Q-PGM application
Q-PGM, currently under development, will enable you to go even further in understanding rare events. Q-PGM uses a quantum approach to the construction of probabilistic graphical models.
Designing the structure of a probabilistic graphical model is a complex problem. It involves exploring all possible solutions for the interconnections between the variables in the graph, in order to identify the combination that best represents the conditional independence relationships between the variables in the problem to be solved.
Due to the complexity of solving this problem, probabilistic graphical models are often neglected for the detection of rare events. This is due to the large number of variables that need to be taken into account to achieve satisfactory performance. Instead, other methods, although less explicable, and requiring dataset rebalancing techniques, are preferred as they offer the advantage of achieving detection performance satisfactory to industry in reasonable computation times.
Our Q-PGM application aims to overcome the complexity of calculating the structure of probabilistic graphical models, in order to make their use more accessible. This will enable future users of this module to take advantage of their explicability capability, which represents a significant added value in many rare event detection processes.
What are the benefits of using Q-PGM?
Thanks to an innovative reformulation, our Q-PGM application makes the most of the features of quantum hardware architectures to optimally solve the problem of finding the structure of a probabilistic graphical model.
A gain in explicability
By its very nature, our probabilistic graphical model approach has the advantage of being explicit. The visualization of conditional dependencies between variables makes it easier to understand the complex relationships between them.
Greater precision
Aware of the importance of balancing the number of events reported by the algorithm, for processing by experts, and the number of potential rare events that remain undetected, our approach is designed to take particular care in respecting this balance.
A more complete representation of the relationships between variables
Learning the structure of probabilistic graphical models becomes extremely complex to solve on conventional computers as the number of variables to be taken into account increases. To overcome this difficulty, it is necessary to reduce the number of variables taken into account or to limit the relationships between them when using classical approaches. Our initial results point to the possibility of increasing the number of connections between variables, while limiting the impact on model structure computation time, thanks to the use of quantum computing.
Examples of Q-PGM use cases
Insurance fraud detection
Computer security
Detecting intrusions
Predictive maintenance
Detect potential faults before they occur
Banking transaction monitoring
Detect fraudulent or unusual transactions