The most significant "new" way to build a neural network in Excel is via the native Python integration
We will train it using .
Go to the tab and click Solver . (If missing, enable it via File -> Options -> Add-ins). Set the Objective Cell to your Total Loss cell. Set the optimization goal to Min (Minimize).
Artificial intelligence often feels like a black box hidden behind complex Python libraries like PyTorch or TensorFlow. However, the fundamental mathematics of deep learning—matrix multiplication, activation functions, and gradient descent—rely on basic arithmetic. build neural network with ms excel new
A3: b₁₁ B3: (initial bias, e.g., -0.2)
If you are working with even larger networks, , but note that 2026 has brought powerful no‑VBA alternatives . For instance, there is now a minimal GPT model implemented entirely in Excel formulas —no VBA, no external dependencies. It unrolls the attention loop, computes queries, keys, and values for each token, and generates plausible output, all within spreadsheet cells. This shows that modern Excel is capable of far more than most people imagine.
If you want to tailor this network to a specific project, let me know: What or problem are you trying to solve? The most significant "new" way to build a
δ2=(A2−Y)⋅A2⋅(1−A2)delta sub 2 equals open paren cap A sub 2 minus cap Y close paren center dot cap A sub 2 center dot open paren 1 minus cap A sub 2 close paren Excel Formula: =(A_2 - Y) * A_2 * (1 - A_2)
Using Excel's modern matrix multiplication function ( MMULT ), enter this formula in cell D2 and drag it down: =1 / (1 + EXP(-(MMULT(A2:B2, $F$2:$G$3) + $F$4:$G$4)))
Because native Excel formulas do not automatically loop over time, running multiple training iterations ("epochs") requires automation. You can write an (for Excel on the Web/Desktop) or a traditional VBA Macro to create a loop. This script takes the updated weights from the backpropagation step, pastes them back into the weight initialization cells, and repeats the cycle until the network's error rate drops close to zero. 5. Why Excel is a Game-Changer for AI Literacy Set the Objective Cell to your Total Loss cell
Before diving into the steps, let's clarify the scope of our project. We will build a :
Set to your Weight and Bias blocks ( Weights_1, Bias_1, Weights_2, Bias_2 ). Select GRG Nonlinear as the solving method.
In cell S1 , insert your trained output: =1/(1+EXP(-(MMULT(IF(MMULT(Q1:R1, W1)+B1>0, MMULT(Q1:R1, W1)+B1, 0), W2) + B2)))
