Tutorials¶

Automatic differentiation
How to write a custom layer in Apache MxNet Gluon API
- The simplest custom layer
- Hybridization and the difference between Block and HybridBlock
- Adding a custom layer to a network
- Parameters of a custom layer
- Conclusion
Creating custom operators with numpy
- Parameter-less operators
- Parametrized Operator
- Using custom operators with fork
Image Augmentation
- What are the prerequisites?
- Where can I find the augmentation transforms?
- Sample Image
Spatial Augmentation
- RandomResizedCrop
- RandomFlipLeftRight
- RandomFlipTopBottom
Color Augmentation
- RandomBrightness
- RandomContrast
- RandomSaturation
- RandomHue
- RandomColorJitter
- RandomLighting
Composed Augmentations
Gluon Datasets and DataLoader
- Introduction to Datasets
- Introduction to DataLoader
- Machine learning with Datasets and DataLoaders
Using own data with included Datasets
Using own data with custom Datasets
Appendix: Upgrading from Module DataIter to Gluon DataLoader
Gluon - Neural network building blocks
- Define Network
- Initialize Parameters
- Loss Functions
- Updating the weights
Gluon: from experiment to deployment, an end to end tutorial
- Overview
- Prerequisites
- The Data
- Training using Gluon
- Load the model and run inference using the MXNet Module API
- What’s next
- References
Gotchas using NumPy in Apache MXNet
- Asynchronous and non-blocking nature of Apache MXNet
- Converting NDArray to NumPy Array blocks calculation
- NumPy operators vs. NDArray operators
- How to minimize the impact of blocking calls
- Conclusion
Hybrid - Faster training and easy deployment
- Fast, portable neural networks with Gluon HybridBlocks
- A Hybrid of Imperative and Symbolic Programming
- HybridBlock
- Hybridize
- Serializing trained model for deployment
- Operators that do not work with hybridize
- Summary
Image similarity search with InfoGAN
- Generator
- Discriminator
- Training Loop
- Image similarity
- How the Generator learns
Learning Rate Finder
- Simple Idea
- Epoch to Iteration
- Implementation
- Usage
- Wrap Up
Learning Rate Schedules
- Schedules
- Using Schedules
  - Manually setting the learning rate: Gluon API only
- Advanced Schedules
Advanced Learning Rate Schedules
- Custom Schedule Shapes
  - (Slanted) Triangular
  - Cosine
- Custom Schedule Modifiers
Logistic regression using Gluon API explained
- Working with data
- Defining and training the model
- Validating the model
- Putting it all together
- Tip 1: Use only one neuron in the output layer
- Tip 2: Encode classes as 0 and 1
- Tip 3: Use SigmoidBinaryCrossEntropyLoss instead of LogisticRegressionOutput
- Conclusion
Hand-written Digit Recognition
- Prerequisites
- Loading Data
- Approaches
  - Defining a network: Multilayer Perceptron (MLP)
  - Convolutional Neural Network (CNN)
- Summary
Multiple GPUs training with Gluon API
- Prerequisites
- Storing data on GPU
- Storing the network on multiple GPUs
- Multiple GPUs training schema
- Splitting data between GPUs
- Multiple GPUs classification of MNIST images
- Advanced topic
- Conclusion
- Recommended Next Steps
Naming of Gluon Parameter and Blocks
- Naming Blocks
- Naming Parameters
- Name scopes
- Saving and loading
- Replacing Blocks from networks and fine-tuning
NDArray - Scientific computing on CPU and GPU
- Creating NDArray
- Basic attributes
- NDArray Operations
- Using GPU
Using pre-trained models in MXNet
- Loading the model
- Loading the data
- Testing the different networks
- Fine-tuning pre-trained models
  - Transferring knowledge through fine-tuning
  - Fine Tuning an ONNX Model
Saving and Loading Gluon Models
- Setup: build and train a simple model
- Saving model parameters to file
- Loading model parameters from file
- Saving model parameters AND architecture to file
- Loading model parameters AND architecture from file
  - From a different frontend
    - Loading serialized Hybrid networks from C
    - Loading serialized Hybrid networks from Scala
  - From Python
Data Transforms
- Using ToTensor and .transform_first
- Normalize
- Compose
- CenterCrop and Resize
- Augmentation
- Summary
  - Additional Reading

Apache MXNet is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the Apache Incubator. Incubation is required of all newly accepted projects until a further review indicates that the infrastructure, communications, and decision making process have stabilized in a manner consistent with other successful ASF projects. While incubation status is not necessarily a reflection of the completeness or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.

"Copyright © 2017-2018, The Apache Software Foundation Apache MXNet, MXNet, Apache, the Apache feather, and the Apache MXNet project logo are either registered trademarks or trademarks of the Apache Software Foundation."