Road Signs & Machine Learning: Day 3

By: Abhi Mora

More research into teachable machines

For today's research, I wanted to dig a little bit deeper into the teachable machine and maybe see the Python code that it uses. However, I tried my hardest and was still unable to find this. However, to combat this, I decided to make the teachable machine much more accurate, and once I did that, I would research how Machine Learning works in the background, which I would do through our trusty search engine, Google.

One thing I realized yesterday when I was researching was that AI was unable to be one hundred percent sure about the stop signs and no outlet signs. Every once in a while, it would be like 25% that the sign was something other than what it was. If we want to start adding AI into the real world, the only way regulation would allow that would be through getting many more images. But do people have the ability to manually input millions of images?

Well, that is generally not the case. As I researched, I came across the Sidecar Pattern used by software developers around the world. This pattern is usually used in microservice environments and is used to aggregate large sets of data into a smaller set so that it can fit into the file. This seems to be a hack in the system for anyone, and I wonder how the code can do that. So that is what I did.

Through my research, I have understood that the sidecar design pattern allows programmers to add functionality and more capabilities to the program without adding additional configuration when adding other components. This is what I believe Artificial Intelligence uses, especially the teachable machine. The programmers just added the functionality for the program to find points of some images to find similar points, which then adds its capability to identify the image that we added to the teachable machine.

Although I didn't do as much related to teachable machines today, I have learned a ton about how the data is incorporated into Artificial Intelligence, as well as how it can determine and learn using large sets of data, making the process much easier for future engineers.