Apache MXNet AI

1
What Is Apache MXNet ?

• A deep learning framework
• Open source Apache 2.0 license
• Supports distributed gpu cluster
  training/deployment
• Of deep neural networks
• It supports a variety of language bindings
• Supports hybridize for increased
  speed/optimization
• Supports near linear scaling on gpu / host
  clusters
• Provides support for the Horovod framework

2
MXNet Language Bindings

• MXNet has a Python based API
• MXNet also supports the following language
  bindings
• Scala
• Julia
• Clojure
• Java
• C
• R
• Perl

3
MXNet Related Terms

• Horovod
• MMS
• DGL
• ONNX
• Hyperparameter
• D2l.ai
• KVStore
• DMLC

A distributed deep learning framework from
Uber MXNet Model Server Deep Graph Library Open
Neural Network Exchange A parameter whose value
is used to control the learning process A jupyter
notebook based deep learning book for Mxnet
Key-value store interface used by
MXNet Distributed (Deep) Machine Learning
Community - GitHub
4
MXNet Eco System

• Coach RL
• Deep Graph
• GluonFR
• InsightFace
• Keras-MXNet
• MXBoard
• MXFusion
• MXNet Model
• Optuna
• Sockeye

A Python reinforcement learning framework DGL is
a Python pkg for deep learning on graphs A
community driven toolkit for face detection and
recognition A face detection and recognition
repository A back end of high level API
Keras Logging API's for TensorBoard
visualisation A modular deep probabilistic
programming library A flexible tool for serving
models exported from Mxnet A hyperparameter
optimization framework A sequence to sequence
framework for neural translation
5
MXNet Eco System

• TensorLY
• TVM
• Xfer
• GluonCV
• GluonNLP
• GluonTS

A high level API for tensor methods An open deep
learning stack for GPU's, CPU's etc A library for
the transfer of knowledge in deep nets A computer
vision toolkit with a rich model zoo Deep
learning models for natural language processing A
toolkit for probabilistic time series modelling
6
MXNet User Community
7
MXNet Architecture
8
MXNet Architecture

• Runtime Dependency Engine
• Schedules and executes the operations
• According to their read/write dependency
• Storage Allocator
• Efficiently allocates and recycles memory blocks
• On host (CPU) and devices (GPUs)
• Resource Manager
• Manages global resources, such as
• The random number generator and temporal space
• NDArray
• Dynamic, asynchronous n-dimensional arrays

9
MXNet Architecture

• Symbolic Execution
• Static symbolic graph executor, which provides
• Efficient symbolic graph execution and
  optimization
• Operator
• Operators that define static forward/gradient
  calc (backprop)
• SimpleOp
• Operators that extend NDArray operators and
• Symbolic operators in a unified fashion
• Symbol Construction
• Symbolic construction, which provides a way to
  construct
• A computation graph (net configuration)

10
MXNet Architecture

• KVStore
• Key-value store interface for efficient parameter
  synchronization
• Data Loading(IO)
• Efficient distributed data loading and
  augmentation

11
MXNet Data Loading

• For large data sets data loading is optimized in
  MXNet
• Data format
• Uses dmlc-cores binary recordIO implementation
• Data Loading
• Reduced IO cost by utilizing the threaded
  iterator
• Provided by dmlc-core
• Interface design
• Write MXNet data iterators in just a few lines of
  Python

12
MXNet Dependency Engine

• Helps to parallelize computation across devices
• Helps to synchronize computation when
• We introduce multi-threading
• A run time dependency schedule graph is created
• The graph is then used to
• Optimize processing
• Optimize memory use
• Aid parallelism when using
• GPU / CPU clusters
• For deep learning memory use
• Usage during training gt during prediction

13
MXNet Forward Vs Backward Graph
14
Available Books

• See Big Data Made Easy
• Apress Jan 2015
• See Mastering Apache Spark
• Packt Oct 2015
• See Complete Guide to Open Source Big Data
  Stack
• Apress Jan 2018
• Find the author on Amazon
• www.amazon.com/Michael-Frampton/e/B00NIQDOOM/
• Connect on LinkedIn
• www.linkedin.com/in/mike-frampton-38563020

15
Connect

• Feel free to connect on LinkedIn
• www.linkedin.com/in/mike-frampton-38563020
• See my open source blog at
• open-source-systems.blogspot.com/
• I am always interested in
• New technology
• Opportunities
• Technology based issues
• Big data integration