MindMap Gallery Artificial Intelligence Structural Principles and Their Applications
Introduces the relevant structure and principles of artificial intelligence, and introduces the application of artificial intelligence in banking, autonomous driving, etc.
Edited at 2020-10-12 16:08:04This is a mind map about bacteria, and its main contents include: overview, morphology, types, structure, reproduction, distribution, application, and expansion. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about plant asexual reproduction, and its main contents include: concept, spore reproduction, vegetative reproduction, tissue culture, and buds. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about the reproductive development of animals, and its main contents include: insects, frogs, birds, sexual reproduction, and asexual reproduction. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about bacteria, and its main contents include: overview, morphology, types, structure, reproduction, distribution, application, and expansion. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about plant asexual reproduction, and its main contents include: concept, spore reproduction, vegetative reproduction, tissue culture, and buds. The summary is comprehensive and meticulous, suitable as review materials.
This is a mind map about the reproductive development of animals, and its main contents include: insects, frogs, birds, sexual reproduction, and asexual reproduction. The summary is comprehensive and meticulous, suitable as review materials.
Artificial Intelligence Structural Principles and Their Applications
basic knowledge
A new computer age
Business affairs are almost computerized
There are a large number of mobile terminal users
AI (artificial intelligence) and loT (Internet of Things) era
What is artificial intelligence
AI recommended wine
Enter your feelings after trying the drink
Analyze people’s taste sensitivities
AI recommends a drink to try next time
Analyze again, just try a few drinks and combine the flavors to make accurate recommendations
Interesting facts about artificial intelligence
First appeared at the Dartmouth Conference in 1956
early definition
Collect a lot of news, conduct database searches, and then predict and judge the future
General Artificial Intelligence AGI
It doesn’t exist yet, it’s far from completion.
required technology
Identify species and people
Awareness of surrounding distances and conditions
natural communication communication
Understand the other person’s feelings
Answer the question correctly
judgment on things
Strong AI and weak AI (AGI and special AI)
AGI
Ability to source multiple solutions on your own
Ability to solve problems beyond imagination
Autonomous understanding/autonomous behavior
It is people’s dream since the birth of AI technology
It is difficult to achieve and few measures can be taken
Special AI
Intelligent behavior in individual areas
AI that surpasses human capabilities has already been partially put into practical use
Chess/Go
Autopilot
medical diagnosis
How does the brain identify and judge?
left brain and right brain
Different functions
left brain logical thinking
Right brain sensory perception
Consider methods in AI research
Using computers to create something identical to the human brain itself
Realize the same "thinking and logic" and "perception and perception" functions as humans through computers
Nerve cells and synapses
nerve cells (neurons)
Information processing and transmission of information to other neurons
synapse
connecting elements between neurons
Number recognition method (former method)
pattern matching
if...then form rules
Machine learning and big data
Computers learn by parsing data (big data)
Progress in neural networks is due to the accumulation of big data
Feature vector
It is a feature that a computer analyzes and extracts from a transaction. It is a vector value (a combination of multiple numbers) inside the computer.
Humans extract features from experience
Machine learning that learns like humans
The core of artificial intelligence (AI)
machine learning
Neural Networks
deep learning
"Machine learning" based on "neural networks" (mathematical models) constructed using "deep learning"
The impact of neural networks
google cat
Marking the third boom of artificial intelligence
Google research team "Google X Labs"
After learning thousands of pictures, the neural network discovered the existence of cats and understood the characteristics of cats
Game AI Computer-DQN
computer playing video games
Didn’t learn game rules and scoring methods in advance
Learn the rules of the game and get high scores on your own
Deep learning achieves decisive victory in image recognition competition
International competition related to the image database "ImageNet" developed by Stanford University
The computer first recognizes the image, and then lets it guess what was captured in the image.
"Super Vision" led by University of Toronto's Jeffrey Hinton has an error rate of 17%
AlphaGo beats powerful Go player
Four wins and one loss
Crash-free car
"2016 International Consumer Electronics Show" in Las Vegas, USA
Developed by Toyota and Nippon Telegraph and Telephone Company
Autonomous driving and autonomous learning
The car makes decisions and drives autonomously
Driving routes are arbitrary and not predetermined
Infrastructure to support the cars of the future
Interact with the surrounding environment
Surrounding street camera interaction
Vehicles in front share information
Artificial Intelligence Principles
Machine learning methods
supervised learning
Mark the correct answer
Read large amounts of tag data
Understanding the "eigenvector" of the correct answer
unsupervised learning
Learn from data without correct answer labels
Statistically extract commonalities, connections and correlations
A learning method for developing systems that derive correlation functions
Classification and regression problems
Classification problem
Distinguish after identification
regression problem
"Unsupervised learning" effectively solves regression problems
"Unsupervised learning" is learning based on data whose output solution cannot be determined
Export statistical functions
To calculate the predicted value
Analyze data, discover features and calculate functions
Regression problems are often used in
data observation
statistics
Continuously changing values (stock price information)
reinforcement learning
It is a machine learning method that combines supervised learning and unsupervised learning.
