Combinatorial optimization and machine learning are computational problems that are found everywhere in our society. They are highly relevant for everyday tasks such as traffic flow optimization, video surveillance, search engine optimization or the planning of shipping routes and are thus essential to ensure that our society runs as efficiently as possible. Most of these problems are however known to be NP-hard, which means that the computational resources required to solve them increase significantly with the problem’s size. For many real-world applications, finding a solution requires immense computational power and must be performed on large high-performance computer clusters that consume vast amounts of energy.
Our project PINCH (photonic Ising machines) aims to create and investigate a radically new photonic-based computation platform that can solve some of the most difficult problems in combinatorial optimization and machine learning. Our platform is a dedicated piece of hardware utilizes the fact these NP-hard problems can be directly mapped onto a simple physical system, called the Ising model. By implementing such Ising models with physical systems, so-called Ising machines, solutions can be found by the natural tendency of the Ising machine to evolve to the lowest energy state, which in turn represents the optimal solution. This powerful natural computing approach allows Ising machines to forego many of the fundamental limitations of traditional computing platforms when specifically tackling optimization problems and thus mitigate the scaling problems of digital hardware. However, a scalable ising machine does not yet exist due to the fundamental challenges, both in the theoretical understanding as well as in the physical realization of them.
The overall objective of the project is to develop, build and test photonic Ising machines, leading to a platform that we envision as an augmentation of existing computing systems. Photonics presents an ideal platform for Ising machines as the interaction of spins is naturally contained within the interference of different coherent light modes. Photonics can also achieve a high energy efficiency at an inherently high analog bandwidth, which is essential for efficient Ising machines. The parallelism of photonics enables coupling massive amounts of spins, making photonics the key technology to obtain large-scale spin systems. Finally, an additional advantage stems from the ability of photonics to be scaled to full quantum optical systems with the potential for improved performance.
In PINCH, three research groups from three Belgian universities will bundle their know-how and expertise. These groups are the Applied Physics research group from the Vrije Universiteit Brussel, the Neuromorphic Photonics Group from the Universiteit Gent and the Laboratoire d'Information Quantique from the Université Libre de Bruxelles.
Published papers on ising machines
- Noise-injected analog Ising machines enable ultrafast statistical sampling and machine learning
- Order-of-magnitude differences in computational performance of analog Ising machines induced by the choice of nonlinearity
- A poor man’s coherent Ising machine based on opto-electronic feedback systems for solving optimization problems
Research teams
VUB
-
Prof. Dr. Guy Verschaffelt
-
Prof. Dr. Guy Van Der Sande
-
Ian Bauwens
-
Toon Sevenants
-
Jacob Lamers
ULB
-
Prof. Dr Serge Massar
-
Dr. Nicolas Englebert
UGent
-
Prof. Dr. peter bienstman
-
Ruqi Shi