Visualization
Assuming the weights and biases are in cells E2:E7, and the inputs are in cells A2:B5, the formulas would be:
No TensorFlow. No PyTorch. Just cells, formulas, and gradient descent. build neural network with ms excel full
δout=(ypred−Y)⋅ypred⋅(1−ypred)delta sub o u t end-sub equals open paren y sub p r e d end-sub minus cap Y close paren center dot y sub p r e d end-sub center dot open paren 1 minus y sub p r e d end-sub close paren Excel Formula: = (K11 - C11) * K11 * (1 - K11) Hidden Layer Error Gradients
Excel's Solver engine will run backpropagation iterations behind the scenes, rapidly adjusting your parameters until the Total Error drops near zero. 6. Verifying the Results Once Solver finishes, look back at your training table. Compare your target outputs ( ) to your predictions ( Visualization Assuming the weights and biases are in
Building this system entirely within Excel strips away the abstractions of high-level programming. Seeing backpropagation update cells via native sheet formulas demystifies deep learning, turning abstract algorithms into clear, visible arithmetic.
For hidden bias B7 : = B7 - $M$1 * (2*(G3 - H2)*(G3*(1-G3)) * (C3*(1-C3)) ) — wait, the bias derivative doesn't multiply by an input. It's just the delta of the neuron. Compare your target outputs ( ) to your
Building a Complete Neural Network From Scratch in Microsoft Excel
Debugging tips