AI Classification

Imagine a tunnel with several filters inside it:

1. one filter is cupboard with some holes in it
2. one filter is a pile of pillows where you could squeeze through

Small animals like a cat can pass through all filters. Large animals like a dog cannot.

If something comes out of the tunnel, it must be small (cat → 1) If nothing comes out, it was too large (dog → 0)

This is AI classification.

The architecture defines what filters are used.

Example: Classification in Code

Here is how the tunnel classifier works in code:

import torch
import torch.nn as nn

class TunnelClassifier(nn.Module):
    def __init__(self):
        super().__init__()

        # Each layer acts like a filter in the tunnel
        self.filters = nn.Sequential(
            nn.Linear(10, 8),   # first filter
            nn.ReLU(),

            nn.Linear(8, 4),    # second filter (stricter)
            nn.ReLU(),

            nn.Linear(4, 1)     # final decision
        )

    def forward(self, x):
        x = self.filters(x)
        return torch.sigmoid(x)  # output between 0 and 1

How Data Flows Through

Input data → Filter 1 → Filter 2 → Decision

• Input (x): animal
• Each Linear layer: a filter in the tunnel
• ReLU: blocks values that don't fit
• Final output:
  • close to 1 → cat
  • close to 0 → dog

If the data can pass all filters → it reaches the end → cat (1) If it fails → signal collapses → dog (0)

Example Inference

model = TunnelClassifier()

# Example animal features
cat_like = torch.rand(1, 10)   # fits filters
dog_like = torch.rand(1, 10) * 0.1  # likely blocked

print("Cat prediction:", model(cat_like).item())
print("Dog prediction:", model(dog_like).item())

In the code example above, torch.rand(1, 10) represents 10 features of an animal.