Functional Magic Behind an AI Model.

Is it possible to explain the magic behind the functionality of an AI model simplistically???

Yes, It is just a mathematical model (neural network) that "figures out" the function on its own, during training of the model, based on the data that we feed it upon. Whenever, then, you are given an input and the model will give you back an output by performing "that" function.? Don't you believe in it ?

Let me create a very simple "neural network" with only one layer that accepts 10 numbers as input and a scalar value as an output.??

I train this "neural network model" with input data of random numbers but ensuring that the output should be "5" as it is my lucky number.?

I chose SGD (Stochastic Gradient Descent) optimizer for back-propagation which is the backbone for the model to come up with a relevant function. I ran the training 200 (epochs) times.

That is, it. Run the following code and observe the results.

import torch 
import torch.nn as nn 
import torch.optim as optim 
import torchmetrics  
	  
model = nn.Linear(10,1) 
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.1)
criterion = nn.MSELoss()
    
target = torch.tensor([[5.0]])

for epoch in  range(200):
    optimizer.zero_grad()
    x = torch.rand(size=(1,10), requires_grad=True)
    output = model(x)
    loss = criterion(output, target)
    loss.backward()
    optimizer.step()
	    
torch.save(model,"layer.pth")

 #Load the saved model back
loaded_model = torch.load("layer.pth")
    
# Regression metrics
mse_metric = torchmetrics.MeanSquaredError()
mae_metric = torchmetrics.MeanAbsoluteError()

def  test_model():
    input_vector = torch.rand(size=(1,10), requires_grad=True)
    prediction = loaded_model(input_vector)
    target_result = target
    mse = mse_metric(prediction, target_result)
    mae = mae_metric(prediction, target_result)
    print(f"Prediction: {prediction[0][0]}, Target: {target_result[0][0]}")
    print(f"MSE: {mse}, MAE: {mae}")
    print(f"="*40)

for i in  range(10):
    test_model()        

NOTE:?

Of course, you will not get the output exactly as 5 but values closer to 5 in most of the cases. In order to improve the accuracy of the output, the model requires to be trained with a considerable number of inputs. In reality, the neural networks will have very many layers and the size of the dataset will be huge. But, I hope that you would have got an idea about the "magic" behind the AI models.