- TensorFlow

|
|- Basic Concepts
| |- Tensors
| | |- Scalars
| | | |- Float, Int, Bool
| | |- Vectors
| | | |- Operations (Element-wise, Dot Product)
| | |- Matrices
| | | |- Matrix Operations (Transpose, Inverse)
| | |- n-dimensional Arrays
| | |- Tensor Indexing and Slicing
| | |- Broadcasting
| | |- Tensor Concatenation and Splitting
| |- Operations
| | |- Element-wise Operations (Add, Subtract, Multiply, Divide)
| | |- Matrix Operations (Matmul, Convolution, Pooling)
| | |- Activation Functions (ReLU, Sigmoid, Tanh)
| | |- Derivatives and Gradients
| | |- Softmax Function
| | |- Custom Activation Functions
| |- Computational Graphs
| |- Nodes (Operations)
| | |- Input and Output Nodes
| | | |- Placeholder Nodes
| | | |- Constant Nodes
| | | |- Variable Nodes
| | |- Control Flow Operations (Conditional, Loop)
| |- Edges (Tensors)
| | |- Data Flow
| | |- Control Dependencies
| | |- Tensors as Inputs and Outputs
| |- Directed Acyclic Graphs (DAGs)
| |- Topological Sorting
| |- Dependency Resolution
| |- Dynamic Graphs
| |- tf.function and AutoGraph
| |- Eager Execution Mode
|
|- Installation and Setup
| |- Pip Installation
| | |- Virtual Environments
| | | |- Conda, Virtualenv
| | |- Dependency Management
| | |- Requirements Files
| | |- Package Installation
| |- Conda Installation
| | |- Environment Management
| | | |- Creating, Activating, Deactivating Environments
| | | |- Environment Configuration
| | |- Package Installation
| |- GPU Support Setup
| | |- CUDA, cuDNN Installation
| | |- GPU Driver Installation
| | |- Compatibility Checking
| |- IDE Integration (e.g., Jupyter Notebook, Google Colab)
| |- Notebook Features (Magic Commands, Markdown Cells)
| |- Colab Features (Cloud-based, Hardware Acceleration)
| |- Integration with Version Control (Git)
| |- External Libraries (TensorFlow Addons, TensorBoard)
|
|- Core TensorFlow
| |- Variables
| | |- Initialization Methods (Random, Zeros, Ones)
| | |- Regularization (L1, L2)
| | |- Variable Scopes and Namespaces
| | |- Sharing Variables
| | |- Variable Collections
| | |- Variable Partitioning
| |- Constants
| | |- tf.constant() Function
| | |- Immutable Nature
| | |- Creating Constants from NumPy Arrays
| | |- Constant Folding
| |- Placeholders
| | |- tf.placeholder() Function
| | |- Placeholder Shape Inference
| | |- Feed Dictionary
| | |- Placeholder Deprecation
| |- Sessions
| |- Session Configuration
| | |- GPU Memory Allocation
| | |- ConfigProto Options
| | |- Session Management
| | |- Session Timeout
| |- Context Managers
| | |- with tf.Session() as sess:
| | |- tf.compat.v1.Session()
| |- Distributed Sessions
| |- Distributed TensorFlow Architecture
| |- Distributed Execution Modes
| | |- Parameter Servers
| | |- All-Reduce Communication
| | |- TensorFlow on Spark
| |- Distributed Tensor Operations
| |- Collective Operations
| |- Data Parallelism
| |- Model Parallelism
|
|- Neural Networks with TensorFlow
| |- High-level API (tf.keras)
| | |- Sequential Model
| | | |- Stacking Layers
| | | |- Model Compilation
| | | |- Loss Functions and Metrics
| | | |- Optimizers and Learning Rate Schedules
| | |- Functional API
| | | |- Graph-like Model Architectures
| | | | |- Multiple Inputs and Outputs
| | | | |- Branching and Merging
| | | |- Shared Layers
| | | |- Reusable Layer Instances
| | | |- Layer Sharing with Functional API
| | |- Model Subclassing
| | |- Custom Forward Pass
| | |- Dynamic Model Construction
| | |- Inheriting from Keras Models
| |- Pre-trained Models (e.g., VGG, ResNet)
| | |- Model Architecture Overview
| | |- Pre-processing and Input Pipelines
| | |- Transfer Learning Strategies
| | |- Feature Extraction vs. Fine-tuning
| | |- Model Zoo (TensorFlow Hub)
| | |- Customizing Pre-trained Models
| |- Custom Layers
| |- Layer Implementation
| | |- Layer Initialization
| | |- Forward and Backward Pass
| | |- Layer Configuration
| |- Integration with Existing Models
| |- Adding Custom Layers to Pre-trained Models
| |- Feature Extraction and Adaptation
| |- Layer Freezing and Unfreezing
|- Custom Models
| |- Model Architecture Design
| | |- Choosing Layer Types and Configurations
| | |- Model Complexity and Overfitting
| | |- Hyperparameter Tuning
| |- Loss Functions
| | |- Custom Loss Function Implementation
| | |- Weighted Losses
| | |- Handling Class Imbalance
| |- Optimizers
| | |- Custom Optimizer Implementation
| | |- Learning Rate Schedules
| | |- Warmup Schedules
| |- Regularization Techniques
| | |- Dropout, Batch Normalization
| | |- Weight Regularization (L1, L2)
| | |- Mixup, CutMix
