Deeplearning in a nutshell

General introduction

This project is different from the other projects on my Account.
The main focus is not on the programming related stuff, but on the deep learning related theory.

This project was, like many of my projects, made for school. To be spesific for POS logik. It's not a important fact, but i feel like i need to tell it.
I didn't do all of the work myself. This Project was made in cooperation with two friends @LuckyForce (Adrian Schauer) @Marko (Marko Gavranic).

What is this Project really about.

Like already told is the main focus on the Deep Learning theory. The explanation part is done in google docs, not here.

The reason for that is, if someting is changed, everyone with the link has the update right away. It make things a lot easier, in the aspect of the cooperation with my teammembers. If you don't want to rely on the link, you can also simply download it as any file format you like! :)

You can view the document here: https://docs.google.com/document/d/1VaagX4wZUiy-Vo_qol5mipthvHf0_rwtcMO2qcA_nUs/edit?usp=sharing

The whole document is unfortunately written in german:(

You are probably wondering, if everything is on the google docs, for what is this Repository then?

The answer is simple. In the explanation, is a part, where the concepts are shown in action. In an Example. In this repository is this example.

What is the example like?

The example is a simple CNN, which is trained to recognize the digits from 0 to 9.
I know it has been seen multiple times, but we needed a simple example, because the google docs document was so much work. We used the classic MNist dataset.

In the example you can draw digits and see the result. The predictions aren't always accurate, but very often it succeeds.
By the way, if you try to draw a 7 don't draw the line through the middle. In the Dataset it is not drawn like that.

How you can guess, the we used supervised learning and a CNN. You can see the structure of the NN in the image above.

We know, that a smaller network would work to, but for educational resosons, we used a bigger network.

Installation

1. Install Python 3.9: https://www.python.org/downloads/release/python-3913/
2. Cone the Github repository

    git clone https://github.com/Maxi1324/DeepLearningInANutshell
    cd DeepLearningInANutshell

3. create a new Virtual Enviornment

    py -m venv DigitRecognition
    .\DigitRecognition\Scripts\activate
    pip -r .\requirements.txt

4. Run the Application

    .\DigitRecognition\Scripts\activate
    py .\MINST\GUI.py

5. Enjoy!

What did we learn?

We learned a lot of things. My teammates had nothing done with AI in the past. So all the theory was new for them.

For me most things weren't new. But i still deepend my understanding of the gradient descent algorithm and backpropagation and many other topics.

Who did the what?

• Maximilian Fischer(me): The example project and most of the document.
• Adrian Schauer: In the document, the paragraph about backpropagation.
• Marko: In the document, the paragraph about AI in our Society.

Tools used in the Project?

• Tkinter
• Tensorflow
• MNist Dataset
• google :)
• python
• Google Docs

A detailed list of the sources for the google docs, can be found in the document