Linear Inseparability

Definition

Linear inseparability means no single linear boundary can separate the classes correctly.

In 2D:

no single line works

For XOR:

class 0: [0,0], [1,1]
class 1: [0,1], [1,0]

XOR is linearly inseparable in its original input space.

---

What Linear Separability Would Require

Linear separability would require one line such that:

class 1 points are on one side
class 0 points are on the other side

For XOR, that means:

[0,1] and [1,0] together
[0,0] and [1,1] together

No single line can do that.

XOR requires separating opposite corners from opposite corners.

---

Why One Line Fails

The XOR classes alternate around the square.

Move around the square:

[0,0] -> class 0
[1,0] -> class 1
[1,1] -> class 0
[0,1] -> class 1

The class changes at every corner.

A single line can split the square into two sides, but it cannot alternate labels around the corners like that.

A straight line creates two regions, not alternating corner regions.

---

Boundary Shape Problem

A linear classifier produces a flat boundary.

In 2D:

line

XOR needs a decision structure that acts like:

positive in two separated regions
negative in two separated regions

One line cannot create two separated positive regions.

XOR requires more structure than one flat boundary provides.

---

Not A Training Failure

If one perceptron fails on XOR, it is not just because training was poor.

Even perfect weight search cannot find a single linear boundary that solves XOR.

The needed boundary does not exist in the original 2D representation.

The failure is representational, not just procedural.

---

Connection To Features

Linear inseparability depends on the feature space.

In the original features:

x1, x2

XOR is not linearly separable.

But if we create a better representation, the classes may become separable.

Changing representation can change separability.

---

Why It Matters

Linear inseparability is the reason XOR became important historically and conceptually.

It shows:

simple linear classifiers have real limits
representation can matter more than the final classifier

XOR forces the shift from linear decision rules to representation transformation.

---

Final Summary

• Linear inseparability means no single linear boundary can separate the classes.
• XOR is linearly inseparable in its original 2D input space.
• XOR classes occupy alternating corners.
• One line cannot create the needed separation.
• The perceptron failure is not just a training failure.
• A different representation can change the problem.