Index
A
- A/B testing, A/B Testing, A/B Testing Hosted Models-Using the Experiment in Code
- accountability, Accountability and Explainability
- action space, Configure the action space-Configure the action space
- activation functions, Activation Functions
- actor-critic approach, Policy based or value based—why not both?
- Adam optimizer, Set Training Parameters
- Advanced Driver Assistance Systems (ADAS), Autonomous Cars
- Advanced Virtual RISC (AVR), Arduino
- Advantage function, Policy based or value based—why not both?
- AI Driving Olympics (AI-DO), AI Driving Olympics
- Airbnb, Airbnb’s Photo Classification
- AirSim, Deep Learning, Autonomous Driving, and the Data Problem, Setup and Requirements
- AlchemyVision, IBM Watson Visual Recognition
- Algorithmia, Algorithmia
- Alibaba, Speaker Recognition by Alibaba
- alpha users, The Life Cycle of a Food Classifier App
- Amazon Rekognition, Amazon Rekognition
- Amazon SageMaker, A Brief Introduction to Reinforcement Learning
- AMD GPUs, Installation
- Anaconda Python, Installation
- Android Studio, Building a Real-Time Object Recognition App
- annotations, Data
- Annoy (Approximate Nearest Neighbors Oh Yeah), Annoy
- Apollo simulation, Deep Learning, Autonomous Driving, and the Data Problem
- app size, reducing, Reducing App Size-Use Create ML
- Apple’s machine learning architecture, Apple’s Machine Learning Architecture-ML Performance Primitives, Approach 2: Use Create ML
- approximate nearest neighbor (ANN), Approximate Nearest-Neighbor Benchmark
- architecture, Model Architecture
- Arduino, Arduino
- artificial general intelligence, Summary
- artificial intelligence (AI)
- browser-based, AI in the Browser with TensorFlow.js and ml5.js-Summary
- definition of, What Is AI?
- on embedded devices, Becoming a Maker: Exploring Embedded AI at the Edge-Summary
- examples of, An Apology, Motivating Examples-Motivating Examples (see also case studies)
- frequently asked questions, Frequently Asked Questions-Frequently Asked Questions
- history of, A Brief History of AI-How Deep Learning Became a Thing
- versus human accuracy, ImageNet Dataset
- introduction to, The Real Introduction-What Is AI?
- mobile-based, Real-Time Object Classification on iOS with Core ML-Summary, Shazam for Food: Developing Android Apps with TensorFlow Lite and ML Kit-Summary
- perfect deep learning solution, Recipe for the Perfect Deep Learning Solution-Hardware, What’s in the Picture: Image Classification with Keras
- resources for learning about, Education and Exploration, Further Exploration
- responsible AI, Responsible AI-Privacy
- serving prediction requests, Landscape of Serving AI Predictions
- tools, tips, and tricks, Practical Tools, Tips, and Tricks-One Last Question
- aspect ratio, Effect of Change in Aspect Ratio on Transfer Learning
- asynchronous communication, Prefetch Data
- augmentations, AutoAugment
- AutoAugment, Tools of the Trade, AutoAugment
- AutoKeras, Tools of the Trade, AutoKeras
- Automatic Mixed Precision, Use Automatic Mixed Precision
- automatic tuning, AutoAugment
- autonomous driving
- data collection necessary for, Deep Learning, Autonomous Driving, and the Data Problem
- data exploration and preparation, Data Exploration and Preparation-Dataset Imbalance and Driving Strategies
- deep learning applications in, Deep Learning, Autonomous Driving, and the Data Problem
- further exploration, Further Exploration-Reinforcement Learning
- history of, A Glimmer of Hope, A Brief History of Autonomous Driving
- introduction to, Simulating a Self-Driving Car Using End-to-End Deep Learning with Keras
- landscape of, Why Learn Reinforcement Learning with an Autonomous Car?
- model deployment, Deploying Our Autonomous Driving Model-Deploying Our Autonomous Driving Model
- model training, Training Our Autonomous Driving Model-Callbacks
- object detection in, Autonomous Cars
- real-world applications of, Further Exploration
- SAE levels of driving automation, Simulating a Self-Driving Car Using End-to-End Deep Learning with Keras
- simulation and, Deep Learning, Autonomous Driving, and the Data Problem, Practical Deep Reinforcement Learning with DeepRacer, Sim2Real transfer
- steering through simulated environments, The “Hello, World!” of Autonomous Driving: Steering Through a Simulated Environment-Setup and Requirements
- Autonomous Driving Cookbook, Deep Learning, Autonomous Driving, and the Data Problem, Reinforcement Learning
- availability, High Availability, Real-Time Object Classification on iOS with Core ML
- AWS DeepRacer
- for advanced developers, Advanced AWS DeepRacer
- AI Driving Olympics, AI Driving Olympics
- creating AWS accounts, Building Our First Reinforcement Learning
- DeepRacer league, DeepRacer League
- DIY Robocars, DIY Robocars
- improving learning models, Step 5: Improving Reinforcement Learning Models-Racing the AWS DeepRacer Car
- model creation, Step 1: Create Model
- model evaluation, Step 4: Evaluating the Performance of the Model
- model training, Step 3: Model Training
- motivation behind, A Brief Introduction to Reinforcement Learning
- racing the car, Racing the AWS DeepRacer Car-Sim2Real transfer
- reinforcement learning algorithm, Reinforcement Learning Algorithm in AWS DeepRacer
- reinforcement learning in action, How Does a Reinforcement Learning System Learn?
