book

The Deep Learning Architect's Handbook

Name: The Deep Learning Architect's Handbook
Author: Ee Kin Chin
ISBN: 9781803243795

by Ee Kin Chin

December 2023

Intermediate to advanced

516 pages

15h 25m

English

Packt Publishing

Read now

Unlock full access

The Deep Learning Architect’s Handbook
ContributorsAbout the authorAbout the reviewers
Preface
Who this book is forWhat this book coversTo get the most out of this bookDownload the example code filesConventions usedGet in touchShare Your ThoughtsShare Your ThoughtsDownload a free PDF copy of this book
Part 1 – Foundational Methods
Chapter 1: Deep Learning Life Cycle
Technical requirementsUnderstanding the machine learning life cycleStrategizing the construction of a deep learning systemStarting the journeyEvaluating deep learning’s worthinessDefining successPlanning resourcesPreparing dataDeep learning problem typesAcquiring dataMaking sense of data through exploratory data analysis (EDA)Data pre-processingDeveloping deep learning modelsDeep learning model familiesThe model development strategyDelivering model insightsManaging risksEthical and regulatory risksBusiness context mismatchData collection and annotation risksData security riskSummaryFurther reading
Chapter 2: Designing Deep Learning Architectures
Technical requirementsExploring the foundations of neural networks using an MLPUnderstanding neural network gradientsUnderstanding gradient descentImplementing an MLP from scratchImplementing MLP using deep learning frameworksRegularizationDesigning an MLPSummary
Chapter 3: Understanding Convolutional Neural Networks
Technical requirementsUnderstanding the convolutional neural network layerUnderstanding the pooling layerBuilding a CNN architectureDesigning a CNN architecture for practical usageExploring the CNN architecture familiesUnderstanding the ResNet model familyUnderstanding the DenseNet architecture familyUnderstanding the EfficientNet architecture familyUnderstanding small and fast CNN architecture families for small-scale edge devicesUnderstanding SqueezeNetUnderstanding MobileNetUnderstanding MicroNet, the current state-of-the-art architecture for the edgeSummary
Chapter 4: Understanding Recurrent Neural Networks
Technical requirementsUnderstanding LSTMDecoding the forget mechanism of LSTMsDecoding the learn mechanism of LSTMsDecoding the remember mechanism of LSTMsDecoding the “information-using” mechanism of LSTMsBuilding a full LSTM networkUnderstanding GRUDecoding the reset gate of GRUDecoding the update gate of GRUUnderstanding advancements over the standard GRU and LSTM layersDecoding bidirectional RNNAdding peepholes to LSTMsAdding working memory to exceed the peephole connection limitations for LSTMSummary
Chapter 5: Understanding Autoencoders
Technical requirementsDecoding the standard autoencoderExploring autoencoder variationsBuilding a CNN autoencoderSummary
Chapter 6: Understanding Neural Network Transformers
Exploring neural network transformersDecoding the original transformer architecture holisticallyUncovering transformer improvements using only the encoderImproving the encoder only pre-training tasks and objectivesImproving the encoder-only transformer’s architectural compactness and efficiencyImproving the encoder-only transformers’ core functional architectureUncovering encoder-only transformers’ adaptations to other data modalitiesUncovering transformer improvements using only the decoderDiving into the GPT model familyDiving into the XLNet modelDiscussing additional advancements for a decoder-only transformer modelSummary
Chapter 7: Deep Neural Architecture Search
Technical requirementsUnderstanding the big picture of NASUnderstanding general hyperparameter search-based NASSearching neural architectures by using successive halvingSearching neural architectures by using HyperbandSearching neural architectures by using Bayesian hyperparameter optimizationUnderstanding RL-based NASUnderstanding founding NAS based on RLUnderstanding ENASUnderstanding MNASSummarizing NAS with RL methodsUnderstanding non-RL-based NASUnderstanding path elimination-based NASUnderstanding progressive growth-based NASSummary

