Repository navigation
Version 1.21.0.0
This release adds a single-input artificial neuron to the RC2014 program collection — because apparently implementing neural-network fundamentals on a Z80 in Microsoft BASIC seemed like a perfectly reasonable thing to do!
It is deliberately the smallest useful neuron: one numeric input, one weight, one bias and a step activation function. There is no training yet; the purpose is to expose every stage of a forward pass and make the underlying calculation visible.
What's new
- Single Input Artificial Neuron — implements a simple artificial neuron using:
- One numeric input
- One weight
- One bias
- A step activation function
- Shows the complete forward-pass calculation:
- Weighted input
- Bias
- pre-activation value
- Final output
- Includes worked examples demonstrating how weight and bias determine the activation threshold
- Provides experiments for changing weights, biases and activation behaviour
Hardware variants
Three versions are included:
neuron.bas— terminal-only implementationneuron_io.bas— adds binary output using the RC2014 Digital I/O cardneuron_io_lcd.bas— adds both Digital I/O LEDs and output on a 16×2 LCD
With the Digital I/O versions, the neuron's binary output becomes physical: output 0 switches the LEDs off, while output 1 illuminates LED 0.
The LCD version goes one step further and displays both the neuron output and its binary representation.
Learning rather than learning
The neuron itself does not learn yet. Its weight and bias remain fixed during execution.
That is intentional.
The aim of this first implementation is to make the mechanics of a forward pass completely visible before moving on to multiple inputs, loss functions, training and eventually multilayer networks.
Which makes the whole exercise gloriously pointless from a practical computing perspective, but rather useful from a learning one.
And there is something particularly satisfying about using a Z80 computer to explore one of the fundamental building blocks of modern machine learning.