why will multiple time training generate different result as I initialize network manually?

I have created a function fitting neural network with different input parameters so I need to compare them. But here is the problem that multiple time training with a specific set of input generates different results and there is no stable answer to compare with performance of other networks. I want to know what the problem is. As I initialized the networks with specific weights and biases and inputs. I hope to hear about my problem ASAP.

Answer 1

[Greg Heath]

Although you avoided random weight initialization by assigning initial weights, you did not avoid the random data division.

Why not initialize the RNG to the same initial state via

rng(0) net = fitnet(10);

Hope this helps.

Thank you for formally accepting my answer

Greg