Artificial Intelligence By Example电子书

Title Page

Copyright and Credits

Artificial Intelligence By Example

Dedication

Packt Upsell

Why subscribe?

PacktPub.com

Contributors

About the author

About the reviewers

Packt is searching for authors like you

Preface

Who this book is for

What this book covers

To get the most out of this book

Download the example code files

Download the color images

Code in Action

Conventions used

Get in touch

Reviews

Become an Adaptive Thinker

Technical requirements

How to be an adaptive thinker

Addressing real-life issues before coding a solution

Step 1 – MDP in natural language

Step 2 – the mathematical representation of the Bellman equation and MDP

From MDP to the Bellman equation

Step 3 – implementing the solution in Python

The lessons of reinforcement learning

How to use the outputs

Machine learning versus traditional applications

Summary

Questions

Further reading

Think like a Machine

Technical requirements

Designing datasets – where the dream stops and the hard work begins

Designing datasets in natural language meetings

Using the McCulloch-Pitts neuron

The McCulloch-Pitts neuron

The architecture of Python TensorFlow

Logistic activation functions and classifiers

Overall architecture

Logistic classifier

Logistic function

Softmax

Summary

Questions

Further reading

Apply Machine Thinking to a Human Problem

Technical requirements

Determining what and how to measure

Convergence

Implicit convergence

Numerical – controlled convergence

Applying machine thinking to a human problem

Evaluating a position in a chess game

Applying the evaluation and convergence process to a business problem

Using supervised learning to evaluate result quality

Summary

Questions

Further reading

Become an Unconventional Innovator

Technical requirements

The XOR limit of the original perceptron

XOR and linearly separable models

Linearly separable models

The XOR limit of a linear model, such as the original perceptron

Building a feedforward neural network from scratch

Step 1 – Defining a feedforward neural network

Step 2 – how two children solve the XOR problem every day

Implementing a vintage XOR solution in Python with an FNN and backpropagation

A simplified version of a cost function and gradient descent

Linear separability was achieved

Applying the FNN XOR solution to a case study to optimize subsets of data

Summary

Questions

Further reading

Manage the Power of Machine Learning and Deep Learning

Technical requirements

Building the architecture of an FNN with TensorFlow

Writing code using the data flow graph as an architectural roadmap

A data flow graph translated into source code

The input data layer

The hidden layer

The output layer

The cost or loss function

Gradient descent and backpropagation

Running the session

Checking linear separability

Using TensorBoard to design the architecture of your machine learning and deep learning solutions

Designing the architecture of the data flow graph

Displaying the data flow graph in TensorBoard

The final source code with TensorFlow and TensorBoard

Using TensorBoard in a corporate environment

Using TensorBoard to explain the concept of classifying customer products to a CEO

Will your views on the project survive this meeting?

Summary

Questions

Further reading

References

Don't Get Lost in Techniques – Focus on Optimizing Your Solutions

Technical requirements

Dataset optimization and control

Designing a dataset and choosing an ML/DL model

Approval of the design matrix

Agreeing on the format of the design matrix

Dimensionality reduction

The volume of a training dataset

Implementing a k-means clustering solution

The vision

The data

Conditioning management

The strategy

The k-means clustering program

The mathematical definition of k-means clustering

Lloyd's algorithm

The goal of k-means clustering in this case study

The Python program

1 – The training dataset

2 – Hyperparameters

3 – The k-means clustering algorithm

4 – Defining the result labels

5 – Displaying the results – data points and clusters

Test dataset and prediction

Analyzing and presenting the results

AGV virtual clusters as a solution

Summary

Questions

Further reading

When and How to Use Artificial Intelligence

Technical requirements

Checking whether AI can be avoided

Data volume and applying k-means clustering

Proving your point

NP-hard – the meaning of P

NP-hard – The meaning of non-deterministic

The meaning of hard

Random sampling

The law of large numbers – LLN

The central limit theorem

Using a Monte Carlo estimator

Random sampling applications

Cloud solutions – AWS

Preparing your baseline model

Training the full sample training dataset

Training a random sample of the training dataset

Shuffling as an alternative to random sampling

AWS – data management

Buckets

Uploading files

Access to output results

SageMaker notebook

Creating a job

Running a job

Reading the results

Recommended strategy

Summary

Questions

Further reading

Revolutions Designed for Some Corporations and Disruptive Innovations for Small to Large Companies

Technical requirements

Is AI disruptive?