Reinforcement learning with AlphaGo
Study past chess games played by chess players
Games between computers (approximately 30 million moves)
Use learning content to play against professional Go players
Experience and compensation
Robot learning to ride a bike
Earn distance and get rewarded with points
Go smoothly, save time and get paid
Learn to get paid more
advantage
Automate algorithm debugging
Reduces the workload of engineers
Neural network principle
unitary learning theory
All information processing within the human brain is through the same pattern recognition
computer application
Computer simulates human brain using "neural network" mathematical model
Simultaneously simulates the recognition mode of brain nerves
sensor
The simplest perceptron
input layer
output layer
For the same input, the output is not constant
deep learning
thinking model
perceptual model
input layer
output layer
middle layer
The output layer considers the results of the intermediate layer and then outputs it.
"Deep Neural Network" (DNN)
Increase the number of nerves in the middle layer
Increase the number of intermediate layers
Machine learning using DNN is "deep learning"
The load of parallel computing is huge
CNN and RNN
Convolutional Neural Network (CNN)
Processing photo image recognition and analysis
convolution
Terminology for image compression/decompression and radio communications
Excellent identification of feature vectors
Excellent analysis of still images
Decompose and analyze the material in a surprising way, and gradually expand it to a wider range for processing
When analyzing in a small range, the relationship around the image is strong and the relationship in the separated parts is weak.
Recurrent Neural Network (RNN)
Identification and analysis of important information situations in time series, such as numerical changes, etc.
time series information
Contextually important information
CNN is used for voiceprints, RNN is suitable for conversations
Back Propagation
Back-to-back learning method
Applied to both RNN and CNN
Cognitive Systems and A Chatbot
What is IBM Watson
Computers surpass humans for the first time
chess game
Deep Blue's first victory in May 1997
IBM Watson developed for IBM
Intelligence test surpasses humans
Tested on the quiz show "Jeopardy!"
Must understand human natural language
Watson active in the medical field
Find appropriate answers based on the questions and inquiries received, and return the answers to the customer
Process relevant data of hospital patients in a timely manner to derive the best answers and related predictions
Watson helps treat cancer and leukemia
Shorten the development time of effective new drugs
Search cancer-related papers, diagnose the patient's symptoms, and infer the name of the disease and treatment methods
What is cognitive system
third generation computers
cognitive systems era
Technology that allows the system to answer questions through autonomous learning
structured data
Designed to be understood and read by computers
unstructured data
Humans can read it, but computers can’t.
Cognitive systems must support the ability to read and understand "unstructured" data
What is Watson's entity?
No dedicated database
Dedicated solutions (Offering)
A framework that defines a design for use in a specific domain
Product
Watson Explorer
Watson Browser
Watson Analytics
Cloud analysis tools
Use natural language query to connect to the database in the background
Application
Program published as application software or WebService
Platform
Watson Developer Cloud
For developers
Watson Zone on Bluemix
Provide IBM general development tools group
Six major functions of IBM Watson Japanese version
Natural Language Classifier
Dialog
Document Conversion
Search and Rank (Retrieve and Rank)
Speech to Text
Text to Speech
Watson import example (1)-call center
Mizuho Bank call center imports Watson
Artificial Intelligence and Robots - Bank Reception
Let customers play intelligence test or lottery game in waiting area, appease customers, marketing of insurance products
Ability to respond to customer questions with high accuracy
Future customer reception
Identify customers through facial recognition
Guide customers to a dedicated reception room, and deliver content to relevant staff quickly and in real time
Robot opinion as second opinion for reference
Watson's import example (2)-sales support
Watson autonomously self-corrects and answers more correctly next time
The principle of Watson answering questions (6 Japanese version APIs)
Use "NLC" and "DLG"
After natural language processing, dialogue between
Use "NLC" and "R&R"
After natural language processing, a search engine is used to query the database
Use "STT" and "TTS"
Sound to text conversion and text to sound conversion
IBM Watson Japanese version solution package
background
High cost of use
Quantity-based payment system
Charges based on data transaction volume
Have function
AI chatbot
hitTO
AI-Q
Email reply support
technomarkCloud
Watson-linked Pepper reception and customer guidance
e-Reception Manager for Guide
The key to importing AI into chatbots
For internal use within the company
Answer employee questions and improve efficiency
AI chatbot provided
Language input problem
Automatically answer questions and provide the most appropriate response
corpus
Data that combines text and vocalizations into a database
Feedback on answer results
Correlate the correct answer with AI based on feedback
Watson's import example (3)-email reply support
Watson automatically inserts accurate sentences
In addition to operator answers, machine learning (additional consultation mode)
Analyze personality, emotion and tone of voice from a tweet or email
Analyzing personality from Twitter
"Tone Analyzer"
Analyze the tone of an article
Analyze expression tendencies such as "anger" and "sincerity"
Artificial Intelligence in Practical Applications
Applications in call centers and customer reception
Big Japanese banks use AI and bots
Customer reception by Watson and robots
At the bank reception desk, facial recognition determines who the customer is
Listen to customer needs and guide them to Pepper
Pepper accepts detailed inquiries from customers, provides answers, and sends messages to customers’ mobile phones.
