Application of TensorFlow as a backend tool for Image Classification

Transcript

1. By Mercy Oseni Application of TF as a backend tool for Image Classification
2. Speaker Bio - Software Developer at Andela - ML/AI Enthusiast - Maths Lover :)
3. Session Objectives ❖ What is TensorFlow? ❖ What is Keras? ❖ Applications of TF and Keras ❖ When to use TF and/or Keras ❖ Why I chose TF backend + Keras and not core TF ❖ We will build a model that predicts dog breeds
4. What is TensorFlow? ❖ Open Source software library by Google. ❖ Originally created for tasks with heavy numerical computations. ❖ Main application is in ML and DNN. ❖ C++ backend. ❖ Uses a structure known as data flow graphs.
5. What is Keras? ❖ Keras is a high-level neural networks API, written in Python. ❖ It is an abstraction layer that uses other underlying deep libraries like TensorFlow, CNTK, or Theano. ❖ Allows for easy and fast prototyping (through user friendliness, modularity, and extensibility). ❖ Supports both convolutional networks and recurrent networks, as well as combinations of the two. ❖ Runs seamlessly on CPU and GPU. ❖ Keras is the official high-level API for TF.
6. “TF core” VS “TF backend + Keras” ❖ TF core provides lower layer programming for deep learning than Keras, hence runs more efficiently. ❖ A better understanding of the Mathematical concepts behind Neural Networks. ❖ Higher learning curve.
7. “TF backend + Keras” VS “TF core” ❖ Keras isolates lots of complex methods in few lines of code, making it easier to learn and use. ❖ It provides a faster track when you need to quickly train and test a model.
8. Why I chose to use TF backend + Keras ❖ We will be building our model in Keras because it’s easier to use and provides a lot of abstractions.
9. Keras and TF shouldn’t be compared ❖ Both libraries operates at different levels; ➢ Lower level: This is where mathematical operations like generalised matrix-matrix multiplication and neural network primitives like convolutional operations are implemented. E.g TF, Theano ➢ Higher level: At this level, the lower level primitives are used to implement NN abstractions like layers and models. E.g. Keras
10. Predicting Dog Breeds
11. Materials We are going to use Udacity’s Dog Project and follow the instructions to build our model. Link: https://github.com/udacity/dog -project
12. Our Expectations
17. Time to lay our hands on our Notebook :)
18. Reflection ❖ We learnt about the Deep Learning framework ❖ When to apply the framework ❖ We built a model that predicts a dog’s breed
19. Questions? Thanks for your time :) Here’s the link to the project repository: https://github.com/mehrceey04/ML-Udacity/tree/mast er/dog-project