🧠 The Simplest Neural Network That Actually Works

Before tackling multi-layer or transformer architectures, I built the simplest neural network I could — a single-layer perceptron to classify 0s and 1s from the MNIST dataset.

Project Highlights:

Framework: TensorFlow + Keras

Architecture: 1 Dense layer, 1 neuron, sigmoid activation

Optimizer: SGD

Accuracy: 99.9% test accuracy

Dataset: MNIST (filtered to digits 0 and 1)

Key Takeaway:
Even a one-layer model can teach core ML principles:

Data normalization

Gradient descent

Binary cross-entropy

Evaluation with precision, recall, and F1

Explore the notebook here 👇
🔗 Simple Neural Network Project

📢 Follow my AI builds & insights:
🐦 @MarcusMayoAI
| 🧠 Dev.to/marcusmayo
| 💻 GitHub/marcusmayo
| 💼 LinkedIn