Handwritten Digits Classification but with a twist

I recently completed a personal project where I trained an ANN(Artificial Neural Network) to classify handwritten digits from the MNIST dataset.

Use this link to read the story if you’re not a member: Link

Sounds pretty basic for a final-year student, right? But there’s a twist — I implemented the network entirely from scratch, without using any ML libraries like TensorFlow or PyTorch. Just pure Python code and math. Although, to perform calculations efficiently, I used NumPy. Additional libraries included (which were not used in the actual “training” of the network):

• Pandas- For loading the data
• Matplotlib- For plotting loss and accuracy curves
• tqdm- For tracking model training progress

Why This Project?

We all know how the world was taken by the AI wave and I too was caught up by it. To avoid the FOMO, I quickly picked by ML libraries like sklearn and Tensorflow. I was calling myself an “ML Engineer” just by importing a model and training it with the .train() method without even understanding what was going on under the hood, and how this algorithm even works? After making the “world-changing,” “revolutionary,” “eureka” project of “Iris Flower Classification,” I was under the delusion that…