The document discusses several deep learning frameworks including TensorFlow, Keras, PyTorch, Theano, Deep Learning 4 Java, Caffe, Chainer, and Microsoft CNTK. TensorFlow was developed by Google Brain Team and uses dataflow graphs to process data. Keras is a high-level neural network API that runs on top of TensorFlow, Theano, and CNTK. PyTorch was designed for flexibility and speed using CUDA and C++ libraries. Theano defines and evaluates mathematical expressions involving multi-dimensional arrays efficiently in Python. Deep Learning 4 Java integrates with Hadoop and Apache Spark to bring AI to business environments. Caffe focuses on image detection and classification using C++ and Python. Chainer was developed in collaboration with several companies
8. TensorFlow
Developed by
Google Brain Team
Supports languages Uses dataflow graphs to
process data
Easy to build Machine
Learning models
Robust Machine Learning
production
9. TensorFlow
Developed by
Google Brain Team
Supports languages Uses dataflow graphs to
process data
Easy to build Machine
Learning models
Powerful experimentation for
research
Robust Machine Learning
production
10. TensorFlow
Developed by
Google Brain Team
Supports languages Uses dataflow graphs to
process data
TensorBoard for data
visualization
Easy to build Machine
Learning models
Powerful experimentation for
research
Robust Machine Learning
production
12. Keras
Francois Chollet is the author of
keras, with over 350,000 users
and 700+ open-source
contributors
13. Keras
High-level neural network API,
written in Python
Francois Chollet is the author of
keras, with over 350,000 users
and 700+ open-source
contributors
14. Keras
Runs on top of TF,
Theano, CNTK
CNTK
Francois Chollet is the author of
keras, with over 350,000 users
and 700+ open-source
contributors
High-level neural network API,
written in Python
15. Keras
CNTK
Used in several startups,
research labs, and companies
High-level neural network API,
written in Python
Runs on top of TF,
Theano, CNTK
Francois Chollet is the author of
keras, with over 350,000 users
and 700+ open-source
contributors
16. Keras
CNTK
Used in several startups,
research labs, and companies
User-friendly as it offers simple APIs and provides clear
and actionable feedback upon user error
Provides modularity as a sequence or a graph of standalone,
fully configurable modules that can be plugged together with
as few restrictions as possible
Easily extensible as new modules are simple to add. This feature
makes Keras suitable for advanced research
1
2
3
Runs on top of TF,
Theano, CNTK
Francois Chollet is the author of
keras, with over 350,000 users
and 700+ open-source
contributors
High-level neural network API,
written in Python
19. PyTorch
Authored by Adam Paszke, Sam Gross, Soumith
Chintala and Gregory Chanan
Lua based scientific computing framework for Machine
Learning and Deep Learning algorithms
20. PyTorch
Authored by Adam Paszke, Sam Gross, Soumith
Chintala and Gregory Chanan
Lua based scientific computing framework for Machine
Learning and Deep Learning algorithms
It employed CUDA along with C/C++ libraries for
processing and was designed to scale the
production of building models and overall flexibility
21. PyTorch
Authored by Adam Paszke, Sam Gross, Soumith
Chintala and Gregory Chanan
Lua based scientific computing framework for Machine
Learning and Deep Learning algorithms
Widely used in companies like
It employed CUDA along with C/C++ libraries for
processing and was designed to scale the
production of building models and overall flexibility
22. PyTorch
Authored by Adam Paszke, Sam Gross, Soumith
Chintala and Gregory Chanan
Lua based scientific computing framework for Machine
Learning and Deep Learning algorithms
Widely used in companies like
Provides flexibility and speed due to its hybrid front-end
Enables scalable distributed training and performance
optimization in research and production using
torch.distributed backend
Deep integration into Python allows popular libraries and
packages to be used for quickly writing neural network
layers in Python.
