Welcome to session on Custom Model Building in Tensorflow.
Until now, you have worked with the simple and traditional kind of approach to build a model using the keras wrapper in Tensorflow. Most of the built-in layers and pre-trained models do the job but what if you want to create your own custom models or layers?
The Tensorflow framework allows us to create models ranging from simple to complex architecture. In the world of deep learning, developers may come from a varied set of background or experience levels. TensorFlow understands this and provides you with different approaches whether you’re a beginner or an experienced person.
In the next video, let’s understand how we can do that and look at different type of approaches which you can utilise for model building.
The tf.keras is TensorFlow’s implementation of the keras API specification. Now keras has become a high-level API to build and train models in TensorFlow 2.0 and higher versions. As explained in the video, there are three methods to implement your own neural network architectures in TensorFlow:
- Sequential API
- Functional API
- Model Subclassing
In this session
In this session, you will understand how to work with custom models and layers using the TensorFlow framework, This session will introduce you to the following topics:
- Introduction
- Sequential vs Functional API
- Model Subclassing
- Build Custom Models
- Building Custom Layers
- Building Datasets
- Building Custom Training Loops
People you will hear in this session
Subject Matter Expert:
Sandeep kumar H
Instructor, upGrad