Single-Layer Perceptron and Simple Thinking in Philosophy

I accidentally came across the single-layer perceptron again. To deepen my understanding, I wrote code to test it. Its computational speed is indeed very fast. Using the Threshold activation function, it recognized the first thousand handwritten sample images with extremely fast speed. But for data that has already been trained, the effect is also only average, around 77%. For data that has not been recognized, it is almost unbearable--almost all wrong. This is the characteristic of the perceptron. It performs linearly separable operations at low-dimensional levels, but when encountering linearly inseparable problems, such situations arise.

This can also be abstracted to reality and philosophy. The single-layer perceptron is like being on a plane, attempting to divide things into regions by drawing lines, and this division is either-or. If there are many classified items mixed together, there must be a function that conforms to the target's characteristics to better identify things. But this is often impossible in reality.

If one only classifies matters at extremely low dimensions--that is, being rigidly dogmatic and believing everything is either-or--then no matter how much correct understanding of objective things one has, after education, errors will still abound. Therefore, considering only a single division condition when thinking about problems cannot be established.

The understanding of yin and yang is also like this. Yin and yang are always relative. This relativity is actually assigning different weights to yin and yang in different environments and conditions. But merely doing this is not enough. If one limits division to a single aspect and uses yin and yang to understand things, one will inevitably encounter great problems.

In cases like the perceptron, directly judging based on the characteristics of known things, determining how much they differ, and continuously adjusting one's understanding--while this is beneficial for correctly understanding things, it also brings strong empiricism. The result of empiricism is that it becomes difficult to accept new things, always leading to incorrect judgments.

In most cases, classification errors do not mean that objective things cannot be classified, but rather that incorrect classification methods lead to errors. This point must be clear.