- reinforcement learning summary, Deep Reinforcement Learning Summary with DeepRacer as an Example
- reinforcement learning terminology, Practical Deep Reinforcement Learning with DeepRacer
- reinforcement learning theory, Reinforcement Learning Theory
- reward function in, Practical Deep Reinforcement Learning with DeepRacer
- Roborace, Roborace
- training configuration, Step 2: Configure Training-Configure stop conditions
B
- backpropagation, A Glimmer of Hope, Batch Size, Backpropagation
- batch size
- Bayesian Optimization, Training
- BDD100K, Data
- benchmarks, Examine Industry Benchmarks
- bias, Bias, Class Activation Maps, Image Captioning, Bias-Bias
- binary classification, Binary classification
- Binder, Installation
- Bing Image Search API, Approach 3: Web Scraper Using Bing Image Search API
- blind and low-vision community, Face Detection in Seeing AI
- bottleneck features, Image Similarity
- browser-based AI
- benchmarking and practical considerations, Benchmarking and Practical Considerations
- case studies, Case Studies-GAN Lab
- introduction to, AI in the Browser with TensorFlow.js and ml5.js
- JavaScript-based ML libraries, JavaScript-Based Machine Learning Libraries: A Brief History-TensorFlow.js
- ml5.js, ml5.js
- model conversion, Model Conversion for the Browser
- pix2pix, pix2pix-pix2pix
- PoseNet, PoseNet
- pretrained models using TensorFlow.js, Running Pretrained Models Using TensorFlow.js
- TensorFlow.js architecture, TensorFlow.js Architecture
- training, Training in the Browser-GPU Utilization
- brute-force algorithm, Brute Force
C
- caching, Cache Data
- callbacks, Callbacks
- Carla, Deep Learning, Autonomous Driving, and the Data Problem
- case studies
- browser-based AI
- computer vision
- embedded AI devices
- mobile-based AI
- face filter–like functionality, Face Contours in ML Kit
- HomeCourt, HomeCourt
- InstaSaber, InstaSaber + YoPuppet
- Lose It!, Lose It!
- Magic Sudoku, Magic Sudoku
- portrait mode on Pixel 3 phones, Portrait Mode on Pixel 3 Phones
- Seeing AI, Seeing AI
- speaker recognition by Alibaba, Speaker Recognition by Alibaba
- video segmentation, Real-Time Video Segmentation in YouTube Stories
- YoPuppet, InstaSaber + YoPuppet
- object detection
- reverse image search
- categorical_crossentropy loss function, Start Training
- category unawareness, ImageNet Dataset
- celebrity doppelgangers, Celebrity Doppelgangers
- channels, Predicting an Image’s Category
- chroma keying technique, Real-Time Video Segmentation in YouTube Stories
- Clarifai, Clarifai
- class activation maps (heat maps), Class Activation Maps-Class Activation Maps, Heatmap visualization
- classification problems (see also custom classifiers; image classification)
- classification with localization, Localization
- Cloud AutoML, Building a Custom Detector Without Any Code
- cloud computing (see also inference serving)
- comparing custom classification APIs, Comparing Custom Classification APIs
- example scenario using, Cloud APIs for Computer Vision: Up and Running in 15 Minutes
- getting up and running with, Getting Up and Running with Cloud APIs
- hosted models, Hosted Models-Using the Experiment in Code
- performance tuning for cloud APIs, Performance Tuning for Cloud APIs-Effect of Resizing on OCR APIs
- prebuilt object detection APIs, Invoking Prebuilt Cloud-Based Object Detection APIs-Invoking Prebuilt Cloud-Based Object Detection APIs
- reduced costs of mobile-based AI, Real-Time Object Classification on iOS with Core ML
- training custom classifiers, Training Our Own Custom Classifier-Training Our Own Custom Classifier
- troubleshooting classifiers, Top Reasons Why Our Classifier Does Not Work Satisfactorily
- visual recognition APIs available, The Landscape of Visual Recognition APIs-What’s unique about this API?
- visual recognition APIs comparison, Comparing Visual Recognition APIs-Bias
- co-occurrences, Class Activation Maps, Approach 2: Fatkun Chrome Browser Extension
- COCO (Microsoft Common Objects in COntext), Datasets, Image Captioning
- code examples
- code-free training, A continuously evolving landscape, Building a Custom Detector Without Any Code
- Cognata, Deep Learning, Autonomous Driving, and the Data Problem
- Cognitive Services, Microsoft Cognitive Services, Invoking Prebuilt Cloud-Based Object Detection APIs
- CometML, Training
- components, Predicting an Image’s Category
- compression, Performance Tuning for Cloud APIs
- computer vision
- case studies, Case Studies-InDro Robotics
- comparing visual recognition APIs, Comparing Visual Recognition APIs-Bias
- custom classifer training, Training Our Own Custom Classifier-Training Our Own Custom Classifier
- custom classification APIs comparison, Comparing Custom Classification APIs-Comparing Custom Classification APIs
- example scenario using, Cloud APIs for Computer Vision: Up and Running in 15 Minutes-Cloud APIs for Computer Vision: Up and Running in 15 Minutes
- performance tuning for cloud APIs, Performance Tuning for Cloud APIs-Effect of Resizing on OCR APIs
- time estimation for achieving, How Deep Learning Became a Thing
- troubleshooting classifiers, Top Reasons Why Our Classifier Does Not Work Satisfactorily
- types of computer-vision tasks, Types of Computer-Vision Tasks-Approaches to Object Detection
- using cloud APIs, Getting Up and Running with Cloud APIs-Getting Up and Running with Cloud APIs
- visual recognition APIs available, The Landscape of Visual Recognition APIs-What’s unique about this API?
