- A research team led by Ludwig Maximilian University of Munich suggests that quantum neural networks (QNNs) could become a tool for chemical and pharmaceutical companies to test surrogate functions.
- A surrogate function is a simplified function that is used to model processes in tasks, such as drug interactions.
- The team’s findings suggest QNN will outperform classical neural networks.
A key challenge in the chemical and pharmaceutical industries is finding cost-effective ways to model and evaluate surrogate functions that approximate those of a complex black box. In the pharmaceutical industry, surrogate functions — which are simple functions used to model processes — are often relied upon in tasks, such as drug interactions, to optimize and streamline development efforts.
Conventional methods of classical machine learning often struggle to accurately solve this problem due to the limited and noisy data sets that are common in practical applications. As a result, chemical companies around the world are excited to explore new approaches to addressing this problem.
In a Study, a research team led by Ludwig Maximilian University of Munich suggests that quantum neural networks (QNNs) offer a very promising solution. These networks, based on the principles of quantum mechanics, show the potential to outpace their classical counterparts when trained on small, noisy data sets.
The researchers conclude in what they claim is the first practical exploration of using QNNs as surrogate models for realistic, higher-dimensional data: “In this paper, we have shown that a Quantum Substitute Model based on QNNs can offer advantages over classical ANNs in terms of prediction accuracy. for data sets that are much more difficult than used in the previous literature, when sample sizes are scarce and there is substantial interference.
Through extensive experimentation, the team reports its QNN outperforms classic minimalistic neural networks when handling noisy and limited data sets. These results provide empirical evidence supporting the superiority of the quantum surrogate model.
The researchers also measured the performance of the current NISQ (Noisy Intermediate-Scale Quantum) hardware in the experiment and estimated the gate precision required to replicate the simulation results, demonstrating the practical feasibility of implementing QNN in real-world scenarios.
The implications of these findings are very important for the chemical and pharmaceutical industries, according to the team. By leveraging QNN as a surrogate model, companies can increase their ability to accurately approach complex functions, even with limited and noisy data. This has the potential to streamline development processes, leading to more efficient and cost-effective solutions.
There are limits, the team points out. While QNN is promising, the field of quantum computing is still in its infancy. The hardware and gateway fidelity required for large-scale implementation is yet to be fully realized.
The team posted their findings at ArXiv preprint server. The pre-print server allows researchers to share their findings but technically this is not a peer-reviewed study.
Team members include: Jonas Stein, Michael Poppel, Philip Adamczyk, Ramona Fabry, Zixin Wu, Michael Kolle Jonas Nußlein, Danielle Schuman, Philipp Altmann and Claudia Linnhoff-Popien, all from LMU Munich; Thomas Ehmer, from Merck KGaA and Vijay Narasimhan, from EMD Electronics.