Demystifying Machine Learning on the Web with Tensorflow.js

This talk would give a little background into what machine learning is, and an introduction to TensorFlowJS. We would talk briefly about the benefits of machine learning on the browser and show some practical applications. We’d also look into the different methods, and platforms in which it can be used and some demos on how it can be used.

This talk was delivered at the First Open Source Festival in Africa

Transcript

1. Demystifying Machine
   Learning on the Web
   with Tensorflow.js. Shedrack Akintayo
   RelianceHMO
   @coder_blvck
   

   View Slide

2. Meet Shedrack!
   - Frontend Engineer.‍
   - Loves music and movies.
   - Loves all things javascript.
   Shedrack Akintayo
   RelianceHMO
   @coder_blvck
   

   View Slide

3. I’m NOT a Machine Learning
   Engineer, I just happen to
   know Javascript ‍♂
   

   View Slide

4. What is Machine Learning
   though?
   

   View Slide

5. Open Source
   Festival 2020
   Machine Learning
   Machine Learning is the ability of computers to learn by themselves from
   specific data called datasets without being explicitly programmed.
   

   View Slide

6. Open Source
   Festival 2020
   Types of Machine Learning
   ● Supervised Learning
   ● Unsupervised Learning
   ● Reinforcement Learning
   

   View Slide

7. What is Tensorflow &&
   Tensorflow.js
   

   View Slide

8. Photo
   comes here
   Tensorflow
   Tensorflow is an end to end open source
   machine learning platform with a
   comprehensive and flexible ecosystem of
   tools, libraries, community support and
   resources.
   

   View Slide

9. Open Source
   Festival 2020
   Tensorflow.js
   This is an open source javascript library built
   on deeplearn.js(Tensor Flow) used for
   creating deep learning modules on the
   browser. With it we can create CNNs, RNNs,
   etc. directly on the browser and train these
   modules using the Client’s GPU processing
   power. Hence, a server GPU is not necessary
   to train the modules.
   

   View Slide

10. Open Source
    Festival 2020
    TensorFlow.js Stack.
    ● TensorFlow.js Core: A flexible low-level API, formerly known as
    deeplearn.js
    ● TensorFlow.js Layers: A high-level API which implements functionality
    similar to keras
    ● TensorFlow.js Data: A simple API to load and prepare data
    ● TensorFlow.js Converter: tools to import a TensorFlow SavedModel to
    TensorFlow.js
    

    View Slide

11. Overview of Tensorflow.js
    APIs ⛩
    

    View Slide

12. Tensorflow.js is powered by WebGL and it
    offers high-level layers API for defining
    models, and a low-level API for linear
    Algebra and automatic differentiation.
    TensorFlow.js also supports importing
    TensorFlow SavedModels and Keras
    Models
    

    View Slide

13. View Slide

14. Platforms that support
    Tensorflow.js
    

    View Slide

15. Open Source
    Festival 2020
    Browser React Native
    (coming soon)
    NodeJS
    

    View Slide

16. Why Tensorflow.js
    though? ‍♂
    

    View Slide

17. Why Tensorflow.js
    ● You can import an existing, pre-trained model for inference.
    ● You can retrain an imported model
    ● You can author models directly in the browser
    ● Write once, run everywhere.
    ● No drivers or installs
    ● Easy to Visualize and interact with data
    

    View Slide

18. Important Terms used in
    TensorFlow.js
    

    View Slide

19. Terms
    ● Tensor: This is the central unit of values in Tensorflow, it consists
    of a set of values shaped into an array of one or more dimensions
    ● Models: A set of functions that looks for a relationship between
    features and labels of data
    ● Epoch: An epoch is the entire processing or iteration by the
    learning algorithm of the entire data-set.
    ● Learning rate: is a configurable hyperparameter used in the
    training of neural networks that has a small positive value, often in
    the range between 0.0 and 1.0.
    

    View Slide

20. Pretrained Models in
    TensorFlow.js
    

    View Slide

21. Pretrained Models
    These are machine learning models that have already been written in
    tensorflow. It is also a saved network that was previously trained on a
    large dataset, typically on a large-scale image-classification task.
    They are trained to either deal with data in form of audio, image or text.
    The have very easy to use APIs, you just have to install or link them from
    their respective CDNS to use them
    

    View Slide

22. Some Pretrained Models available in
    Tensorflow.js
    ● Object Classification
    ● Body Segmentation
    ● Image classification
    ● Speech Command Recognition
    ● Sentence Encoding.
    

    View Slide

23. Limitations of Tensorflow.js
    ● Needs a large amount of data(unless using pre-trained models)
    ● Training on the web is slow. It takes a lot of time
    ● Bias and Ethics
    ● Speech Command Recognition
    ● Sentence Encoding.
    

    View Slide

24. Useful links &&
    Resources
    

    View Slide

25. Demo Links for Tensorflow.js
    ● Game Demo by Ashinze Ekene
    ● TensorFlow.js website
    ● Pretrained Models
    ● Deep Learning with javascript book
    ● Tensorflow.js Demos
    

    View Slide

26. Learning Resources and References
    ● Machine Learning Course from stanford
    ● Machine Learning Crash Course by Google
    ● Machine Learning for frontend developers by Charlie Gerard
    ● Machine Learning Magic for your Javascript Application
    ● Machine Learning for frontend developer by Charlie Gerard| JS Heroes
    2019
    ● Machine Learning on the browser with Tensorflow.js by Ashinze
    Ekene[slides]
    ● A Gentle Introduction to Tensorflow.js
    

    View Slide

27. Thank
    you! Shedrack Akintayo
    RelianceHMO
    @coder_blvck
    

    View Slide

28. View Slide