- concept drift, Model Versioning
- conditional GANs , pix2pix
- contact information for this book, How to Contact Us
- continuous action space, Configure the action space
- contrastive loss function, Siamese Networks for One-Shot Face Verification
- conventions used in this book, Conventions Used in This Book
- ConvNetJS, ConvNetJS
- convolution filters, Convolution
- Convolutional Neural Networks (CNNs)
- accuracy and feature lengths, Length of Feature Vectors
- creating diagrams of, Model
- end-to-end learning example pipeline, End-to-End Deep Learning Example Pipeline-Basic Custom Network Pipeline
- hyperparameter effect on accuracy, How Hyperparameters Affect Accuracy-Effect of Change in Aspect Ratio on Transfer Learning
- image classification using, Predicting an Image’s Category, A Shallow Dive into Convolutional Neural Networks
- improving model accuracy, From Novice to Master Predictor: Maximizing Convolutional Neural Network Accuracy-From Novice to Master Predictor: Maximizing Convolutional Neural Network Accuracy
- progress over past decade, How Deep Learning Became a Thing
- resources for learning about, Further Reading-Further Reading
- structure of, A Shallow Dive into Convolutional Neural Networks, Structure of a CNN
- techniques for ML experimentation, Common Techniques for Machine Learning Experimentation-Reproducible Experiments
- tools for automated tuning, Tools to Automate Tuning for Maximum Accuracy-AutoKeras
- tools to reduce code and effort, Tools of the Trade-tf-explain
- transfer learning using, Transfer Learning
- versus transfer learning, Approach 2: Use Create ML
- Coral USB accelerator, Google Coral USB Accelerator, Speeding Up with the Google Coral USB Accelerator
- Core ML
- alternatives to, Alternatives to Core ML
- building real-time object recognition apps, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- conversion to, Conversion to Core ML, Model Conversion Using Core ML Tools
- dynamic model deployment, Dynamic Model Deployment
- ecosystem, Apple’s Machine Learning Architecture-ML Performance Primitives
- history of, A Brief History of Core ML
- measuring energy consumption, Measuring Energy Impact-Benchmarking Load
- Not Hotdog app, Not Hotdog on iOS with Core ML and Create ML-Summary
- on-device training, On-Device Training
- performance analysis, Performance Analysis-Benchmarking Models on iPhones
- reducing app size, Reducing App Size-Use Create ML
- counting people, Crowd Counting
- Create ML, Use Create ML, Approach 2: Use Create ML-Approach 2: Use Create ML, Building a Custom Detector Without Any Code
- crowd counting, Crowd Counting
- cucumber sorting device, Cucumber Sorter
- curse of dimensionality, Reducing Feature-Length with PCA
- custom classifiers
- custom detectors, building without code, Building a Custom Detector Without Any Code-Building a Custom Detector Without Any Code
- (see also object detection)
- CustomVision.ai, Approach 1: Use Web UI-based Tools-Approach 1: Use Web UI-based Tools, Model Conversion Using Core ML Tools, Building a Custom Detector Without Any Code
D
- data augmentation, Data Augmentation-Data Augmentation, Data Augmentation, Data Augmentation
- data caching, Cache Data
- data generators, Drive Data Generator
- data inspection, Data Inspection
- data preparation (TensorFlow)
- data reading (TensorFlow)
- data size
- datamash tool, nvidia-smi
- dataset imbalance, Dataset Imbalance and Driving Strategies
- Dataset.cache() function, Cache Data
- DatasetList.com, Data
- datasets
- annotating data, Data
- approaches to collecting data, Collecting Data-Approach 3: Web Scraper Using Bing Image Search API
- collecting hundreds of images, Data, Approach 2: Fatkun Chrome Browser Extension
- downloading data in custom ways, Data
- finding larger, Data
- finding prebuilt, Data
- finding smaller, Data
- for mobile AI app development, How Do I Collect Initial Data?, Data Collection
- Google Dataset Search, Data
- ImageNet dataset, Predicting an Image’s Category, ImageNet Dataset-ImageNet Dataset
- largest labeled, Data
- MS COCO (Microsoft Common Objects in COntext), Datasets, Image Captioning
- need for labeled, Datasets
- of negative classes, Data
- NIST and MNIST datasets, A Glimmer of Hope
- publicly available, Datasets, Approach 1: Find or Collect a Dataset
- reading popular, Data
- ready-to-use, TensorFlow Datasets
- scraping images using Bing Image Search API, Approach 3: Web Scraper Using Bing Image Search API
- scraping images using command line, Data
- sequential data, Training on Sequential Data
- splitting into train, validation, and test, Breaking the Data: Train, Validation, Test
- synthetic datasets, Which Library Should I Use?, Data
- TensorFlow Datasets, TensorFlow Datasets, Use TensorFlow Datasets, Data
- for unique problems, Data
- versioning tools for, Data
- video datasets, Data
- DAVE-2 system, Further Exploration
- DAWNBench benchmarks, Examine Industry Benchmarks
- debugging, Training
- deep learning
- Deeplearning.ai, Education and Exploration
- detection, Detection (see also facial recognition; object detection)
- Digital Data Divide, Data