| |- Custom Metrics
| |- Metric Implementation
| |- Custom Evaluation Metrics
| |- Monitoring Training and Validation Metrics
|
|- Data Loading and Preprocessing
| |- Data Formats (e.g., CSV, JSON, Image)
| | |- Parsing Techniques
| | |- Handling Missing Data
| | |- Working with Text and Image Data
| | |- Tokenization and Vectorization
| | |- Image Augmentation Techniques
| | |- Image Data Generators
| |- TensorFlow Dataset API
| | |- Creating Datasets from Arrays
| | |- Reading from Files (Text, TFRecord)
| | |- Batching and Shuffling
| | |- Buffering and Prefetching
| | |- Dataset Caching
| | |- Parallelizing Data Loading
| |- Data Augmentation
| | |- Image Augmentation Techniques
| | | |- Random Crop, Resize
| | | |- Flipping, Rotation
| | | |- Color Jittering, Brightness Adjustment
| | |- Text Augmentation Techniques
| | |- Text Tokenization
| | |- Synonym Replacement, Random Insertion
| | |- Sentence Mixup, Back-Translation
| |- Normalization and Standardization
| | |- Min-Max Scaling
| | |- Z-score Normalization
| | |- Feature-wise Transformation
| |- Batching and Shuffling
| |- Importance of Balanced Batches
| |- Shuffling for Generalization
| |- Handling Imbalanced Datasets
| |- Performance Considerations (Buffer Sizes)
|
|- Training and Evaluation
| |- Training Loop
| | |- Epochs and Iterations
| | |- Batch Training
| | |- Gradient Descent
| | | |- Mini-batch Gradient Descent
| | | |- Stochastic Gradient Descent
| | | |- Batch Gradient Descent
| |- Optimizers
| | |- Gradient Descent Variants
| | |- Adaptive Learning Rates (Adam, RMSprop)
| | |- Momentum and Nesterov Accelerated Gradient
| | |- Custom Optimizer Implementation
| |- Learning Rate Scheduling
| | |- Step Decay, Exponential Decay
| | |- Performance-based Scheduling
| | |- Cyclical Learning Rates
| |- Early Stopping
| | |- Monitoring Validation Loss
| | |- Patience and Restarts
| |- Model Checkpointing
| |- Saving Best Models
| |- Saving Multiple Checkpoints
| |- Resuming Training from Checkpoints
| |- TensorBoard Integration
| |- Logging Scalars, Images, Histograms
| |- Graph Visualization
| |- Hyperparameter Tuning
| |- Model Evaluation
| |- Accuracy, Precision, Recall, F1-score
| |- Confusion Matrix
| |- ROC Curve, AUC Score
| |- Visualizing Model Outputs
|
|- TensorFlow Extended (TFX)
| |- Data Validation
| | |- Schema Definition
| | |- Anomaly Detection
| | | |- Outlier Detection Techniques
| | | |- Data Drift Detection
| |- Data Transformation
| | |- Feature Engineering Techniques
| | | |- Feature Crosses
| | | |- Embedding Transformation
| | |- Scaling and Normalization
| | | |- Min-Max Scaling
| | | |- Z-score Normalization
| | |- Handling Categorical Features
| | |- One-hot Encoding
| | |- Embedding Layers
| |- Model Analysis
| | |- Fairness Indicators
| | |- Slicing Metrics
| | |- Model Interpretability
| | |- Feature Importance
| | |- SHAP Values
| |- Serving
| | |- Model Exporting
| | |- TensorFlow Serving Configuration
| | |- gRPC Interface
| | |- REST API
| |- Metadata Management
| |- Tracking Data Provenance
| |- Model Versioning
| |- Lineage and Attribution
| |- Pipeline Orchestration
| |- TFX Pipelines
| |- Airflow Integration
| |- Kubeflow Pipelines
| |- Pipeline DSLs (e.g., Beam, Apache NiFi)
|
|- Deployment
| |- TensorFlow Serving
| | |- Dockerizing Models
| | |- Kubernetes Deployment
| | |- Multi-model Serving
| | |- TensorRT Integration
| | |- TensorFlow Serving with GPUs
| |- TensorFlow Lite
| | |- Model Optimization Techniques
| | | |- Quantization (Dynamic, Post-training)
| | | |- Pruning
| | | |- Model Compression (Knowledge Distillation)
| | |- TensorFlow Lite for Microcontrollers
| | |- Porting Models to Microcontrollers
| | |- Optimization for Memory and Speed
| |- TensorFlow.js
| |- Model Conversion
| |- Web Integration
| |- Transfer Learning in the Browser
| |- Real-time Inference
|
|- Advanced Topics
|- Distributed TensorFlow
| |- Parameter Servers
| |- Asynchronous Training
| |- TensorFlow on Spark
| |- TensorFlow with Horovod
|- TPU Usage
| |- TPU Architecture Overview
| |- TPU-specific Optimizers
| |- TPU Training Strategies
| |- TPU Pod Configuration
|- Mixed Precision Training
| |- Floating-point Precision
| |- Precision Scaling
| |- Loss Scaling
| |- Mixed Precision in TensorFlow
|- Custom Operations
| |- C++ API
| |- CUDA Integration
| |- Performance Optimization
| |- Extending TensorFlow with Custom Ops
|- Research Papers and New Features
|- Review of Latest TensorFlow Features
|- Implementing State-of-the-Art Models
|- Research Contributions to TensorFlow Ecosystem