Machines That Transform Space
Machines That Transform Space
So far, our vectors have simply existed. Each vector represented a student. A customer. A song. A photograph. A sentence. But what if we wanted to change them? Not one vector. Every vector. At once. Imagine opening a photograph on your phone. You zoom in. Rotate it. Stretch it. Shrink it. Flip it. The picture changes. But the object remains the same. Something has transformed it. Machine Learning asks the same question. How do we systematically transform an entire space of vectors? Suppose every house in our dataset needs its prices doubled. Every student's marks must be normalized. Every image must be rotated. Every GPS location must be transformed into another coordinate system. Doing this one vector at a time would be painfully inefficient. Mathematics looked for a universal transformation. It found one. The Matrix. Most textbooks introduce a matrix as a rectangular arrangement of numbers. That is true. But it hides its real purpose. A matrix is not just a table. A matrix is a transformation machine. It accepts a vector. Performs a transformation. Produces another vector. The input changes. The rules remain the same. Different matrices perform different transformations. One stretches space. Another compresses it. Another rotates it. Another reflects it. Another projects it onto a lower-dimensional surface. The numbers inside the matrix are simply instructions telling the transformation how to behave. Every matrix is a machine. Every vector is its raw material. Now imagine repeatedly applying the same transformation. Take a vector. Transform it. Transform the result again. And again. And again. Surely every vector should behave differently. Surprisingly... they don't. Something extraordinary begins to happen. Some vectors keep changing direction after every transformation. Others refuse. No matter how many times the transformation is applied... they continue pointing in exactly the same direction. They only become longer... or shorter. It's as if these vectors have discovered a hidden path through the transformation. A path the matrix cannot bend. Only stretch. Or compress. Mathematicians found this behaviour so remarkable that they gave these vectors a special name. Eigenvectors. And the amount by which they stretch or shrink... became known as the Eigenvalue. This raises a fascinating question. Why do certain directions remain unchanged while every other direction twists and turns? And more importantly... could those special directions reveal the most important structure hidden inside our data? That question leads us to one of the most beautiful discoveries in linear algebra. U1. 8 — Why Nature Has Preferred Directions.