Artificial Intelligence by Example

Contents......Page 3
Preface......Page 14
1 Adaptive Thinker......Page 20
How to be an adaptive thinker......Page 21
Addressing real-life issues before coding a solution......Page 22
Step 1 – MDP in natural language......Page 23
From MDP to the Bellman equation......Page 26
Step 3 – implementing the solution in Python......Page 30
The lessons of reinforcement learning......Page 32
How to use the outputs......Page 33
Machine learning versus traditional applications......Page 37
Summary......Page 38
Further reading......Page 39
2 Think like Machine......Page 40
Technical requirements......Page 41
Designing datasets in natural language meetings......Page 42
Using the McCulloch-Pitts neuron ......Page 43
The McCulloch-Pitts neuron......Page 44
The architecture of Python TensorFlow......Page 48
Overall architecture......Page 50
Logistic function......Page 51
Softmax......Page 52
Summary......Page 56
Further reading......Page 57
3 Machine Thinking to Human Problem......Page 58
Determining what and how to measure......Page 59
Convergence......Page 61
Numerical –  controlled convergence......Page 62
Evaluating a position in a chess game......Page 64
Applying the evaluation and convergence process to a business problem......Page 68
Using supervised learning to evaluate result quality......Page 70
Summary......Page 74
Further reading......Page 75
4 Unconventional Innovator......Page 76
XOR and linearly separable models......Page 77
Linearly separable models......Page 78
The XOR limit of a linear model, such as the original perceptron......Page 79
Step 1 – Defining a feedforward neural network......Page 80
Step 2 – how two children solve the XOR problem every day......Page 81
Implementing a vintage XOR solution in Python with an FNN and backpropagation......Page 85
A simplified version of a cost function and gradient descent......Page 87
Linear separability was achieved......Page 90
Applying the FNN XOR solution to a case study to optimize subsets of data......Page 92
Summary......Page 98
Further reading......Page 99
5 Machine Learning & Deep Learning......Page 100
Building the architecture of an FNN with TensorFlow......Page 101
Writing code using the data flow graph as an architectural roadmap......Page 102
The input data layer......Page 103
The hidden layer......Page 104
The output layer......Page 105
Gradient descent and backpropagation......Page 106
Running the session......Page 108
Checking linear separability......Page 109
Designing the architecture of the data flow graph......Page 110
The final source code with TensorFlow and TensorBoard......Page 112
Using TensorBoard in a corporate environment......Page 113
Will your views on the project survive this meeting?......Page 114
Summary......Page 117
References......Page 118
6 Optimizing Solutions......Page 119
Dataset optimization and control......Page 120
Designing a dataset and choosing an ML/DL model......Page 121
Agreeing on the format of the design matrix......Page 122
Dimensionality reduction......Page 124
Implementing a k-means clustering solution......Page 125
The data......Page 126
Conditioning management......Page 127
The k-means clustering program......Page 128
The mathematical definition of k-means clustering......Page 130
The goal of k-means clustering in this case study......Page 131
1 – The training dataset......Page 132
3 – The k-means clustering algorithm......Page 133
5 – Displaying the results – data points and clusters......Page 134
Test dataset and prediction......Page 135
Analyzing and presenting the results......Page 136
AGV virtual clusters as a solution......Page 137
Questions......Page 139
Further reading......Page 140
7 When & How to use AI......Page 141
Checking whether AI can be avoided......Page 142
Data volume and applying k-means clustering......Page 143
NP-hard – the meaning of P......Page 144
Random sampling......Page 145
The law of large numbers – LLN......Page 146
Using a Monte Carlo estimator......Page 147
Training the full sample training dataset......Page 148
Training a random sample of the training dataset......Page 149
Shuffling as an alternative to random sampling......Page 151
Buckets......Page 153
Access to output results......Page 154
SageMaker notebook......Page 155
Creating a job......Page 156
Running a job......Page 158
Recommended strategy......Page 159
Questions......Page 160
Further reading......Page 161
8 Revolutions & Disruptive Innovations......Page 162
Is AI disruptive?......Page 163
AI is based on mathematical theories that are not new......Page 164
Cloud server power, data volumes, and web sharing of the early 21st century started to make AI disruptive......Page 165
