These slides elaborate on frameworks for Deep Learning.
1. Hello and welcome to this lecture on Deep Learning frameworks.
2. In the previous lectures we spoke about the theory. Now you know all the building blocks to develop a neural network on your own. But of course you don’t need to develop everything from scratch. There are many different Deep Learning Frameworks, that provide all the building blocks we spoke in the earlier lectures, and you can take them exactly like building blocks and stack them together.
a. The most famous is Tensorflow. Tensorflow was originally developed by google, but it was released as open source end of 2015. It can perform its operations on CPU and GPU and is available for all operating systems, including mobile platforms like Android or iOS. It is written in Python, this is why you need to install Python as well, because we implicitly will use it from R.
b. Others frameworks are mxnet, Caffe, CNTK, or Theano. Since these frameworks are low-level and hard to work with, a high-level API was developed.
c. The one we will use is called Keras. It runs on top of Tensorflow, CNTK, and Theano. It is very user friendly and easier to use than Tensorflow. It is modular and so can stack together a neural network very easily. The layers, loss functions, optimizers, activation functions, and more are defined in separate models.
d. I highlighted the combination that we are going to work with in this course: Keras on top of Tensorflow. We will only interact with Keras, and Keras independently will interact with Tensorflow in the backend. Tensorflow is written in Python, so you need to install Python, I recommend to install anaconda. Then you can install the Keras package in RStudio and this will take care of creating an environment, including Tensorflow. We will cover this in a separate lecture.
That’s it for this lecture. Thank you very much for watching and see you in the next one.
2. In the previous lectures we spoke about the theory. Now you know all the building blocks to develop a neural network on your own. But of course you don’t need to develop everything from scratch. There are many different Deep Learning Frameworks, that provide all the building blocks we spoke in the earlier lectures, and you can take them exactly like building blocks and stack them together.
a. The most famous is Tensorflow. Tensorflow was originally developed by google, but it was released as open source end of 2015. It can perform its operations on CPU and GPU and is available for all operating systems, including mobile platforms like Android or iOS. It is written in Python, this is why you need to install Python as well, because we implicitly will use it from R.
b. Others frameworks are mxnet, Caffe, CNTK, or Theano. Since these frameworks are low-level and hard to work with, a high-level API was developed.
c. The one we will use is called Keras. It runs on top of Tensorflow, CNTK, and Theano. It is very user friendly and easier to use than Tensorflow. It is modular and so can stack together a neural network very easily. The layers, loss functions, optimizers, activation functions, and more are defined in separate models.
d. I highlighted the combination that we are going to work with in this course: Keras on top of Tensorflow. We will only interact with Keras, and Keras independently will interact with Tensorflow in the backend. Tensorflow is written in Python, so you need to install Python, I recommend to install anaconda. Then you can install the Keras package in RStudio and this will take care of creating an environment, including Tensorflow. We will cover this in a separate lecture.
That’s it for this lecture. Thank you very much for watching and see you in the next one.