Enterprise

MindMap Gallery Artificial Intelligence Basics

Artificial Intelligence Basics

Introducing several major developments and basic principles of artificial intelligence.

Edited at 2020-08-17 08:52:49

PlotWizard

Recent works View more works>>

Artificial Intelligence Basics

PlotWizard

Recent works View more works>>

Recommended to you
Outline

Artificial Intelligence Basics

1. The beginning of a new era

machine learning

definition

Simulating human cognitive abilities through machines

how to learn

from data

supervised learning

unsupervised learning

from behavior

reinforcement learning

AI application

Medical/Safety/Manufacturing/Autonomous Driving

AI history

The first wave (1956-1974)

1964~1966 The first natural language conversation program ELIZA

The second wave (1980-1987)

expert system

Solve problems in specific areas

subtopic

The third wave (2011-)

Model and algorithm development

statistical learning

Support Vector Machines

Probabilistic graphical model

Born out of nowhere

1950: :The Turing Test

1951: Minsky's first neural network machine, SNARC

1955: Sou Su Reasoning

2. Observe differences and identify flowers

2.1 Classification

Images, videos, text, sounds

2.2 Extracting features

Feature vector

vector

Feature points and feature space

Feature points: represented points

Feature space: composed of feature points

2.3 Classifier

function from feature vector to predicted class

train classifier

train

Algorithm: Perceptron

Optimization: Adjust parameters to minimize the loss function

Loss function: a mathematical measure of the error in a classifier’s output

Algorithm: Support Vector Machine

Classifier with maximum classification interval in feature space

2.4

application

judge good or bad

test

actual use

2.5 Multi-category classification

normalized exponential function

Compress one vector into another vector

Each element is between 0 and 1

The sum of the elements is 1

2.62 Practical application of classification

Face prediction/cancer prediction

3. Image recognition

3.1 Image classification based on manual features

computer recognition image

matrix:

Pixel: small grid

Resolution: number of grid rows * number of columns

Image features

Humans: Quick Discrimination

Machine: convolution operation

Edge feature extraction

Example: Oriented gradient histogram

3. Histogram stitching

2. Statistics based on characteristics

1. Extract features

3.2 Image classification based on neural network

Feature design to feature learning

Deep neural network structure

multiple sequentially connected layers

convolution layer

1. The first convolutional layer: image input

2. The second convolution layer~: the feature map input of the previous layer

Convert feature map to feature vector

Fully connected layer

Transform the feature vector

Normalized exponential layer

The last layer of the classification network

nonlinear activation layer

The transformation effects of convolutional layers and fully connected layers are preserved

Pooling layer

Reduce the number of calculations and parameters

Artificial neural network and biological neural network

Algorithm: Backpropagation

1. Chain propagation

2. Stochastic Gradient Descent

3.3 Development and Challenges of Deep Neural Networks

"deep" of depth

year 2010

ALEX net: 5 convolutional layers

2016

polynet: 10 convolutional layers

Deep "difficulty"

Too much is not enough

overfitting

Training set OK

Catering to the data Doesn’t perform well on new data

Underfitting

limited ability

poor performance

gradient disappears

Batch processing

Cross-layer connection

3.4 Practical application of image classification

3.4.1 Face recognition

4. Sound distinction

4.1 Listening

Human ear: Acceptable frequency range: 85~1100HZ

sound digitization

Sampling/Quantization/Encoding

sound

1. Voice

2.Music

rhythmic

understand sounds

loudness

tone

Spectrum

timbre

different hearing effects

formant

relatively concentrated areas on the spectrum

4.2 Music style classification

Computer "in-the-ear" style

Feature extraction

feature

Classification

Style classification

Mel frequency cepstral coefficient

low dimensionality

Describe the energy level of different frequencies

9. Go master

9.1 AlphaGo chess network

Supervised Learning Policy Network

KGS platform obtains data

reinforcement learning

Feedback is evaluative

Interaction between agent and environment

Find the best strategy for maximum returns

Reinforcement Learning Policy Network

9.2 AlphaGo’s overall view

Valuation network

situation value judgment

fast moving network

9.3 Alpha yuan

Strategy iteration

strategic assessment

Strategy improvements

constantly alternating

self-play

Integrated into Monte Carlo search

through random deduction

8. Create pictures

8.1 Data space and data distribution

data space

The space where the data is located

Data distribution

The distribution of data in space

Generative Adversarial Network

Generate network

Randomly generate observation data

discriminative network

Distinguish whether data is true or false

potential space

Simple distribution masters the space of complex distribution

8.2 Generating Network

Random point changes to similar images in the dataset

8.3 Discrimination network

Input different pictures to improve judgment ability

8.4 Generative Adversarial Networks

Generative and discriminative networks train each other

7. Understand the text

7.1 Mission characteristics

latent semantic analysis

Corpus: massive text data

Document: stand-alone text

Subject: the main content of the document

7.2 Text features

bag of words model

Chinese word segmentation

Stop words and low frequency words

word frequency

The quotient of the number of times a word appears in a text divided by the total number of words

Inverse document frequency

Negative logarithm of document frequency

7.3 Discover potential themes in the text

topic model

Core: document word frequency = topic proportion * topic word frequency

Matrix multiplication

latent semantic analysis

7.4 Text-based search and recommendation

question

synonyms

Polysemy

6.Category

6.1 First contact with things

supervised learning

Annotation information required for training data

unsupervised learning

No annotation information

6.2 K-means clustering

clustering

The aggregation of data in the feature space is divided into different groups

algorithm

Randomly select K samples

Each sample is assigned to the category corresponding to the nearest cluster center, and a new division method is obtained.

Recalculate the cluster center of each type of sample

6.3 Practical application-crowd classification

Face Detection

face straightening

Feature extraction

face clustering

6.4 Hierarchical clustering and biological clustering

Each sample is separated into 1 column

Repeatedly merge the two most similar classes

5. Identify videos

5.1 Image to video

persistence of vision

video

Hundreds of photos taken continuously

5.2 Video behavior recognition

difficulty

Large differences in behaviors in the same category

Behavior is not well defined

Environmental background varies greatly

behavioral identification features

sports

light flow

Calculation: Correspondence of the same points between 2 frames

definition

A two-dimensional (2D) instantaneous velocity composed of all pixels in the image

hypothesis

The motion of two adjacent frames is small

The colors of two adjacent frames are basically unchanged.

5.3 Video behavior based on deep learning

Single frame recognition

widely varying behavior

Watch TV/write information

video

Recognition: pixel components

level

vertical

Long time: sequential segmented network

Short time: two-stream convolutional neural network

static

spatial flow convolutional neural

dynamic

Time flow convolutional neural