Little Bits of Artificial Intelligence

1. Imagine a Teacher Grading Papers

Suppose a student hands teacher a paper, and teacher’s goal is to predict the grade based on:

• Neatness (score out of 10)
• Correctness (score out of 10)
• Creativity (score out of 10)

Teacher is using his experience (rules he has learned) to make a decision. In a neural network, this is like going forward through the layers using known logic to produce a result.

2. What is “Feed Forward”?

Feed forward means: Taking the input, doing some math, and sending it through the network to get the final output. No feedback, no correction. Just straight calculation.

3. Step-by-Step Story

1. Input Layer – The student gives teacher scores:
   Neatness = 7, Correctness = 9, Creativity = 6
2. Weights & Biases – Teacher secretly has his personal rules:
   Grade = (Neatness × 0.2) + (Correctness × 0.5) + (Creativity × 0.3)
3. Hidden Layer (Optional) – Teacher might first convert the scores into intermediate feelings:
   Feeling = f((7×0.2 + 9×0.5 + 6×0.3) + bias)
4. Activation Function – Teacher passes the result through an emotional filter (e.g., “How good is this feeling?”):
   • ReLU, Sigmoid, etc., decide whether you’re excited or not.
5. Output Layer – You finally decide:
   Final Grade = 8.4

That’s it! Teacher didn’t change his rules (weights) yet. Just calculated and gave the result. That’s feed forward.

What It’s Not

• Not Learning: It’s not adjusting anything yet.
• Not Backpropagation: That’s when the teacher finds out he gave the wrong grade and fixes his rules.

Real-Life Analogy

Imagine a vending machine:

• The button is pressed (input)
• The circuit activates some connections (weights)
• The motor turns (activation)
• A candy pops out (output)

No learning involved. Just input → process → output. That’s feed forward.

4. How feed forward helps a neural network

What is Feed Forward, Really?

Feed forward is the first step in how a neural network “thinks”.It’s like making a guess based on what it already knows (its weights and biases), without changing anything.

So, How Does It Help?

1. Prediction Engine

Feed forward is how the neural network makes predictions.

• Example: Input is a photo of a cat.
• It feeds the pixel values forward.
• Output says: “I think it’s a cat (95% sure).”

Without feed forward → no output → no prediction.

2. Transforms Raw Data into Meaningful Signals

Raw data (like numbers or pixels) isn’t useful by itself. Feed forward:

• Applies weights to inputs
• Combines them into hidden patterns
• Produces something meaningful at the output

Like turning student scores into a grade. Or stock history into a price prediction.

3. Foundation for Learning

Feed forward provides the base output for comparison.

• We use this output to calculate error (i.e., how wrong the guess was).
• Then we do backpropagation to update the weights and improve.

Without feed forward, the network wouldn’t know what answer it gave, so it couldn’t fix anything.

4. Used in Both Training and Testing

• During training → Feed forward → Compute error → Learn
• During testing → Feed forward only → Predict the result

It’s like solving a problem, and then checking if we were right.

5. Layer-by-Layer Abstraction

• Learns a bit of the pattern
• Passes it on to the next layer
• Final layer gives the result

Example:

• Layer 1: Detects lines in an image
• Layer 2: Combines lines into shapes
• Layer 3: Sees it’s a cat

Summary: How Feed Forward Helps