Remember your first programming tutorial? It probably began by teaching you how to print "Hello World" to the terminal. The equivalent for AI is to begin with classifying handwritten digits with the MNIST dataset or classifying clothes with FashionMNIST.
While these tasks appear simplistic, having datasets like this is incredibly important. They act as a baseline for the experimentation done by researchers and allow us to compare the strength of approaches or delve more into the nature of neural networks.
Sample data and labels from the MNIST dataset.
MNIST (Classifying Hand-written Digits)
Experimenting with the
steps used in backprop
FashionMNIST (Classifying Clothing Items)
The first hackathon I went to after learning about the foundations of AI was WolfHacks 2019. My friends and I were the runner up in this hackathon for building a waste management solution. Inspired by how centrifuges use centrifugal force to separate components in a fluid, I thought of a rotary sorter that would do the same for waste. Through a combination of sensors and computer vision, waste could be correctly classified and placed into an appropriate bin for collection. Much of the principles used to classify hand-written digits or clothes carry over in other applied AI projects. The only thing that changes is the scale of the dataset, complexity of the model, and the use-case of the model in a greater system.
Later that year at Hack The North 2019, my friends and I noticed that there were a lot of potholes and irregularities on the road. On our drive to the hackathon we started thinking of ways this could be improved and came up with an idea that leveraged AI, hardware, a cloud database and a publicly available website. Our idea was to build a pothole detector that could passively run from the dashboard of cars or government owned vehicles such as buses in the public transit system. We trained a detector to identify potholes and then deployed it on a Raspberry Pi with a GPS module plus a camera. Every time a pothole was identified, an image and a location marker was dropped on a publicly available map. I was amazed at the potential of AI to solve such massive problems. A simple neural network could become another pair of eyes on the road and watch over our public infrastructure to ensure it's usable for all.