We need the derivative of the error with respect to each parameter. Using the chain rule of calculus:
=MMULT(InputData, W1) + TRANSPOSE(b1)
Copy S1 down to S4 .
Building a Neural Network from Scratch in Microsoft Excel: The Modern No-Code Guide build neural network with ms excel new
Your journey into Excel AI is just beginning. Here are some excellent resources to guide you:
: As of late 2025, Microsoft Copilot's Agent Mode can generate the structure of a predictive model or neural network by simply describing the task in plain English. 2. Step-by-Step Build (Traditional Formula Approach)
The (η) is a small positive number (try 0.1) that controls the step size in the direction of the negative gradient. If you set it too high, the network might overshoot the optimal solution; too low, and training will be very slow. After updating all weights and biases, the new values are used for the next forward pass, and the cycle repeats. We need the derivative of the error with
Building a neural network this way strips away the abstraction of Python libraries like TensorFlow or PyTorch. It forces you to visualize exactly how data streams through matrices, math functions, and optimization loops. If you want to expand this spreadsheet model, let me know:
Choose GRG Nonlinear (Generalized Reduced Gradient), which handles non-linear equations like our sigmoid function. Click Solve.
: Use the Sigmoid function to normalize the output between 0 and 1. The formula is: =1/(1+EXP(-WeightedSum)) . Here are some excellent resources to guide you:
=(Prediction - Target) * Prediction * (1 - Prediction) 2. Hidden Layer Error Gradients ( δH1delta sub cap H 1 end-sub δH2delta sub cap H 2 end-sub
In a new cell (say O3 ), enter:
Name this range HiddenActivation .