Humans and robots working together
Solve short-term personnel problems
Artificial Intelligence Chatbot
Facebook M Become a Consultant
LINE Customer Connect
Driving chatbots to implement customer support services utilizing "LINE"
By importing this service, businesses can use LINE to respond to inquiries from their web sites or LINE accounts.
Apply to FAQ
By accumulating unsatisfied problems and updating the FAQ through machine learning or manual updating, the proportion of solved problems is increased.
Artificial intelligence begins to be active in the medical field
Examine MRI images and detect abnormalities
Get a diagnosis under the supervision of a professional doctor to detect abnormalities
The final diagnosis was made by a professional doctor
Can be quickly diagnosed
Help doctors discover abnormalities that have been overlooked by doctors
Extract answers from huge amounts of data
Cancer Gene Analysis
Reception, precise search for the best department for patients
Get accurate diagnosis and treatment as early as possible
Reduce doctors’ misdiagnosis and omissions
Consultation by robot
Artificial intelligence for composition in the style of Beatles
People are responsible for arranging and composing lyrics
Artificial Intelligence Composition
Import 14,000 music scores for learning
Choose from 45 Beatles Instrumental Learning Styles
Interactively select the best parts to arrange, mix, and complete the composition
Artificial intelligence that understands emotions
Pepper's two emotion generators
emotion recognizer
Read the other person’s emotions
Analyze voice intonation
Analyze facial expressions
emotion generator
Let robots have human-like emotions
emotion matrix
Similar secretion of hormones, emotional types and physiological responses
emotional map
Similar to the modeling of emotions such as "excitement" and "aggressiveness" that occur with the increase or decrease of hormones
Visualizing the emotions of motorcycles on the track (Honda)
See the video of the ride while the emotions of the motorcycle are in turmoil
Want the car to communicate emotionally with the driver
The future of communication with motorcycles (Kawasaki)
Release concept film
Motorcycles of the future communicate with riders and grow together
Communicate to prevent traffic accidents before they happen
Job Hunting Artificial Intelligence
Provide suggestions for study courses that suit you
The higher the suitability, the higher the probability of receiving credits.
the ability to discover that one has not noticed
Diagnose underlying personality
Recommend job search information that matches the user's job search intentions
Values-based matching
To ensure a high degree of matching between job seekers and companies
matchmaking service
Artificial intelligence that can stop writing novels or news
A short novel in the style of "Hoshi Shinichi"
Not yet capable of writing novels
Write formatted news reports
Associated Press
Using "WordSmith" to write artificial intelligence
Nihon Keizai Shimbun
Fully Automated Artificial Intelligence Final Account Summary
Why writing a novel is difficult
Questions to answer
Is it a fixed format or a non-fixed format?
How much creativity is required?
The most important thing about machine learning mechanism is "reward"
The reward is clear
You can’t score points when writing novels
Evaluation of novels is more subjective
Another way to explain
Reconstructing stories based on people’s thinking creates stories for people
Discovery from big data is the achievement of artificial intelligence
Other application examples
AI monitors the network and detects anomalies
Virus characteristics
When a virus is activated, legacy systems discover files or executable code from build mode and block it.
