Written by Deepansha Singh
Quantum machine learning is a field that lies at the intersection of quantum computing (computing that harnesses quantum physics properties such as superposition and entanglement to enable an exponential speedup) and machine learning (mimicking human intelligence in machines by learning from previous experiences using datasets).
Today, both technologies, quantum computing, and machine learning are being used to find solutions to real-world problems. However, in the future, using a combination of both technologies, quantum machine learning, could be used to provide a massive speedup in current classical machine learning solutions for various real-world problems.
Quantum Computing: A Brief Overview
To get a brief overview of quantum computing and get a better understanding of the field, I’m referring you to my previously published article on the Techvik blog (https://www.techvik.in/post/the-quantum-computing-tech-revolution).
Classical Machine Learning: A Brief Overview
Machine learning (ML) is a very common tech buzzword. From the financial services industry to have personal assistants like Siri, ML is being used in all types of industries today.
In a nutshell, ML is basically training the machine to “think” like a human by using data, which enables the machine to learn from these experiences, and in turn, draw some conclusions.
Specifically, the different types of machine learning are supervised learning (data that is labeled), unsupervised learning (data that is unlabelled), semi-supervised learning (partially labeled data), and reinforcement learning (no labeled data but instead machine gets trained by observing when it’s “rewarded”).
Quantum Machine Learning: Combining Quantum Computing & Machine Learning
Lying at the intersection of quantum computing and machine learning, quantum machine learning takes the classical machine learning algorithm and transforms it into a quantum circuit. This quantum circuit can then be run on a quantum simulator or quantum computer. To get started with creating quantum circuits and running them on a quantum simulator or quantum computer, I recommend using the IBM Q Experience platform (https://quantum-computing.ibm.com).
Quantum machine learning can improve classical machine learning in terms of the run time of the algorithm along with making the training process itself a lot more efficient by learning from fewer data altogether.
Examples of How Quantum Machine Learning Can Improve Classical Machine Learning
A speedup in the algorithm runtime using quantum computing can be done in several different ways.
One method is to construct a hybrid algorithm, where the algorithm utilizes quantum computing for some of the tasks, but classical computing for other tasks. One particular machine learning model is generative modeling, where the computer is given data related to a particular area and the objective is for the computer to generate new samples and data that is similar to this input by learning more about the distribution through extensive training. Quantum computing can provide an immense speed up in this algorithm by performing quantum sampling, which is a lot more efficient than classical sampling, which can lead to an immense speedup.
Furthermore, another popular way (which is currently being used today at several startups & companies like D-Wave) that quantum computing can greatly accelerate the algorithm runtime is by using quantum annealers. A lot of real-world applications of machine learning are optimization related and quantum annealing can perform a lot of optimization tasks a lot faster. Some specific applications of utilizing a quantum annealer to perform this optimization and lead to an immense speedup are in the financial services industry- portfolio asset identification and optimal trading trajectory are just two of the real-world use case applications.
The last method which can be utilized to improve classical machine learning solutions drastically is to use powerful quantum algorithms such as quantum amplitude amplification along with other quantum algorithms that are centered around making linear algebra tasks such as linearization a lot more efficient. One particular real-world application is in the financial services industry in the supply chain management problem, where the problem statement is to meet customer demand while avoiding the unwanted stock. This problem is generally solved using classical machine learning, where the parameters are found by minimizing the least-squares error between the training data and the values predicted by the model. This optimization is done by obtaining the inverse of the training data matrix. Using a classical approach, obtaining the inverse can take too long and can be too computationally expensive. However, in the quantum realm, this linear algebra task of obtaining the inverse of a matrix can be done pretty quickly and is exponentially faster than a classical method (as demonstrated by this research paper- https://arxiv.org/pdf/1204.5242.pdf)!
Apart from the financial industries, there are many more real-world applications of quantum machine learning which I haven’t covered in this article. For instance, a quantum computing approach can be used in quantum chemistry applications where the goal generally is to minimize complex cost functions. This speedup can be provided by quantum algorithms that perform linear algebra tasks such as eigensolving a lot more efficiently than classical methods.
Quantum machine learning will be able to handle a lot larger datasets, build more accurate and better models, and develop exponentially faster algorithms. In the near future, quantum machine learning will completely change our world in a positive way! :)