What can one weighted signal separate?
Compare a single threshold with a two-hidden-unit network on all four XOR inputs.
One unit thresholds a weighted sum. The two-unit network computes OR and NAND, then ANDs their outputs.
Cut these cards apart. Use the original example first, then let a partner supply a changed case. Keep sealed/final cards with the teacher until the learner commits a rule.
Input A: 0
Input B: 0
Expected XOR: 0
My prediction and reason:
Input A: 0
Input B: 1
Expected XOR: 1
My prediction and reason:
Input A: 1
Input B: 0
Expected XOR: 1
My prediction and reason:
Input A: 1
Input B: 1
Expected XOR: 0
My prediction and reason:
The hand-designed XOR network demonstrates representation capacity, not a training result.