Little Bits of Artificial Intelligence

1. Story: “Chef Neural and the Recipe Tuning”

Imagine a chef named Neural who is baking cakes using a machine (neural network). Chef Neural doesn’t bake the cake himself but sets the recipe knobs—how much flour, sugar, how long to bake, etc.—and lets the machine do the baking.

These recipe knobs are not learned by the baking machine. Instead, Neural chooses them based on experience or testing. In the world of neural networks, these knobs are called hyperparameters.

Here are some recipe knobs (hyperparameters):

• Number of layers in the network (how many layers of mixing?)
• Number of neurons per layer (how much batter to mix?)
• Learning rate (how fast to mix ingredients?)
• Batch size (how many cakes to bake at once?)
• Number of epochs (how many times to re-bake with same recipe?)

If Neural sets these wrong—say, too much heat (high learning rate), too many cakes at once (large batch size), or not baking long enough (fewer epochs)—he ends up with burnt, undercooked, or inconsistent cakes (poor neural network performance).

2. Why Hyperparameter Tuning is Important?

• Better Accuracy: Right tuning gives better predictions.
• Avoid Overfitting/Underfitting: Prevents model from memorizing or ignoring data patterns.
• Optimized Training Time: Balances speed vs performance.
• Model Stability: Ensures convergence and stable learning.