Key players insights


Quantum Computing has received largely increased attention and investments in the past years and this growth is expected to be even greater in the near future. The effort is directed towards improving both on the software and on the hardware, where for the latter it is known that current devices show some limitations and are not yet suited for most use cases that Quantum Computing aims to solve – either more efficiently or altogether.

However, Reply’s expertise and focused attention to customers’ business performances has led to exploiting Quantum Computing approaches to create value already today in production, while Quantum hardware still needs to be improved and solve computational problems more efficiently and effectively.  
MegaQUBO, Reply’s Quantum-Inspired accelerator to solve large combinatorial optimization problems formulated as QUBO (Quadratic Unconstrained Binary Optimization), has been widely used to combine a Quantum Computing formalism with the high-performance hardware already available in order to bring business value while Quantum devices are being improved.

The next sections illustrate two major success stories on solving combinatorial optimization problems from different fields and industries and how Reply has employed a different approach, universal quantum computing, for Quantum Machine Learning.


A multinational energy company, a major integrated player in the global energy, gas and renewable energy markets, has benefitted from Quantum Computing and the QUBO formulation to minimize costs by making more efficient use of their resources.
The goal is to optimize the planning of maintenance work carried out by the units operating in increasingly large territorial areas. A large number of interventions needs to be performed by a finite number of crews. Overall, internal processes require continuous improvement and impressive technological infrastructure, hence posing great organizational challenges.

With MegaQUBO, Reply has solved the complex combinatorial problem quantistically formulated as a QUBO and on GPUs the execution of the algorithm has achieved a significant speed-up with respect to the existing solution, while also greatly enhancing the quality of the solutions found, thus improving the KPIs.

It was not long after that that the appreciation for the innovative Quantum Computing approach was shared also by the company [1] [2].


One of the biggest telco operator in Italy has resorted to Quantum Computing and QUBO to find the optimal configuration of Physical Cells Identifiers (PCIs) of radio cells so that it would be possible to optimize the 4.5G and 5G networks planning

Dealing with interferences require a complex mathematical formulation that conventional solutions can hardly handle. On the other hand, the QUBO model from Quantum Computing was a highly suitable fit that in turn paved the way for finding high quality solutions and thus increased business performances.

Together with Reply, the company has been able to search through the potential combinations that would solve the problem and find one that greatly improved the KPIs with respect to the existing solution, while still being able to significantly reduce the computational time required by the algorithm.

Early Adaption to Quantum Machine Learning

For applications in machine learning, this approach does not work so easily as the underlying problem is not readily formulated as a QUBO. Therefore, quantum machine learning requires so-called universal quantum computers. As those are still difficult to scale to many qubits, resource economical algorithms are needed, which can get the most out of such “noisy intermediate-scale quantum devices”. It turns out that enriching classic machine learning algorithms with elements from quantum computing can greatly improve their performance. The experiments of Reply have shown an increase of about 5% in accuracy for some algorithms. Neural networks, for instance, are ubiquitous in modern machine learning and they have applications to image analysis, natural language processing or the modelling of sensor data.
In their contribution to the quantum computing challenge hosted by car manufacturer BMW and cloud provider AWS – who already make some quantum hardware usable as part of a cloud workflow – the team of Reply has enriched a neural network trained for image classification with a quantum layer. They could thus demonstrate a full machine learning workflow for quality control in manufacturing already with a few qubits, in fact with a resource need less than what current quantum hardware makes available. Even better, this architecture is versatile and can be adapted either to a different problem, like time series analysis, or even cope with multiple modes of data as obtained by for instance adding some text to the pictures which have to be evaluated. The jury found this approach to be so promising that the team of Reply has reached the final of this challenge.

From Optimization to Machine Learning, from Cybersecurity to simulations in Finance and Chemistry, Reply is currently investigating even more potential use cases and aims to consistently benchmark various approaches – quantum, classical and hybrid – against each other to deliver the best solution to their clients.
To learn more about projects, proof of concept, and research activities carried on by Reply and how your business could already benefit from Quantum Computing solutions, visit the link