- discount factor (γ), Delayed rewards and discount factor (γ)
- discrete action space, Configure the action space
- Distributed Deep Reinforcement Learning for Autonomous Driving, Reinforcement Learning
- distributed training, Distribute Training
- DIY Robocars, DIY Robocars
- Docker, Installation
- driving automation, Simulating a Self-Driving Car Using End-to-End Deep Learning with Keras
- (see also autonomous driving)
E
- eager execution, Use tf.function
- early stopping, Early Stopping, Callbacks
- edge devices (see embedded AI devices)
- embedded AI devices
- embeddings
- encryption, Privacy
- energy consumption, of mobile apps, Measuring Energy Impact-Benchmarking Load
- episodes, Practical Deep Reinforcement Learning with DeepRacer
- Euclidean distance, Similarity Search, Reducing Feature-Length with PCA
- experience buffer, Practical Deep Reinforcement Learning with DeepRacer
- experimental optimizations, Turn on Experimental Optimizations
- explainability, Accountability and Explainability
F
- facial recognition, Siamese Networks for One-Shot Face Verification, Face Contours in ML Kit, Face Detection in Seeing AI
- failure handling, Failure Handling
- fairing, Fairing
- Faiss, Faiss
- Fast.ai, Education and Exploration
- FastProgress, Training
- Fatkun Batch Download Image, Data, Approach 2: Fatkun Chrome Browser Extension
- feature extraction, Feature Extraction-Feature Extraction, Feature Extraction
- feature scaling, Predicting an Image’s Category
- feature vectors
- federated learning, Federated Learning
- Field Programmable Gate Arrays (FPGAs), FPGA + PYNQ
- filter operations, Turn on Experimental Optimizations
- filter_fusion option, Filter fusion
- fine tuning, Fine Tuning-How Much to Fine Tune, Improving Accuracy with Fine Tuning-Fine Tuning Without Fully Connected Layers
- fine-grained recognition, ImageNet Dataset
- Firebase, ML Kit, ML Kit + Firebase
- fixed round-robin order, Enable Nondeterministic Ordering
- Flask, Flask: Build Your Own Server-Cons of Using Flask
- Flickr, Flickr
- food classifier apps
- building real-time object recognition apps, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- challenges of, Shazam for Food: Developing Android Apps with TensorFlow Lite and ML Kit
- life cycle of, The Life Cycle of a Food Classifier App
- life cycle of mobile AI development, A Holistic Look at the Mobile AI App Development Cycle-How Do I Update the Model on My Users’ Phones?
- using ML Kit + Firebase, ML Kit + Firebase-Using the Experiment in Code
- model conversion to TensorFlow Lite, Model Conversion to TensorFlow Lite
- Shazam for Food app, Shazam for Food: Developing Android Apps with TensorFlow Lite and ML Kit
- TensorFlow Lite architecture, TensorFlow Lite Architecture
- TensorFlow Lite on iOS, TensorFlow Lite on iOS
- TensorFlow Lite overview, An Overview of TensorFlow Lite
- tools for, The Life Cycle of a Food Classifier App
- forward pass, Batch Size
- FPGAs (Field Programmable Gate Arrays), FPGA + PYNQ
- Fritz, Fritz, Fritz-Fritz
- fused operations, Use Fused Operations
G
- GAN Lab, GAN Lab
- General Data Protection Regulation (GDPR), Real-Time Object Classification on iOS with Core ML
- generalization, Overtrain, and Then Generalize, Approach 2: Fatkun Chrome Browser Extension
- Generative Adversarial Networks (GANs), pix2pix
- geographic availability, Geographic Availability
- Giphy, Giphy
- Giraffe-Tree problem, Image Captioning
- Google Cloud ML Engine
- Google Cloud Platform (GCP), Uploading a hosted model
- Google Cloud Vision, Google Cloud Vision, Invoking Prebuilt Cloud-Based Object Detection APIs
- Google Coral USB accelerator, Google Coral USB Accelerator, Speeding Up with the Google Coral USB Accelerator
- Google Dataset Search, Data
- Google Edge Tensor Processing Unit (TPU), Google Coral USB Accelerator
- Google Seedbank, Education and Exploration
- Google's AI doodle, AI in the Browser with TensorFlow.js and ml5.js
- GPU persistence, Enable GPU Persistence
- GPU utilization, GPU Utilization
- Graphics Processing Unit (GPU) starvation
- GUI-based model training tools, A continuously evolving landscape
I
- IBM Watson Visual Recognition, IBM Watson Visual Recognition
- image captioning, Image Captioning
- image classification
- building custom classifiers in Keras, Building a Custom Classifier in Keras with Transfer Learning-Analyzing the Results
- category unawareness, ImageNet Dataset
- class activation maps (heat maps), Class Activation Maps-Class Activation Maps
- using CNNs, A Shallow Dive into Convolutional Neural Networks
- desired properties of image classifiers, Desired Properties of an Image Classifier
- fine-grained recognition, ImageNet Dataset
- how images store information, Predicting an Image’s Category
- ImageNet dataset, ImageNet Dataset-ImageNet Dataset
- using Keras, Predicting an Image’s Category-Predicting an Image’s Category
- model zoos, Model Zoos
- preprocessing images, Predicting an Image’s Category
- simple pipeline for, Predicting an Image’s Category
- image labeling APIs, Effect of Resizing on Image Labeling APIs, How Do I Label My Data?
