Quantum: the key to self-supervised machine learning

What is it about?

In recent years, machine learning (ML) has revolutionized scientific studies. The key to these successes is deep learning, which uses many artificial neural network (ANN) layers stacked together. But deep learning has a problem: the results depend on how well the ANN has been trained. Usually, a human would need to label and sort all the training data. This is called "supervised learning". The problem with supervised learning is that it takes a lot of work and time to label training data. As ML needs grow, the sheer volume of labelling will become too great. Self-supervised learning, however, can help solve this problem. Instead of reading human labels, a model takes advantage of the structure in the data itself to learn. However, self-supervised learning is a significantly more complex task. Finding complex correlations requires more training data, more time, and larger network capacity. Classical hardware is unlikely to support the growing future needs of self-supervised ML. According to a latest study, quantum neural networks (QNNs) could be the solution to this issue. QNNs might be more suited to self-learning algorithms and could allow more powerful architectures.

Featured Image

Photo by and machines on Unsplash

Photo by and machines on Unsplash

Why is it important?

ML has led to breakthrough results in fields like visualizing protein folding, imaging black holes, and treating heart disease. In this latest article, scientists test quantum-classical hybrid ML methods. Their results show an advantage for ML using small-scale QNN over classical networks. It shows that the best current quantum model is already equally accurate to the equivalent classical model for some applications. Classical ML has solved many non-visual problems in biology and chemistry. Quantum self-supervised learning can be applied to complex quantum problems in the natural sciences. KEY TAKEAWAY: Quantum computing could propel the development of self-supervised machine learning methods. QNNs have already shown they can match the accuracy of their classical counterparts. This could have far-reaching applications across the natural sciences.