Model: 2 neurons

The first fictional neural network was created with the goal of predicting whether a table was round or not without seeing it, using only its area and radius, and without needing to know geometry
Choose a random area and radius of a table and see for yourself whether it is round or not

Architecture:
  • Framework used: PyTorch
  • Model: MLP with two neurons
  • Strategy: Custom algorithm for learning rate adjustment
  • Hardware: Entry-level GPU (NVIDIA GTX 1050 2GB), training optimized for efficiencyPantalla mostrando modelo con 4 neuronas
Questions:
  • Why doesn't the machine have total precision?
  • Why can't the machine predict what's in the box if the tables have a radius greater than 50 cm or less than 20 cm?
  • What would happen if the machine had 4 neurons instead of 2?