The tutorial about the AI pipeline deals with the topic of no-code AI throughout and teaches all the necessary basics to be able to use artificial intelligence yourself without having to be a scientist or developer. In this tutorial everything is explained so that you understand the basics and can use an artificial intelligence yourself. Therefore, as an author, I usually present projects where we roll up our sleeves together and get started. I then guide you step by step to success. Also in this tutorial about the No-Code AI Pipeline, it goes straight to the topic of AI without drifting into theory. There are already plenty of very good books on the topic of AI, which generally approach the complex topics of artificial intelligence theoretically. I will link a few of them here, but for me the focus is on doing it myself. The theory can follow later if everyone is interested.

Table of Contents

Open Source and No-Code

The project presented here, the Open Source No-Code AI Pipeline, offers a simple and comprehensive introduction to the topic of artificial intelligence (AI) with a focus on image recognition. This way you will quickly get a personal impression which problems you might be able to solve with an artificial intelligence or a neural network or not yet. The target groups are interested private persons like pupils, students or simply NERDs as well as companies that want to open up the topic of artificial intelligence in a simple and cost-effective way.

No-Code AI Pipeline overview

With the implementation as a project presented here, it is possible to concentrate on the essentials of artificial intelligence. The most important thing is the data needed for the training of the neural networks. Because within the AI pipeline already well pre-trained neural networks are used. Using pre-trained neural networks is the standard and saves us not only time but also a lot of energy or power costs. On top of these already pre-trained networks, we add our professional requirements for image recognition, such as the recognition of specific objects. This includes, for example, the recognition of small plastic game pieces, plants, screws, land mines and much more that can be represented in images.

The workload for the acquisition and preparation of the image data for the training of the neural networks in relation to the effort of configuring the parameters for the training of a neural network is divided in a ratio of about 90:10. That is, 90% of the time is required to provide good data so that the neural network can efficiently and correctly learn its domain requirement. Only 10% of the time is used to change parameters of the neural network in order to optimize the trained network. The training itself runs in the background on a server under your desk or in your own data center depending on the available possibilities.

The three components of the AI pipelin

All components presented here are publicly available and have been published as open source projects on GitHub. The corresponding channels on GitHub offer help with problems and you can give a lot back to the community by improving the software and ideas for new features.

No Code AI Pipeline Prozess Beschreibung

In my case, it is a larger self-built Deep Learning computer under the desk with whose support and assistance from work colleagues at BMW I was able to write this tutorial. Behind the software of the AI Pipeline is in large parts the Tech-Office Munich of the car manufacturer BMW from Munich.

About the author

Ingmar Stapel studied Computer Engineering at the University of Applied Sciences in Würzburg and is currently working internationally in the field of innovation, i.e. topics that will shape our working world in the future; in addition, he has been intensively involved with the Raspberry Pi and robotics for years. He likes to share this knowledge with interested people from the hobbyist scene at lectures on robotics. On his private blogs, he also writes about many current technology trends and reports from his travels.