top of page

Wait, but HOW do machines .... learn? Part - 2

Updated: Nov 28, 2023

In the previous post, we saw how we can use "machine learning" to solve simple numerical problems involving multiplication. We also began to understand the structure of the human neuron.

Below is a the computer model of the neuron. It's called the perceptron, and it's very simple. Each perceptron has 2 or more inputs. Think of it as music coming from multiple telephones. Each input is a telephone line, and the final volume inside the neuron is the result of all the music from each telephone line playing at once. Once the inside of the neuron gets "loud" enough, it too starts producing a tone of music. If the inside of the neuron gets quiet below a threshold, it will stop producing the tone.

A neural network perceptron, with 3 incoming lines for input.
A neural network perceptron

But there is a catch. Notice the dials on each of the incoming telephone lines. That's where the learning takes place.

How does the neuron know which knob to turn up and which to turn down? Well, each neuron has a function. It's purpose in the whole mind. Each neuron has it's own purpose and is connected to the neuron which are relevant to that purpose. It fires when it detects the right conditions.

Lets assume that our neuron is responsible for finding Rock Music. So if we think of the music coming from the 3 telephone lines, as 1. Rock and Roll FM, 2. Sweet Jazz FM and 3. Hip-Hop FM, over time, it should increase the volume of the first input (Rock and Roll FM) over the other 2 telephone lines.

So, "learning" for a neuron essentially is learning the volume dial setting for it's incoming signals. The loudness of the music from the telephone lines gets "multiplied" by the setting of the dial. Even if Hip-Hop FM is playing at full blast, the volume knob would be turned down to not impact the loudness in the neuron, and hence it does not fire. Whereas the knob dial on Rock and Roll FM is set to boost, where even a whisper coming from that telephone line will boom across the neuron, and make it fire.

The mechanism for learning is simple. Imagine you are the neuron. You're sitting in the room listening to the telephone lines.

A man sitting in a room listening to music.
A man sitting in a room listening to music, trying to find rock music.

Music is played across the 3 telephone lines, and you are told when rock music is being heard. Listen to the telephone signals and give a prediction based on the music you hear. Maybe it's hip-hop, maybe it's rock. Now let's say you're told that rock music is indeed, playing.

Move the knobs for each line, one by one, a little bit each time. Note the impact. Do you hear rock music more or less? If turning a knob down brings you closer to the correct answer, do it. If turning the knob up brings you closer to the correct answer, then do that. Keep doing this until you only hear rock music and very little of other music.

And that's what machine learning is. The rest of the mathematics and data-science is just optimizations for speed and quality.


Core Maitri is an enterprise software consultancy specializing in Excel-to-Web, AI Integration, and Enterprise Application Development services. Our approach is deeply consultative, rooted in understanding problems at their core and then validating our solutions through iterative feedback.

Recent Posts

See All


bottom of page