- image saliency, Class Activation Maps
- image segmentation, Image Segmentation
- image similarity, Image Similarity-Image Similarity
- image size, Performance Tuning for Cloud APIs
- ImageN, Data
- ImageNet dataset, Predicting an Image’s Category, ImageNet Dataset-ImageNet Dataset
- ImageNet Large Scale Visual Recognition Challenge (ILSVRC), How Deep Learning Became a Thing, ImageNet Dataset
- ImageNet-Utils, Approach 1: Find or Collect a Dataset
- Imagenette, Data
- iMerit, Data
- implicit bias, Bias
- in-group/out-group bias, Bias
- Inception architectures, Google Cloud Vision
- InDro Robotics, InDro Robotics
- inference serving
- desirable qualities for, Desirable Qualities in a Production-Level Serving System-Support for Multiple Machine Learning Libraries
- frameworks for, Frameworks
- frequently asked questions, Scalable Inference Serving on Cloud with TensorFlow Serving and KubeFlow
- Google Cloud ML Engine, Google Cloud ML Engine: A Managed Cloud AI Serving Stack-Building a Classification API
- KubeFlow, KubeFlow-Installation
- price versus performance, Price Versus Performance Considerations-Cost Analysis of Building Your Own Stack
- server building with Flask, Flask: Build Your Own Server-Cons of Using Flask
- TensorFlow Serving, TensorFlow Serving
- tools, libraries, and cloud services, Landscape of Serving AI Predictions
- inference time, Inference Time
- Inference-as-a-Service, Google Cloud ML Engine: A Managed Cloud AI Serving Stack, Price Versus Performance Considerations
- installation, Installation-Installation
- InstaLooter, Data
- instance retrieval, Building a Reverse Image Search Engine: Understanding Embeddings
- instance-level segmentation, Instance-level segmentation
- InstaSaber, InstaSaber + YoPuppet
- Instruction Set Architecture (ISA), FPGAs
- Intel Movidius Neural Compute Stick 2, Intel Movidius Neural Compute Stick
- Interactive Telecommunications Program (ITP), PoseNet
- interactivity, Real-Time Object Classification on iOS with Core ML
- Intersection over Union (IoU), Intersection over Union
- iOS models (see Core ML)
K
- Keras
- benefits of, A continuously evolving landscape, Introducing Keras
- callbacks in, Callbacks
- conversion to Core ML, Conversion from Keras, Model Conversion Using Core ML Tools
- custom classifiers using transfer learning, Building a Custom Classifier in Keras with Transfer Learning-Analyzing the Results
- deploying models to Flask, Deploying a Keras Model to Flask
- fine tuning object classifiers with, Approach 3: Fine Tuning Using Keras
- history of, Keras
- image classification pipeline, Predicting an Image’s Category-Predicting an Image’s Category
- key function for feature extraction, Feature Extraction
- model accuracy, Investigating the Model-ImageNet Dataset
- pretrained models in, Model Zoos
- Keras functions
- Keras Tuner, Tools of the Trade, Keras Tuner-Keras Tuner, Training
- Keras.js, Keras.js
- keypoint detection, PoseNet
- Knock Knock, Training
- KubeFlow
- Kubernetes, KubeFlow
L
- labeling, Datasets, Data, How Do I Label My Data?, Labeling the Data
- latency, Low Latency, Real-Time Object Classification on iOS with Core ML
- layers, effect on accuracy, Effect of Number of Layers Fine-Tuned in Transfer Learning
- lazy learning, Similarity Search
- learning rate
- localization, Localization
- Lose It!, Lose It!
- loss function, Set Training Parameters, Siamese Networks for One-Shot Face Verification
- low latency, Low Latency
M
- machine learning (ML)
- Apple’s machine learning architecture, Apple’s Machine Learning Architecture-ML Performance Primitives, Approach 2: Use Create ML
- common techniques for, Common Techniques for Machine Learning Experimentation-Reproducible Experiments
- definition of, What Is AI?
- JavaScript-based ML libraries, JavaScript-Based Machine Learning Libraries: A Brief History-TensorFlow.js
- overview of, Machine Learning
- resources for learning about, Education and Exploration, Further Exploration
- support for multiple ML libraries, Support for Multiple Machine Learning Libraries
- terminology used in, Bias
- Magic Sudoku, Magic Sudoku
- magnitude-based weight pruning, Prune the Model
- map operations, Turn on Experimental Optimizations
- map_and_filter_fusion option, Map and filter fusion
- map_func function, Parallelize I/O and Processing
- map_fusion option, Map fusion
- Markov decision process (MDP), The Markov decision process
- masks, Segmentation
- Mean Average Precision (mAP), Mean Average Precision
- Metacar, Metacar
- metric learning, Siamese Networks for One-Shot Face Verification
- metrics, Set Training Parameters
- MicroController Units (MCUs), TensorFlow Lite Architecture, Arduino
- Microsoft Cognitive Services, Microsoft Cognitive Services, Invoking Prebuilt Cloud-Based Object Detection APIs
- minibatches, Use Larger Batch Size
- MirroredStrategy, Distribute Training
- ML Kit
- ml5.js, ml5.js
- MLPerf, Examine Industry Benchmarks
- MMdnn, Model
- MNIST dataset, A Glimmer of Hope
- mobile-based AI
- Apple’s machine learning architecture, Apple’s Machine Learning Architecture-ML Performance Primitives, Approach 2: Use Create ML
- building real-time object recognition apps, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- case studies, Case Studies-InstaSaber + YoPuppet, Case Studies-Real-Time Video Segmentation in YouTube Stories
- challenges of, Real-Time Object Classification on iOS with Core ML
- conversion to Core ML, Conversion to Core ML
- Core ML alternatives, Alternatives to Core ML
- Core ML history, A Brief History of Core ML
- development life cycle for, The Development Life Cycle for Artificial Intelligence on Mobile
- dynamic model deployment, Dynamic Model Deployment
- frequently asked questions, Real-Time Object Classification on iOS with Core ML, A Holistic Look at the Mobile AI App Development Cycle-How Do I Update the Model on My Users’ Phones?