1
2
3
It employed CUDA along with C/C++ libraries for
processing and was designed to scale the
production of building models and overall flexibility
26. Theano
Python library that allows to define, optimize, and
evaluate mathematical expressions involving multi-
dimensional arrays efficiently
Developed by Written in
27. Theano
Python library that allows to define, optimize, and
evaluate mathematical expressions involving multi-
dimensional arrays efficiently
• Theano has tight integration with NumPy for data computations
• Uses GPU’s to perform data-intensive computations which are much
faster than on a CPU
• Has extensive unit-testing and self-verification that can detect and
diagnose many types of errors
Developed by Written in
28. How Big Data evolved?
Microsoft CNTKDeep Learning 4 Java
29. Deep Learning 4 Java
Developed by a Machine Learning
group led by Adam Gibson
30. Deep Learning 4 Java
Written in Java and Scala, DL4J supports different neural
networks like CNN, RNN, and LSTM
Developed by a Machine Learning
group led by Adam Gibson
31. Deep Learning 4 Java
DL4J was contributed to Eclipse Foundation.
Integrated with Hadoop and Apache Spark, DL4J
brings AI to business environments for use on
distributed CPUs and GPUs
Written in Java and Scala, DL4J supports different neural
networks like CNN, RNN, and LSTM
Developed by a Machine Learning
group led by Adam Gibson
32. Deep Learning 4 Java
DL4J was contributed to Eclipse Foundation.
Integrated with Hadoop and Apache Spark, DL4J
brings AI to business environments for use on
distributed CPUs and GPUs
• Provides a distributed computing framework as training
with DL4J occurs in a cluster
• Includes an n-dimensional array class using ND4J that
allows scientific computing in Java and Scala
• Offers a vector space modeling and topic modeling
toolkit that is designed to handle large text sets and
perform NLP
Written in Java and Scala, DL4J supports different neural
networks like CNN, RNN, and LSTM
Developed by a Machine Learning
group led by Adam Gibson
35. Caffe
Caffe stands for Convolutional Architecture for Fast Feature EmbeddingDeveloped at
36. Caffe
Caffe stands for Convolutional Architecture for Fast Feature Embedding
Written in C++, with a Python interface
Developed at
37. Caffe
Caffe stands for Convolutional Architecture for Fast Feature Embedding
Written in C++, with a Python interface
Generally used for image detection and
classification
Developed at
38. Caffe
Caffe stands for Convolutional Architecture for Fast Feature Embedding
Written in C++, with a Python interface
Generally used for image detection and
classification
• Used in academic research projects, startup
prototypes, and large-scale industrial applications in
vision, speech, and multimedia
• Caffe supports GPU and CPU-based acceleration
computational kernel libraries such as NVIDIA, cuDNN
and, IntelMLK
• Offers high speed can process over 60M images per
day with a single NVIDIA K40 GPU
Developed at
41. Chainer
Preferred Networks in collaboration with
IBM, Intel, Microsoft, and Nvidia
Written purely in Python, it runs on top of
Numpy and CuPy Python libraries
Developed by
42. Chainer
Preferred Networks in collaboration with
IBM, Intel, Microsoft, and Nvidia
Written purely in Python, it runs on top of
Numpy and CuPy Python libraries
Provides a number of extended libraries
Developed by
43. Chainer
Developed by
Preferred Networks in collaboration with
IBM, Intel, Microsoft, and Nvidia
Written purely in Python, it runs on top of
Numpy and CuPy Python libraries
Provides a number of extended libraries • Supports CUDA computation. It only requires a few
lines of code to leverage a GPU. It also runs on
multiple GPUs with little effort
• Provides various network architectures, including
feed-forward nets, convnets, recurrent nets, and
recursive nets
44. How Big Data evolved?
Microsoft CNTKMicrosoft CNTK
46. Microsoft Cognitive Toolkit
CNTK
CNTK is a Deep Learning framework that builds a neural network as a
series of computational steps via a directed graph
Developed by
47. Microsoft Cognitive Toolkit
CNTK
Supports interfaces such as Python, C++
CNTK is a Deep Learning framework that builds a neural network as a
series of computational steps via a directed graph
Developed by
48. Microsoft Cognitive Toolkit
CNTK
Supports interfaces such as Python, C++
CNTK is a Deep Learning framework that builds a neural network as a
series of computational steps via a directed graph
Used mainly for
Developed by
49. Microsoft Cognitive Toolkit
CNTK
Developed by
Supports interfaces such as Python, C++
CNTK is a Deep Learning framework that builds a neural network as a
series of computational steps via a directed graph
Used mainly for
• Designed for speed and efficiency, CNTK scales well in
production using GPUs but has limited support from the
community
• Supports both RNN and CNN type of neural models
capable of handling image, handwriting and speech
recognition problems