What is new and what isn't in AI

AI is based on mathematical theories that are not new

Neural networks are not new

Cloud server power, data volumes, and web sharing of the early 21st century started to make AI disruptive

Public awareness contributed to making AI disruptive

Inventions versus innovations

Revolutionary versus disruptive solutions

Where to start?

Discover a world of opportunities with Google Translate

Getting started

The program

The header

Implementing Google's translation service

Google Translate from a linguist's perspective

Playing with the tool

Linguistic assessment of Google Translate

Lexical field theory

Jargon

Translating is not just translating but interpreting

How to check a translation

AI as a new frontier

Lexical field and polysemy

Exploring the frontier – the program

k-nearest neighbor algorithm

The KNN algorithm

The knn_polysemy.py program

Implementing the KNN compressed function in Google_Translate_Customized.py

Conclusions on the Google Translate customized experiment

The disruptive revolutionary loop

Summary

Questions

Further reading

Getting Your Neurons to Work

Technical requirements

Defining a CNN

Defining a CNN

Initializing the CNN

Adding a 2D convolution

Kernel

Intuitive approach

Developers' approach

Mathematical approach

Shape

ReLu

Pooling

Next convolution and pooling layer

Flattening

Dense layers

Dense activation functions

Training a CNN model

The goal

Compiling the model

Loss function

Quadratic loss function

Binary cross-entropy

Adam optimizer

Metrics

Training dataset

Data augmentation

Loading the data

Testing dataset

Data augmentation

Loading the data

Training with the classifier

Saving the model

Next steps

Summary

Questions

Further reading and references

Applying Biomimicking to Artificial Intelligence

Technical requirements

Human biomimicking

TensorFlow, an open source machine learning framework

Does deep learning represent our brain or our mind?

A TensorBoard representation of our mind

Input data

Layer 1 – managing the inputs to the network

Weights, biases, and preactivation

Displaying the details of the activation function through the preactivation process

The activation function of Layer 1

Dropout and Layer 2

Layer 2

Measuring the precision of prediction of a network through accuracy values

Correct prediction

accuracy

Cross-entropy

Training

Optimizing speed with Google's Tensor Processing Unit

Summary

Questions

Further reading

Conceptual Representation Learning

Technical requirements

Generate profit with transfer learning

The motivation of transfer learning

Inductive thinking

Inductive abstraction

The problem AI needs to solve

The Γ gap concept

Loading the Keras model after training

Loading the model to optimize training

Loading the model to use it

Using transfer learning to be profitable or see a project stopped

Defining the strategy

Applying the model

Making the model profitable by using it for another problem

Where transfer learning ends and domain learning begins

Domain learning

How to use the programs

The trained models used in this section

The training model program

GAP – loaded or unloaded

GAP – jammed or open lanes

The gap dataset

Generalizing the Γ(gap conceptual dataset)

Generative adversarial networks

Generating conceptual representations

The use of autoencoders

The motivation of conceptual representation learning meta-models

The curse of dimensionality

The blessing of dimensionality

Scheduling and blockchains

Chatbots

Self-driving cars

Summary

Questions

Further reading

Automated Planning and Scheduling

Technical requirements

Planning and scheduling today and tomorrow

A real-time manufacturing process

Amazon must expand its services to face competition

A real-time manufacturing revolution

CRLMM applied to an automated apparel manufacturing process

An apparel manufacturing process

Training the CRLMM

Generalizing the unit-training dataset

Food conveyor belt processing – positive pγ and negative nγ gaps

Apparel conveyor belt processing – undetermined gaps

The beginning of an abstract notion of gaps

Modifying the hyperparameters

Running a prediction program

Building the DQN-CRLMM

A circular process

Implementing a CNN-CRLMM to detect gaps and optimize

Q-Learning – MDP

MDP inputs and outputs

The input is a neutral reward matrix

The standard output of the MDP function

A graph interpretation of the MDP output matrix

The optimizer

The optimizer as a regulator

Implementing Z – squashing the MDP result matrix

Implementing Z – squashing the vertex weights vector

Finding the main target for the MDP function

Circular DQN-CRLMM – a stream-like system that never starts nor ends

Summary

Questions

Further reading

AI and the Internet of Things (IoT)

