neuroAImers

• 

  Gameplay Footage

• 

  Neural Network Visualizer

• 

  Data Generator

In a world where AI is consistently evolving, and becoming harder to truly grasp and understand, neuroAImers AIms to make neural networks accessible to young minds, allowing them to become the Trailblazers of Tomorrow™.

Inspiration

We were inspired by educational games such as Blooket, Kahoot, and Prodigy, which effectively combined appealing gameplay with educational content. However, there are little to no introductory-level games for children that teach them the basics of neural network training.

What it does

The game revolves around surviving endless waves of various enemies, and progressively improving the neural network which aims and shoots at those enemies. The player is limited to moving and deciding when to shoot, making the training process an integral part of the game. Upon defeating enemies, the player will receive coins which can be used to buy epochs to train their AI. Additionally, the player will be able to generate training data for their AI, which will emphasize the importance of both quality and quantity of training data.

How we built it

The entire game was written in pure Python, making use of tools such as numpy and pygame. Numpy was used to optimize the neural network training process, making heavy use of vectorization. Pygame was used to create the minimalistic yet appealing visuals, including the neural network visualizer.

Challenges we ran into

It was difficult to integrate AI into a game. Many hours of work were spent deciding what the input and output of the neural network should be. For example, we initially thought of passing the enemy positions directly into the neural network and outputting a real number angle, but throughout the development process, we realized that passing in a low-resolution image and outputting a rounded classification of the angle was more effective. It is interesting how our final product has connections to the classic digit recognition neural network, which also categorizes low-resolution images.

Additionally, the short time frame of this hackathon required us to efficiently learn and implement relatively new concepts such as neural networks, data generation, and GUI. Much of the time was spent debugging our implementation, but with perseverance and determination, we are proud to present a finished version of neuroAImers.

What's next for neuroAImers

In the future, we plan to add a tutorial feature to allow children to independently play this game without a mentor, and learn how to make meaningful impacts in our futures. We also plan to add another tab to the shop where you can customize the activation functions in your neural network, to give players a more holistic understanding of every aspect of a neural network. Finally, adding more enemies and upgrading the visuals will make the game more appealing to children.

Built With

• numpy
• pygame
• python