Quantum Machine Learning: Week 1
Introduction
Quantum Machine Learning (QML) is a hyped field with seemingly many applications. To gain more insights into the field, as well as spend my summer somewhat more productive, I decided to take part in the Qiskit Global Summer School about QML. This lasts from the 12.07.21 to the 23.07.21. In the following blog post, I’ll try to summarize some things i learned.
Day 1
Day 1 started off with just a general introduction to quantum, first the bra-ket notation followed by gates, and quantum teleportation. A good refresher to use linear algebra again. However, without previous knowledge of the material, I would have really struggled with the material as it was quite fast-paced…
Day 2
Again, a bit of refreshing material. This time, it was about the Deutsch-Jozsa algorithm and Grover’s algorithm. Again, I covered these two before. However, this time, it helped quite a lot to revisit the material. Especially for Grover’s algorithm, my intuition for the way it works increased significantly.
Day 3
Day 3 was definitely a step up compared to the previous days. While the concept of noise in quantum operations and measurements is easy to grasp, I was struggling with the mathematical background a bit. I will definitely revisit this lecture in the future.
Day 4
After the difficulty with Day 3, it’s refreshing to have an easier day with classical machine learning. We mostly covered material such as what the task of machine learning is, and how e.g. SVM works. This set the stage nicely for what would greet me in Day 5…
Day 5
Full push for the last day of the week. After all the pre-requisites have been introduced, we are finally discovering the world of Quantum Machine learning to build a classifier. We started off with a bit of background on how we can encode the (classical) data we have into a quantum state. Here there are multiple options. However, the most powerful construct here is probably the so-called Ansatz, offering a way to parameterize a quantum circuit. Using well-studied Ansatz(‘s?), we can then manipulate the quantum state and extract the label by repeatedly measuring the quantum state (no-cloning theorem and all, so we have to run this circuit multiple times and extract a probability distribution in that way). The performance of the classifier can then be tuned by changing the parameters of the Ansatz, using similar ideas as gradient descent. I will have to revisit this lecture as well, since the way in which the Ansatz is used feels a bit like magic… But what is not magic in quantum…