Technical requirements

The Iotham City project

Setting up the DQN-CRLMM model

Training the CRLMM

The dataset

Training and testing the model

Classifying the parking lots

Adding an SVM function

Motivation – using an SVM to increase safety levels

Definition of a support vector machine

Python function

Running the CRLMM

Finding a parking space

Deciding how to get to the parking lot

Support vector machine

The itinerary graph

The weight vector

Summary

Questions

Further reading

References

Optimizing Blockchains with AI

Technical requirements

Blockchain technology background

Mining bitcoins

Using cryptocurrency

Using blockchains

Using blockchains in the A-F network

Creating a block

Exploring the blocks

Using naive Bayes in a blockchain process

A naive Bayes example

The blockchain anticipation novelty

The goal

Step 1 the dataset

Step 2 frequency

Step 3 likelihood

Step 4 naive Bayes equation

Implementation

Gaussian naive Bayes

The Python program

Implementing your ideas

Summary

Questions

Further reading

Cognitive NLP Chatbots

Technical requirements

IBM Watson

Intents

Testing the subsets

Entities

Dialog flow

Scripting and building up the model

Adding services to a chatbot

A cognitive chatbot service

The case study

A cognitive dataset

Cognitive natural language processing

Activating an image + word cognitive chat

Solving the problem

Implementation

Summary

Questions

Further reading

Improve the Emotional Intelligence Deficiencies of Chatbots

Technical requirements

Building a mind

How to read this chapter

The profiling scenario

Restricted Boltzmann Machines

The connections between visible and hidden units

Energy-based models

Gibbs random sampling

Running the epochs and analyzing the results

Sentiment analysis

Parsing the datasets

Conceptual representation learning meta-models

Profiling with images

RNN for data augmentation

RNNs and LSTMs

RNN, LSTM, and vanishing gradients

Prediction as data augmentation

Step1 – providing an input file

Step 2 – running an RNN

Step 3 – producing data augmentation

Word embedding

The Word2vec model

Principal component analysis

Intuitive explanation

Mathematical explanation

Variance

Covariance

Eigenvalues and eigenvectors

Creating the feature vector

Deriving the dataset

Summing it up

TensorBoard Projector

Using Jacobian matrices

Summary

Questions

Further reading

Quantum Computers That Think

Technical requirements

The rising power of quantum computers

Quantum computer speed

Defining a qubit

Representing a qubit

The position of a qubit

Radians, degrees, and rotations

Bloch sphere

Composing a quantum score

Quantum gates with Quirk

A quantum computer score with Quirk

A quantum computer score with IBM Q

A thinking quantum computer

Representing our mind's concepts

Expanding MindX's conceptual representations

Concepts in the mind-dataset of MindX

Positive thinking

Negative thinking

Gaps

Distances

The embedding program

The MindX experiment

Preparing the data

Transformation Functions – the situation function

Transformation functions – the quantum function

Creating and running the score

Using the output

IBM Watson and scripts

Summary

Questions

Further reading

Answers to the Questions

Chapter 1 – Become an Adaptive Thinker

Chapter 2 – Think like a Machine

Chapter 3 – Apply Machine Thinking to a Human Problem

Chapter 4 – Become an Unconventional Innovator

Chapter 5 – Manage the Power of Machine Learning and Deep Learning

Chapter 6 – Don't Get Lost in Techniques, Focus on Optimizing Your Solutions

Chapter 7 – When and How to Use Artificial Intelligence

Chapter 8 – Revolutions Designed for Some Corporations and Disruptive Innovations for Small to Large Companies

Chapter 9 – Getting Your Neurons to Work

Chapter 10 – Applying Biomimicking to AI

Chapter 11 – Conceptual Representation Learning

Chapter 12 – Automated Planning and Scheduling

Chapter 13 – AI and the Internet of Things

Chapter 14 – Optimizing Blockchains with AI

Chapter 15 – Cognitive NLP Chatbots

Chapter 16 – Improve the Emotional Intelligence Deficiencies of Chatbots

Chapter 17 – Quantum Computers That Think