The world is filled with examples of things that people are good at, but computers can’t quite be programmed to do. Writing music that sounds nice, mimicking the style of a painter, or even telling the difference between a cat and a dog by looking at a picture; all of these are doable for humans (some even very easy), but almost impossible to design algorithms for.

Machine learning seeks to fill this gap in our computational abilities, by gathering as many examples of the desired behaviour as possible, and searching an immense computational space for a mathematical function which will mimic that behaviour. For many commercial applications, this requires a large amount of computing power.

There are a few popular machine learning techniques that require us to solve many large, sparse, linear systems. We learned in the previous lesson that this can be sped up significantly using the HHL algorithm on a quantum computer. The central obstacle to this approach is the data input problem: it is not yet known how to input the large noisy datasets used in machine learning into a quantum computer.

Main takeaways

• Machine learning is the discipline of producing mathematical functions that replicate patterns from large amounts of data.
• Many machine learning techniques rely on more basic tasks such as linear system solving for their functioning. These linear equations can be efficiently solved using the HHL algorithm.