- Fritz end-to-end solution for, Fritz-Fritz
- measuring energy consumption, Measuring Energy Impact-Benchmarking Load
- on-device training, On-Device Training
- performance analysis, Performance Analysis-Benchmarking Models on iPhones
- reducing app size, Reducing App Size-Use Create ML
- self-evolving model, The Self-Evolving Model
- MobileNet, Training
- model accuracy
- model architecture, Model Architecture
- model cards, Model
- model compression, Prune the Model
- model definition, Model Definition
- model inference (TensorFlow)
- model size, Model Size
- model testing, Test the Model
- model tips and tricks, Model
- model training
- model versioning, Model Versioning
- model zoos, Model Zoos
- model-free reinforcement learning, Model free versus model based
- ModelDepot.io, Model
- ModelZoo.co, Model
- monitoring, Monitoring
- Movidius Neural Compute Stick 2, Intel Movidius Neural Compute Stick
- MS COCO (Microsoft Common Objects in COntext), Datasets, Image Captioning
- MS Paint, Model
- multiclass classification, Multiclass classification
- multilayer neural networks, A Glimmer of Hope
- multiple libraries support, Support for Multiple Machine Learning Libraries
- multiples of eight, Find the Optimal Learning Rate
- MultiWorkerMirroredStrategy, Distribute Training
- Myriad VPU, Intel Movidius Neural Compute Stick
N
- National Institute of Standards and Technology (NIST), A Glimmer of Hope
- NavLab, A Glimmer of Hope, A Brief History of Autonomous Driving
- nearest-neighbor approach
- negative classes, Data, Approach 2: Fatkun Chrome Browser Extension
- Neighborhood Graph and Tree (NGT), NGT
- Netron, Model
- Neural Architecture Search (NAS), Tools of the Trade, AutoKeras, Training
- neural networks (see also Convolutional Neural Networks)
- New York Times, The New York Times, Scalability
- NIST dataset, A Glimmer of Hope
- NN-SVG, Model
- Non-Maximum Suppression (NMS), Non-Maximum Suppression
- nondeterministic ordering, Enable Nondeterministic Ordering
- normalization, Predicting an Image’s Category
- Not Hotdog app
- number of nines, High Availability
- NVIDIA Data Loading Library (DALI) , tf.image built-in augmentations
- NVIDIA GPU Persistence Daemon, Enable GPU Persistence
- NVIDIA Graphics Processing Units (GPUs), Installation
- NVIDIA Jetson Nano, NVIDIA Jetson Nano, Port to NVIDIA Jetson Nano
- nvidia-smi (NVIDIA System Management Interface), nvidia-smi, Training
O
- object classification (see also food classifier apps)
- building iOS app, Building the iOS App
- building real-time apps for, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- challenges of on mobile devices, Real-Time Object Classification on iOS with Core ML, Shazam for Food: Developing Android Apps with TensorFlow Lite and ML Kit
- collecting data, Collecting Data-Approach 3: Web Scraper Using Bing Image Search API
- in ML Kit, Object Classification in ML Kit
- model conversion using Core ML, Model Conversion Using Core ML Tools
- model training, Training Our Model-Approach 3: Fine Tuning Using Keras
- overview of steps, Not Hotdog on iOS with Core ML and Create ML
- real-time apps for, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- object detection
- approaches to, Approaches to Object Detection
- building custom detectors without code, Building a Custom Detector Without Any Code-Building a Custom Detector Without Any Code
- case studies, Case Studies-Autonomous Cars
- evolution of, The Evolution of Object Detection
- image segmentation, Image Segmentation
- inspecting and training models, Inspecting the Model-Training
- key terms in, Key Terms in Object Detection-Non-Maximum Suppression
- model conversion, Model Conversion
- performance considerations, Performance Considerations
- prebuilt cloud-based APIs, Invoking Prebuilt Cloud-Based Object Detection APIs-Invoking Prebuilt Cloud-Based Object Detection APIs
- reusing pretrained models, Reusing a Pretrained Model-Deploying to a Device
- TensorFlow Object Detection API, Using the TensorFlow Object Detection API to Build Custom Models-Preprocessing the Data
- types of computer-vision tasks, Types of Computer-Vision Tasks-Approaches to Object Detection
- object localization, Detection
- object segmentation, Segmentation
- OCR APIs, Effect of Compression on OCR APIs
- OmniEarth, OmniEarth
- one shot learning, Siamese Networks for One-Shot Face Verification
- online resources
- ONNX.js, ONNX.js
- Open Images V4 , Data
- Open Neural Network Exchange (ONNX), A continuously evolving landscape, Model, ONNX.js
- optimizations (TensorFlow Lite)
- optimizations (TensorFlow) (see also TensorFlow performance checklist)
- Automatic Mixed Precision, Use Automatic Mixed Precision
- better hardware, Use Better Hardware
- distributed training, Distribute Training
- eager execution and tf.function, Use tf.function
- experimental optimizations, Turn on Experimental Optimizations
