Neural Networks are all over the place, you can see them in fields like Robotics in Self-Driving-Cars, in Medical assistance with Watson and many others applications. When people talk about it sounds cryptic and really hard. But here's how you can create one yourself.
Neural Networks are used in Self driving cars. They are seen all over the news today in companies like Tesla, Uber, Google
Build one yourself
The Steps to Build Your Neural Network
- Get your hands on data
- Create your model
- Train your network
- Use in a test dataset
- Go wild and use it into new data
First we need the data
Let's take the example of the MNIST digit recognition competition. We are supplied with hand written digits and our task is
- Given a new image of a handwritten digit, tell what the digit is
For this task we're given the handwritten digit with 28x28 pixels, to use this image as a input for our network we need to flatten the vetor to a 784 one line vector, here's an example with a 2x2 matrix
So now we know that our input digit is going to be 784 long, so we can beggin creating our Neural Network.
If you are not familiar with PyTorch, you can check out here, or if you're new to Neural Networks you can start here. It's pretty straight forward. So we need to create a NN ( Neural Network ) with 784 inputs. The outputs and how many layers we can choose, it's simple to add or remove more of them in your model.
- Import PyTorch
- Create your Model with nn.Sequential
- Specify how many inputs and outputs it should have on nn.Linear(#inputs,#outputs)
- Print your model to see the configuration
Welcome to the club!
There you are. You've created your first Neural Network, doesn't feel good ?