A model is a function that takes an input and produces an output, and it contains internal settings that determine exactly how that transformation happens. These settings are not fixed in advance; they are the parts that change during learning. Picture a simple model that decides whether an email is spam by combining a few signals: how many suspicious words appear, whether the sender is known, and whether the message contains a link. Each signal is multiplied by a weight, and the weighted signals are added together to produce a score. If the score is above a threshold, the model says spam; otherwise it says not spam. The weights and the threshold are the adjustable settings. A large positive weight on suspicious words means the model treats those words as strong evidence of spam; a weight near zero means the model ignores that signal. Learning, in this picture, is the process of finding weights and a threshold that produce the correct output on the training examples. This page shows the structure of such a model and how changing one weight changes its behavior, but it does not yet explain how the settings are found; that comes in the next chapter.
Learn MotionHow Machines Learn from Data
Learning Without Being Told the Rules
The Model's Adjustable Settings
3 / 3
Follow the diagram from left to right. Three signals enter the model: the count of suspicious words, whether the sender is known, and whether the message contains a link. Each signal is multiplied by a weight, and the weighted values are added into a single score. That score is compared against a threshold to produce the final answer. The weights and the threshold are the adjustable settings, and they are the only parts that change when the model learns. Notice the weight on suspicious words is currently large, so that signal dominates the score. If that weight were near zero, the model would ignore suspicious words entirely and rely on the other signals instead. The structure stays the same; only the settings move.
0:00 / 0:00