- industry benchmarks, Examine Industry Benchmarks
- larger batch size, Use Larger Batch Size
- multiples of eight, Use Multiples of Eight
- optimized hardware stacks, Install an Optimized Stack for the Hardware
- optimizing parallel CPU threads, Optimize the Number of Parallel CPU Threads
- overtraining and generalization, Overtrain, and Then Generalize
- optimizers
- ordering, nondeterministic, Enable Nondeterministic Ordering
- overfitting, Data Augmentation, Callbacks
- overtraining, Early Stopping, Overtrain, and Then Generalize
- O’Reilly’s Online Learning platform, Education and Exploration
P
- p-hacking, Reproducibility
- PapersWithCode.com, Model, Education and Exploration
- parallel processing, Parallelize CPU Processing, Optimize the Number of Parallel CPU Threads
- PCA (Principle Component Analysis), Reducing Feature-Length with PCA
- perceptrons, Exciting Beginnings, The Cold and Dark Days, Perceptron
- performance tuning, Performance Tuning for Cloud APIs-Effect of Resizing on OCR APIs (see also TensorFlow performance checklist)
- Personally Identifiable Information (PII), Real-Time Object Classification on iOS with Core ML
- Photobucket , Photobucket
- Pilot Parliaments Benchmark (PPB), Bias
- Pinterest, Pinterest
- pipelines
- pix2pix, pix2pix-pix2pix
- Pixel 3 phones, Portrait Mode on Pixel 3 Phones
- pixels, Predicting an Image’s Category
- PlotNeuralNet, Model
- policy-based reinforcement learning, Policy based
- pooling operations, Length of Feature Vectors, Pooling
- portrait mode, Portrait Mode on Pixel 3 Phones
- PoseNet, PoseNet, Squatting for Metro Tickets
- pretrained models
- principal components, Reducing Feature-Length with PCA
- privacy, Privacy, Privacy, Real-Time Object Classification on iOS with Core ML
- profiling, TensorFlow Profiler + TensorBoard
- progress bars, Training
- progressive augmentation, Use progressive augmentation
- progressive resizing, Use progressive resizing
- progressive sampling, Use progressive sampling
- Project Oxford, Microsoft Cognitive Services
- proximal policy optimization (PPO) algorithm, Configure the simulation environment, Reinforcement Learning Algorithm in AWS DeepRacer
- pruning, Prune the Model
- PYNQ platform, FPGA + PYNQ
- PyTorch, PyTorch
R
- randomization, Reproducible Experiments
- Raspberry Pi, Raspberry Pi, Hands-On with the Raspberry Pi
- region of interest (ROI), Identifying the Region of Interest
- reinforcement learning
- AWS DeepRacer algorithm, Reinforcement Learning Algorithm in AWS DeepRacer
- AWS DeepRacer car, Racing the AWS DeepRacer Car-Sim2Real transfer
- AWS DeepRacer example, Practical Deep Reinforcement Learning with DeepRacer-Step 4: Evaluating the Performance of the Model
- crux of, Building an Autonomous Car in Under an Hour: Reinforcement Learning with AWS DeepRacer
- Distributed Deep Reinforcement Learning for Autonomous Driving tutorial , Reinforcement Learning
- further exploration, Further Exploration-Roborace
- improving learning models, Step 5: Improving Reinforcement Learning Models-Racing the AWS DeepRacer Car
- inner workings of, Reinforcement Learning in Action
- introduction to, A Brief Introduction to Reinforcement Learning
- learning with autonomous cars, Why Learn Reinforcement Learning with an Autonomous Car?-Why Learn Reinforcement Learning with an Autonomous Car?
- summary of, Deep Reinforcement Learning Summary with DeepRacer as an Example
- terminology used in, Practical Deep Reinforcement Learning with DeepRacer
- theory of, Reinforcement Learning Theory
- replay buffer, Practical Deep Reinforcement Learning with DeepRacer
- reporting bias, Bias
- reproducibility, Reproducibility, Reproducible Experiments
- ResearchCode, Education and Exploration
- resizing
- ResNet, Training
- ResNet-50, Predicting an Image’s Category, Predicting an Image’s Category
- REST APIs, Making a REST API with Flask
- reverse image search
- case studies, Case Studies-Image Captioning
- feature extraction, Feature Extraction-Feature Extraction
- fine tuning for improved accuracy, Improving Accuracy with Fine Tuning-Fine Tuning Without Fully Connected Layers
- image similarity, Image Similarity-Image Similarity
- improving search speed, Improving the Speed of Similarity Search-Reducing Feature-Length with PCA
- introduction to, Building a Reverse Image Search Engine: Understanding Embeddings
- scaling with nearest-neighbor approach, Scaling Similarity Search with Approximate Nearest Neighbors-Faiss
- Siamese networks for, Siamese Networks for One-Shot Face Verification
- similarity search, Similarity Search-Similarity Search
- visualizing image clusters, Visualizing Image Clusters with t-SNE-Visualizing Image Clusters with t-SNE
- reward function, Practical Deep Reinforcement Learning with DeepRacer, Configure reward function-Configure reward function, How Does a Reinforcement Learning System Learn?