But using software such as WMI and PowerShell to collect mechanical energy, it usually goes unnoticed
AI detection principle
AI can monitor all terminals on the network and learn the usual operations and actions of each terminal
After learning the usual operations, regard this state as the normal state
Once a different operation or action is performed, a warning will be issued to draw attention
Follow-up development of AI sommelier
Applications in fields such as fashion and wine
Recommend suits, shoes, coordinated combinations, etc. that are completely suitable for the user
Apply for a patent in the United States
AI system that reads micro-expressions
Affectiva
Affdex is artificial intelligence for facial expression recognition
Has the world's largest emotion database and artificial intelligence that acquires emotions through deep learning
mechanism
computer measuring emotion
The visual receptor (camera) tracks the key points and movements of the human face
Analyze subtle facial movements
Relate complex emotions to data
Determine important facial landmarks (nose, eyebrows, mouth, etc.)
Classified according to color, texture, light grayscale, etc.
Recognize and track the location of dozens of accurate parts of an individual's face and capture subtle muscle movements
Reflected into the database as data
Analyze information and accumulate information on how the corresponding content is related
Emotions and expressions through watching advertisements or taking classes, used for
Medical care, nursing, consultation, etc.
Sentiment analysis of robots
The latest technology in AI computing
Microsoft Cognitive Services (Microsoft Azure)
language
Language understanding intelligent service (LUIS)
Teach the application to understand commands issued by the user
Text Analysis API (Text Analytics API)
Assess sentiment and themes to understand user needs
Web Language Model API (Web Language Model API)
Leverage the power of predictive language models targeting data on the web
Bing Spell Check API (Bing Spell Check API)
Detect and correct typos in applications
Text Translation API (Translator Text API)
Automatic text translation is easy with a simple call
visible
Image processing algorithms that enable personalized content review by returning insights such as images, emotion recognition, and more
Face API (Face API)
Detect, analyze, organize and label faces in photos
Emotion API (Emotion API)
Personalized user experience through emotion recognition
Computer Imaging API (Computer Vision API)
Extract information useful for decision-making from images
content moderator (Content Moderator)
Automatic review of images, text and videos
voice
Handle audio language in your application
Bing Voice API (Bing Speech API)
Convert speech to text and back again, and understand user intent
Speaker Recognition API (Speaker Recognition API)
Use speech recognition and authenticate individual speakers
Speech Translation API (Translator Speech API)
Easily implement real-time speech translation with a simple REST API call
Custom voice service (Custom Speech Service)
Improve accuracy when speech recognition has difficulty responding to issues such as the client's speaking style, surrounding noise, vocabulary, etc.
search
Deepen the linkage with BingSearchAPI to make the use of applications, web pages and other functions more convenient
Bing Search API (Bing Search API)
Search API to search web documents, images, videos, news and get comprehensive results
Bing automatic recommendation API (Bing Autosuggest API)
Provides apps with smart auto-suggestion options for searches
Knowledge
Establish mappings between complex information and data so that tasks such as reasoned recommendations and semantic search can be performed
Recommended API (Recommendations API)
Predict and recommend items customers want
Academic Knowledge API (Academic Knowledge AP1)
Use rich education-related content from Microsoft Academic Graph
Concrete experience of image and animation analysis technology
Microsoft Computer Imaging API
Parse pre-prepared sample images
You can also parse any image
Detect text in images and convert them into text
Google's Cloud Vision API
Cloud-based image analysis
The rest of the functions are the same as Microsoft
Deep learning and GPU
NVIDIA transforms from a "visual computing company" to an "AI computing company"
The role of CPU and GPU
Basic computer processing is handed over to the "CPU"
Calculation processing such as 3D and CG is left to the "GPU"
GPU active in AI computing
Deep learning method, huge amount of calculation processing
Processing requires "matrix operations", and GPU matrix operations are strong
GPU is more than 10 times faster than CPU
AI computing used in autonomous driving and robots
Process camera and sensor information in near real-time and understand your surroundings
AI board "DRIVE PX2"
For autonomous cruise (autonomous driving on highways)
For point-to-point autonomous driving (autonomous driving between specific places)
For fully automatic control
Embedded AI board "JETSON TX1"
humanoid robot
automatic sweeper
Handling robot
Easily implement deep learning frameworks
As long as you use a deep learning library, you can embed deep learning into the system
TensorFlow
Japanese company
Chainer
American University Research Center
Caffe
Theano
Torch
Minerva
GPU bridging library
cuBLAS
cuDNN
Intel is making a comeback relying on A high-speed technology using CP∪
Introduce multi-core technology to CPU and optimize it
CPU FPGA performance
FPGA
field programmable gate array (Field-Programmable Gate Array)
Significantly shorten the development cycle
Good maintainability
High scalability
All processing data is saved in on-chip memory and temporary calculations are performed
Accelerate neural network calculations such as deep learning through composition
test
CPU FPGA
513 pictures per second, power consumption 294W
CPU
51 frames per second, power consumption 130W