Little Bits of Artificial Intelligence

1. Imagine This: A Smart Loan Approval Assistant

A bank wants to build a Smart Loan Approval System using AI. People apply with:

• Age
• Income
• Loan Amount
• Credit Score

The bank wants to predict: Will this person repay the loan or default? To do this, they use a Neural Network trained on thousands of past applicants’ data.

Where Do Tensors Come In?

Think of Tensors as structured containers to hold data:

• A tensor is like a fancy, multidimensional spreadsheet.
• For one applicant:

[Age=35, Income=45000, Loan=10000, CreditScore=750]

• This is a 1D tensor (vector).
• When you put 1000 applicants’ data together: it’s a 2D tensor or matrix — shape [1000, 4]

In programming:

applicants = [
[35, 45000, 10000, 750],
[28, 32000, 5000, 680],
…
] # This is a tensor (2D list)

Where Do Weights Come In?

Imagine the Neural Network is like a decision-making machine.

For each input feature (like Age or Income), the network assigns a weight that determines how important that feature is in predicting repayment.

For example:

• Maybe Credit Score is given a high weight: 0.9
• Income might get: 0.6
• Loan amount: -0.7 (high loan might negatively impact outcome)

These weights adjust themselves during training using something called backpropagation.

Simple simulation:

weights = [0.1, 0.6, -0.7, 0.9]
input = [35, 45000, 10000, 750]

weighted_sum = sum(i * w for i, w in zip(input, weights))

How Neural Network Computing Happens

Neural network computing = data flowing through layers of neurons, like:

1. Input Layer: Takes tensors (your input data)
2. Hidden Layers: Do internal math using weights and activation functions
3. Output Layer: Gives final prediction: e.g., [0.85] = 85% chance of repayment

All of this is matrix math under the hood — tensors multiplied by weights, then adjusted based on prediction correctness.

Real-Life Flow:

Python-Style Summary (No Libraries):

# Input tensor (1 person)
input_tensor = [35, 45000, 10000, 750]

# Initial weights (guessed)
weights = [0.1, 0.6, -0.7, 0.9]

# Weighted sum
z = sum(i * w for i, w in zip(input_tensor, weights))

# Simple activation (sigmoid-like)
output = 1 / (1 + (2.718 ** (-z)))

print("Predicted repayment probability:", output)

Output:

Predicted repayment probability: 1.0

• Tensor stores and transmits the data.
• Weight decides how strongly each input influences the prediction.
• Neural Network computing is the full machinery—applying weights, combining results, learning from errors, and improving predictions.

Final Thoughts

• Tensor stores and transmits the data.
• Weight decides how strongly each input influences the prediction.
• Neural Network computing is the full machinery—applying weights, combining results, learning from errors, and improving predictions.

This combo is what powers systems like:

• Loan approval models
• Credit fraud detection
• Recommendation engines
• Voice recognition