Inventions versus innovations......Page 166
Where to start?......Page 167
Getting started......Page 168
The header......Page 169
Implementing Google\'s translation service ......Page 170
Google Translate from a linguist\'s perspective......Page 171
Lexical field theory......Page 172
Jargon......Page 173
Translating is not just translating but interpreting......Page 174
How to check a translation......Page 175
AI as a new frontier......Page 176
Lexical field and polysemy......Page 177
Exploring the frontier – the program......Page 179
k-nearest neighbor algorithm......Page 180
The KNN algorithm......Page 181
The knn_polysemy.py program......Page 183
Implementing the KNN compressed function in Google_Translate_Customized.py......Page 185
Conclusions on the Google Translate customized experiment......Page 193
Summary......Page 194
Questions......Page 195
Further reading......Page 196
9 Getting Neurons to work......Page 197
Technical requirements......Page 198
Defining a CNN......Page 199
Initializing the CNN......Page 201
Kernel......Page 202
Intuitive approach......Page 203
Developers\' approach......Page 204
Mathematical approach......Page 205
Shape......Page 206
ReLu......Page 207
Pooling......Page 209
Next convolution and pooling layer......Page 210
Dense layers......Page 211
Dense activation functions......Page 212
The goal......Page 213
Quadratic loss function......Page 214
Binary cross-entropy......Page 215
Adam optimizer......Page 216
Training dataset......Page 217
Loading the data......Page 218
Data augmentation......Page 219
Training with the classifier......Page 220
Saving the model......Page 221
Summary......Page 222
Further reading and references......Page 223
10 Applying Biomimicking to AI......Page 224
Technical requirements......Page 225
TensorFlow, an open source machine learning framework......Page 226
Does deep learning represent our brain or our mind?......Page 227
Input data......Page 229
Layer 1 – managing the inputs to the network......Page 231
Weights, biases, and preactivation......Page 232
Displaying the details of the activation function through the preactivation process......Page 235
The activation function of Layer 1......Page 237
Dropout and Layer 2......Page 238
Layer 2......Page 239
Correct prediction......Page 240
accuracy......Page 241
Cross-entropy......Page 243
Training......Page 244
Optimizing speed with Google\'s Tensor Processing Unit......Page 245
Summary......Page 248
Further reading......Page 249
11 Conceptual Representation Learning......Page 250
Technical requirements......Page 251
Inductive thinking......Page 252
The problem AI needs to solve......Page 253
Loading the model to optimize training......Page 255
Loading the model to use it......Page 258
Using transfer learning to be profitable or see a project stopped......Page 261
Applying the model......Page 262
Making the model profitable by using it for another problem......Page 263
The trained models used in this section......Page 264
GAP – loaded or unloaded......Page 265
GAP – jammed or open lanes......Page 268
Generalizing the Γ(gap conceptual dataset)......Page 270
Generative adversarial networks......Page 271
Generating conceptual representations......Page 272
The use of autoencoders......Page 273
The curse of dimensionality ......Page 274
Scheduling and blockchains......Page 275
Chatbots......Page 276
Summary......Page 277
Further reading......Page 278
12 Automated Planning & Scheduling......Page 279
Technical requirements......Page 280
Planning and scheduling today and tomorrow......Page 281
A real-time manufacturing revolution......Page 282
An apparel manufacturing process......Page 286
Training the CRLMM......Page 288
Food conveyor belt processing – positive pγ and negative nγ gaps......Page 289
Apparel conveyor belt processing – undetermined gaps......Page 290
The beginning of an abstract notion of gaps......Page 291
Modifying the hyperparameters......Page 293
Running a prediction program......Page 294
Building the DQN-CRLMM......Page 295
Implementing a CNN-CRLMM to detect gaps and optimize......Page 296
Q-Learning – MDP ......Page 297
The input is a neutral reward matrix......Page 298
The standard output of the MDP function......Page 299
A graph interpretation of the MDP output matrix......Page 300
The optimizer......Page 301
Implementing Z – squashing the MDP result matrix......Page 302
Implementing Z – squashing the vertex weights vector......Page 303
Finding the main target for the MDP function......Page 305
Circular DQN-CRLMM – a stream-like system that never starts nor ends......Page 307
Further reading......Page 312