Chapter 8: Exploring Supervised Deep Learning
Technical requirementsExploring supervised use cases and problem typesImplementing neural network layers for foundational problem typesImplementing the binary classification layerImplementing the multiclass classification layerImplementing a regression layerImplementing representation layersTraining supervised deep learning models effectivelyPreparing the data for DL trainingConfiguring and tuning DL hyperparametersExecuting, visualizing, tracking, and comparing experimentsExploring model-building tipsExploring general techniques to realize and improve supervised deep learning based solutionsBreaking down the multitask paradigm in supervised deep learningMultitask pipelinesTLMultiple objective learningMultimodal NN trainingSummary
Chapter 9: Exploring Unsupervised Deep Learning
Technical requirementsExploring unsupervised deep learning applicationsCreating pretrained network weights for downstream tasksCreating general representations through unsupervised deep learningExploring zero-shot learningExploring the dimensionality reduction component of unsupervised deep learningDetecting anomalies in external dataSummary
Part 2 – Multimodal Model Insights
Chapter 10: Exploring Model Evaluation Methods
Technical requirementsExploring the different model evaluation methodsEngineering the base model evaluation metricExploring custom metrics and their applicationsExploring statistical tests for comparing model metricsRelating the evaluation metric to successDirectly optimizing the metricSummary
Chapter 11: Explaining Neural Network Predictions
Technical requirementsExploring the value of prediction explanationsDemystifying prediction explanation techniquesExploring gradient-based prediction explanationsTrusting and understanding integrated gradientsUsing integrated gradients to aid in understanding predictionsExplaining prediction explanations automaticallyExploring common pitfalls in prediction explanations and how to avoid themSummaryFurther reading
Chapter 12: Interpreting Neural Networks
Technical requirementsInterpreting neuronsFinding neurons to interpretInterpreting learned image patternsExplaining predictions with image input data and integrated gradientsPractically visualizing neurons with image input dataDiscovering the counterfactual explanation strategySummary
Chapter 13: Exploring Bias and Fairness
Technical requirementsExploring the types of biasUnderstanding the source of AI biasDiscovering bias and fairness evaluation methodsEvaluating the bias and fairness of a deep learning modelTailoring bias and fairness measures across use casesMitigating AI biasSummary
Chapter 14: Analyzing Adversarial Performance
Technical requirementsUsing data augmentations for adversarial analysisAnalyzing adversarial performance for audio-based modelsExecuting adversarial performance analysis for speech recognition modelsAnalyzing adversarial performance for image-based modelsExecuting adversarial performance analysis for a face recognition modelExploring adversarial analysis for text-based modelsSummary
Part 3 – DLOps
Chapter 15: Deploying Deep Learning Models to Production
Technical requirementsExploring the crucial components for DL model deploymentIdentifying key DL model deployment requirementsChoosing the right DL model deployment optionsArchitectural choicesComputing hardware choicesModel packaging and frameworksCommunication protocols to useUser interfacesExploring deployment decisions based on practical use casesExploring deployment decisions for a sentiment analysis applicationExploring deployment decisions for a face detection and recognition system for security camerasDiscovering general recommendations for DL deploymentModel safety, trust, and reliability assuranceOptimizing model latencyTools that abstract deploymentDeploying a language model with ONNX, TensorRT, and NVIDIA Triton ServerPractically deploying a DL model with the single pipeline approachSummary
Chapter 16: Governing Deep Learning Models
Technical requirementsGoverning deep learning model utilizationGoverning a deep learning model through monitoringMonitoring a deployed deep learning model with NVIDIA Triton Server, Prometheus, and GrafanaGoverning a deep learning model through maintenanceExploring limitations and risks of using automated tasks triggered by model monitoring alertsSummary
Chapter 17: Managing Drift Effectively in a Dynamic Environment
Technical requirementsExploring the issues of driftExploring the types of driftExploring data drift typesExploring concept driftExploring model driftExploring strategies to handle driftExploring drift detection strategiesAnalyzing the impact of driftExploring strategies to mitigate driftDetecting drift programmaticallyDetecting concept drift programmaticallyDetecting data drift programmaticallyImplementing programmatic data distribution drift detection using evidentlyComparing and contrasting the Evidently and Alibi-Detect libraries for drift detectionSummary
Chapter 18: Exploring the DataRobot AI Platform
Technical requirementsA high-level look into what the DataRobot AI platform providesPreparing data with DataRobotIngesting data for deep learning model developmentExploratory analysis of the dataWrangling data for deep learning model developmentExecuting modeling experiments with DataRobotDeep learning modelingGathering model and prediction insightsMaking batch predictionsDeploying a deep learning blueprintPractically deploying a blueprint in DataRobotGoverning a deployed deep learning blueprintGoverning through model utilization in DataRobotGoverning through model monitoring in DataRobotGoverning through model maintenance in DataRobotExploring some customer success storiesSummary
Chapter 19: Architecting LLM Solutions
Overview of LLM solutionsHandling knowledge for LLM solutionsExploring chunking methodsExploring embedding modelsExploring the knowledge base index typesExploring orchestrator tools for LLM solutionsEvaluating LLM solutionsEvaluating LLM solutions through quantitative metricsEvaluating LLM solutions through qualitative evaluation methodsIdentifying challenges with LLM solutionsTackling challenges with LLM solutionsTackling the output and input limitation challengeTackling the knowledge- and information-related challengeTackling the challenges of accuracy and reliabilityTackling the runtime performance challengeTackling the challenge of ethical implications and societal impactsTackling the overarching challenge of LLM solution adoptionLeveraging LLM to build autonomous agentsExploring LLM solution use casesSummaryFurther reading
Index
Why subscribe?
Other Books You May EnjoyPackt is searching for authors like youShare Your ThoughtsDownload a free PDF copy of this book

Content preview from The Deep Learning Architect's Handbook

5 Understanding Autoencoders

Autoencoders are a type of model that was built mainly to accomplish representation learning. Representation learning is a type of deep learning task that focuses on generating a compact and representative feature to represent any single data sample, be it image, text, audio, video, or multimodal data. After going through some form of representation learning, a model will be able to map inputs into more representable features, which can be used to differentiate itself from other sample inputs. The representation obtained will exist in a latent space where different input samples will co-exist together. These representations are also known as embeddings. The applications of autoencoders will be tied closely to representation ...

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

The Machine Learning Solutions Architect Handbook

Publisher Resources

ISBN: 9781803243795

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills

The Deep Learning Architect's Handbook

by Ee Kin Chin

5

Understanding Autoencoders

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.