- Roborace, Roborace
- robustness, Robustness
- ROCm stack, Installation
- root-mean-square error (RMSE), Set Training Parameters
- round-robin order, Enable Nondeterministic Ordering
- Runway ML, Education and Exploration
S
- SageMaker reinforcement learning, A Brief Introduction to Reinforcement Learning
- SamaSource, Data
- scalability, Scalability
- scaling horizontally, Distribute Training
- ScrapeStorm.com, Data
- Scrapy.org, Data
- Seeing AI, Seeing AI, Face Detection in Seeing AI
- segmentation, Segmentation
- selection bias, Bias
- self-driving cars, A Glimmer of Hope, Autonomous Cars (see also autonomous driving; reinforcement learning)
- self-evolving model, The Self-Evolving Model
- semantic segmentation, Semantic segmentation
- Semi-Conductor, Semi-Conductor
- semi-supervised learning, Approach 3: Web Scraper Using Bing Image Search API
- sequential data, Training on Sequential Data
- Shazam for Food app, Shazam for Food: Developing Android Apps with TensorFlow Lite and ML Kit
- Siamese networks, Siamese Networks for One-Shot Face Verification
- Sim2Real transfer, Sim2Real transfer
- similarity search
- simulation, Deep Learning, Autonomous Driving, and the Data Problem, Practical Deep Reinforcement Learning with DeepRacer, Sim2Real transfer
- simulation-to-real (sim2real) problem, Why Learn Reinforcement Learning with an Autonomous Car?, Practical Deep Reinforcement Learning with DeepRacer, Sim2Real transfer
- smart refrigerators, Smart Refrigerator
- softmax activation function, Start Training
- SOTAWHAT tool, Education and Exploration
- speaker recognition, Speaker Recognition by Alibaba
- Spotify, Spotify
- squat-tracker app, Squatting for Metro Tickets
- SSD MobileNetV2, Reusing a Pretrained Model
- Staples, Staples
- stop conditions, Configure stop conditions
- success, measuring, How Do I Measure the Success of My Model?
- synthetic datasets, Which Library Should I Use?, Data
T
- t-SNE algorithm, Visualizing Image Clusters with t-SNE-Visualizing Image Clusters with t-SNE
- Teachable Machine, Training in the Browser
- Tencent ML Images, Data
- TensorBoard, TensorBoard-What-If Tool, TensorFlow Profiler + TensorBoard
- TensorFlow
- TensorFlow Datasets, TensorFlow Datasets, Use TensorFlow Datasets, Data
- TensorFlow Debugger (tfdbg), Training
- TensorFlow Embedding projector, Visualizing Image Clusters with t-SNE
- TensorFlow Encrypted, Privacy
- TensorFlow Extended (TFX), TensorFlow Serving
- TensorFlow Hub (tfhub.dev), Model
- TensorFlow Lite
- architecture, TensorFlow Lite Architecture
- building real-time object recognition apps, Building a Real-Time Object Recognition App-Building a Real-Time Object Recognition App
- history of, TensorFlow Lite
- model conversion to, Model Conversion to TensorFlow Lite
- model optimization toolkit, TensorFlow Model Optimization Toolkit
- on iOS, TensorFlow Lite on iOS
- overview of, An Overview of TensorFlow Lite
- performance optimizations, Performance Optimizations
- sample apps in repository, Building a Real-Time Object Recognition App
- using models with ML Kit, ML Kit + Firebase, Custom Models in ML Kit
- TensorFlow models repository, Building a Real-Time Object Recognition App, Obtaining the Model
- TensorFlow Object Detection API
- TensorFlow performance checklist
- TensorFlow Playground, TensorFlow.js
- TensorFlow Serving, TensorFlow Serving
- TensorFlow.js
- TensorSpace, TensorSpace
- tf-coreml, Conversion from TensorFlow
- tf-explain, Tools of the Trade, tf-explain
- tf.data, Use tf.data, Enable Nondeterministic Ordering
- tf.data.experimental.AUTOTUNE, Autotune Parameter Values
- tf.data.experimental.OptimizationOptions, Turn on Experimental Optimizations
- tf.function, Use tf.function
- tf.image, tf.image built-in augmentations
- tf.keras, Tools of the Trade
- tf.test.is_gpu_available(), Training
- tfprof (TensorFlow profiler), TensorFlow Profiler + TensorBoard
- TFRecords, Store as TFRecords, Preprocessing the Data
- timeit command, Brute Force
- tools, tips, and tricks
- training parameters, Set Training Parameters
- Transaction Processing Council (TPC) benchmark, Examine Industry Benchmarks
- transfer learning
- adapting pretrained models to new networks, Model
- adapting pretrained models to new tasks, Adapting Pretrained Models to New Tasks-How Much to Fine Tune
- basic pipeline for, Basic Transfer Learning Pipeline
- building custom classifiers using, Building a Custom Classifier in Keras with Transfer Learning-Analyzing the Results
- Create ML, Approach 2: Use Create ML
- definition of, Datasets
- effect of hyperparmeters on accuracy, How Hyperparameters Affect Accuracy-Effect of Change in Aspect Ratio on Transfer Learning
- fine tuning, Fine Tuning-How Much to Fine Tune
- GUI-based model training tools, A continuously evolving landscape
- overview of, Transfer Learning
- versus training from scratch, Transfer Learning Versus Training from Scratch
- triplet loss function, Siamese Networks for One-Shot Face Verification
V
- validation accuracy, Start Training
- value function (V), Practical Deep Reinforcement Learning with DeepRacer
- value-based reinforcement learning, Value based
- vanilla policy gradient algorithm, Reinforcement Learning Algorithm in AWS DeepRacer
- versioning, Training, Data, Model Versioning
- video datasets, Data
- video segmentation, Real-Time Video Segmentation in YouTube Stories
- visual recognition APIs (see also computer vision)
- VisualData.io, Data
- visualizations
W
- Waymo, Deep Learning, Autonomous Driving, and the Data Problem
- web UI-based tools, Approach 1: Use Web UI-based Tools-Approach 1: Use Web UI-based Tools
- WebScraper.io, Data
- weights, Model Architecture
- Weights and Biases, Training
- What-If Tool, Tools of the Trade, What-If Tool-What-If Tool
- wildlife conservation, Wildlife conservation
..................Content has been hidden....................
You can't read the all page of ebook, please click
here login for view all page.