Most Complicated NOT Gate Ever

user interaction

Inspiration

With the recent rise of LLM neural networks like ChatGPT, we decided we wanted to go right back to the root of how those networks are made and create our own from scratch. Almost every example we were able to find of a simple neural network recommended using something like "" to simplify the process but we decided that writing our code to work completely seperate of any such assistance was the only way to go.

What it does

It creates a neural network (of variable size) in this case containing 12 nodes, with two hidden layers of 4 nodes each. It is then trained on a data set we created to teach it to work as a NOT gate. In this case it is trained on 1,000,000 mini-batches of 10 inputs each. (we would demonstrate it on a larger scale but unfortunately that would take time we simply do not have)

How we built it

All written in java in VS code in a main class and network class

Challenges we ran into

We were originally intending to connect the neural network to a CNN and scan some simple images, but due to lack of time we were unable to complete that portion

Accomplishments that we're proud of

We are quite proud of the networks accuracy. We added a couple features, including a Xavier randomized initialization function, that increase the networks function for smaller datasets