Guide to choose right deep Learning framework for your AI project by Rishikesh

As world rolling around Artificial Intelligence (AI), demand for the AI-based product seen exponential growth, so the AI research. Deep learning algorithms and techniques are widely used for research and development of these products. Good news is that year by year Deep Learning has seen its glory in the release of many open source frameworks which ease the pain to develop and implement these algorithms.

As there are many deep learning frameworks out there and it can lead to confusion as to which one is better for your task. And choosing a deep learning framework for an AI project is as important as choosing a programming language to code product, Data science project coupled with the right deep learning framework has truly amplified the overall productivity.

In this talk, I will discuss the common points which help developers to understand which framework will be the perfect fit for solving given business challenges. Also, we will look into some of the most widely used frameworks and comparing with standard benchmarks.

Following deep learning/machine learning frameworks will be discussed:
1. **Tensorflow**
2. **PyTorch**
3. **Chainer** and/or **MXNET**

Key Highlight of this talk:

* define key points to judge any deep learning framework.

* hardware dependencies.

* anatomy of widely used open source frameworks.

* comparison of above-mentioned frameworks as per defined key points.

**Who is the audience?**

Anyone who inspired to code deep learning algorithms.

**Audience Level**: Beginner to Intermediate

Transcript

1. Guide to choose right deep Learning framework for your AI project Rishikesh rishikksh20@gmail.com Data Scientist @ Humonics Global Pvt ltd
2. @whoami • Data Scientist at Humonics Global Private Ltd. • Github: rishikksh20 • Telegram: @rishikksh20 • Contribute to : Keras, Numpy, Scikit-Learn, matplotlib etc.
3. Quick Survey o How many have machine learning background ? o How many have ever used any deep learning framework ? o How many people have used only 1 framework ? o How many of you contributed to any deep learning framework ?
4. o Key points to keep o Identify scope of the project o Bottlenecks o Frameworks o Benchmarking Outline
5. • Easy of coding • Installation and support • Boilerplate examples and community • Stability and Scalability • Speed of Training and Inference Key Points to keep
6. Ease of Coding • Minimum Line of code for complex algorithms • Ease of use in terms of architecture • Deployment friendly • Coding flexibility • How many functions/methods are offered out of the box • High level as well as Low level APIs • Debugging and logging
7. Installation and Support • Single hand installation • Good documentation • OS support • Multiple language support • Support bleeding edge technologies • Hardware Support (Mobile/CPU/GPU/Multi-GPU/TPU/Distributive/Cloud)
8. Boilerplate examples and community • Framework Developers • Kick start examples • Availability of tutorials and training materials • Active communities
9. Stability and Scalability • Maturity of frameworks • Back compatibility • Frequency of release • Number of users • Number of existing issues • Scale to large distributive environment
10. Speed of Training and Inference • How much time it take to train a model ? • How much time it take to inference ?
12. Identify scope of the project • Research Project • Small Project or POC • Medium Scale Project • Large Complex Project
13. Bottleneck • Installation and Dependency management • Hard to keep up with latest development • So far there is no standardization • Inconsistency in low level optimization • Too much Hardware dependencies
14. Tensorflow • Originally developed by Google Brain • Widely used machine learning framework • Backed by excellent and large community • Adequate documentation and huge learning resource available • Probably all necessary functions and features are out-of-the box • Production ready framework
15. • High level API like Keras. • Written in C++ but support many client language APIs like Java, Python, Go etc. • Started as static graph structure now support dynamic graph computation. • Deploy on Android • Good Visualization tool like tensorboard
16. So Any Reason to looks others ? • Less pythonic and flexible. For researchers, Tensorflow is hard to learn and hard to use. Research is all about flexibility, and lack of flexibility is baked into Tensorflow at a deep level.
17. PyTorch • Primarily developed by Facebook and Uber • Pure Pythonic approach and highly flexible (Dynamic computational Graph) • Research friendly • No need high level API like Keras • Simple to learn and code • Butter smooth debugging
18. Cons • Less mature • Not yet Production ready, but is in roadmap • Lack of in-built visualization tool like Tensorboard • Do not have mobile support yet • Scalability
19. • Supported by Microsoft and Amazon • Supports many languages, from C++ to Python, JavaScript, Go, and R • Faster training than other popular frameworks • Highly scalable and production ready • Easy to code using High level API “Gluon”. • Distributed environment • Deploy any where from GPU to mobile MXNet
20. • Lesser user base • Small community Cons
21. Benchmarking
22. Which one is more mature ?
23. Speed tensorflow pytorch mxnet 130 135 140 145 150 155 160 165 170 175 180 173 169 145 ResNet50 Feature Extraction
24. tensorflow pytorch mxnet 28 28.5 29 29.5 30 30.5 31 31.5 30 31 29 IMDB movie reviews
25. Questions ? Contact Details: Github: https:/ /github.com/rishikksh20 Telegram: @rishikksh20 Email : rishikksh20@gmail.com