13 AI & Internet of Things (IoT)......Page 313
Technical requirements......Page 314
Setting up the DQN-CRLMM model......Page 315
The dataset......Page 316
Training and testing the model......Page 317
Motivation – using an SVM to increase safety levels......Page 318
Definition of a support vector machine......Page 320
Python function ......Page 322
Finding a parking space......Page 324
Deciding how to get to the parking lot......Page 327
Support vector machine......Page 328
The itinerary graph......Page 330
The weight vector......Page 331
Questions......Page 332
References......Page 333
14 Optimizing Blockchains with AI......Page 334
Mining bitcoins......Page 335
Using cryptocurrency ......Page 336
Using blockchains......Page 337
Creating a block......Page 339
Exploring the blocks......Page 340
A naive Bayes example......Page 341
The blockchain anticipation novelty......Page 343
Step 1 the dataset......Page 344
Step 2 frequency......Page 345
Step 4 naive Bayes equation......Page 346
Gaussian naive Bayes......Page 347
The Python program......Page 348
Summary......Page 350
Questions......Page 351
Further reading......Page 352
15 Cognitive NLP Chatbots......Page 353
Intents......Page 354
Testing the subsets......Page 356
Entities......Page 357
Dialog flow......Page 359
Scripting and building up the model......Page 360
A cognitive chatbot service......Page 362
A cognitive dataset......Page 363
Cognitive natural language processing......Page 364
Activating an image + word cognitive chat......Page 366
Solving the problem ......Page 368
Implementation......Page 369
Questions......Page 370
Further reading......Page 371
16 Improve Emotional Intelligence Deficiencies of Chatbots......Page 372
Technical requirements......Page 373
Building a mind ......Page 374
How to read this chapter......Page 375
Restricted Boltzmann Machines......Page 376
The connections between visible and hidden units......Page 377
Energy-based models......Page 379
Running the epochs and analyzing the results......Page 380
Parsing the datasets......Page 382
Profiling with images......Page 384
RNN for data augmentation......Page 386
RNNs and LSTMs......Page 387
RNN, LSTM, and vanishing gradients......Page 388
Step 2 – running an RNN......Page 389
Word embedding......Page 390
The Word2vec model......Page 391
Principal component analysis......Page 394
Variance......Page 395
Eigenvalues and eigenvectors......Page 397
TensorBoard Projector......Page 399
Using Jacobian matrices......Page 400
Questions......Page 401
Further reading......Page 402
17 Quantum Computers that think......Page 403
Technical requirements......Page 404
Quantum computer speed......Page 405
Representing a qubit......Page 408
The position of a qubit ......Page 409
Radians, degrees, and rotations......Page 410
Bloch sphere......Page 411
Quantum gates with Quirk......Page 412
A quantum computer score with Quirk......Page 414
A quantum computer score with IBM Q......Page 415
Representing our mind\'s concepts ......Page 418
Expanding MindX\'s conceptual representations......Page 420
Positive thinking......Page 421
Negative thinking......Page 422
Distances......Page 424
The embedding program......Page 425
The MindX experiment......Page 427
Transformation Functions – the situation function......Page 428
Transformation functions – the quantum function......Page 430
Creating and running the score......Page 431
Using the output......Page 432
IBM Watson and scripts......Page 433
Summary......Page 434
Further reading......Page 435
Chapter 1 – Become an Adaptive Thinker......Page 436
Chapter 2 – Think like a Machine......Page 438
Chapter 3 – Apply Machine Thinking to a Human Problem......Page 439
Chapter 4 – Become an Unconventional Innovator......Page 440
Chapter 5 – Manage the Power of Machine Learning and Deep Learning......Page 442
Chapter 6 – Don\'t Get Lost in Techniques, Focus on Optimizing Your Solutions......Page 443
Chapter 7 – When and How to Use Artificial Intelligence......Page 445
Chapter 8 – Revolutions Designed for Some Corporations and Disruptive Innovations for Small to Large Companies......Page 447
Chapter 9 – Getting Your Neurons to Work......Page 449
Chapter 10 – Applying Biomimicking to AI......Page 451
Chapter 11 – Conceptual Representation Learning......Page 453
Chapter 12 – Automated Planning and Scheduling......Page 455
Chapter 13 – AI and the Internet of Things......Page 456
Chapter 14 – Optimizing Blockchains with AI......Page 457
Chapter 15 – Cognitive NLP Chatbots......Page 458
Chapter 16 – Improve the Emotional Intelligence Deficiencies of Chatbots......Page 460
Chapter 17 – Quantum Computers That Think......Page 461
Index......Page 465