Optimization of large-scale systems arising in telecommunication and in artificial intelligence

该研究计划将侧重于开发和改进解决电信和人工智能中出现的大规模系统的优化方法。我们将对大规模组合系统的数学建模和有效求解工具的设计进行理论研究。我们还将考虑特定的应用，主要是在通信系统和人工智能领域，以便应用新的增强工具并在实际实例中验证其效率。在电信网络的应用方面，我们将重点开发用于单播和多播流量下的网络设计和管理的增强工具。事实上，随着电信垄断的逐渐消失和带宽成本的下降（几年来每年30%），电信行业已经遇到了深刻的变革，导致在核心和接入网中更合理地使用设备，因此需要优化工具来更有效地管理这些网络。据预测，在未来五年内，光网络的流量将每年翻一番。经过过去四年的投资，电信行业准备对通信基础设施进行再投资。具有新功能和更好性能的新一代设备正在出现。无论是对于住宅用户还是工业用户，它都可以通过更动态的物理层更好地集成语音、数据和视频。因此，开发高效的网络维化工具变得越来越重要。在人工智能的应用方面，第一个目标是将先前为命题概率逻辑开发的工具推广到可决实例的一阶概率逻辑。第二个目标是为自动化机制设计设计有效的解决工具，并将其应用于覆盖网络的自组织和自管理。

The research proposal will focus on the development and improvement of optimization methods for solving large scale systems arising in telecommunications and in artificial intelligence. We will conduct theoretical research on the mathematical modeling and the design of efficient solution tools for large scale combinatorial systems. We will also consider particular applications, mainly in the area of communication systems and of artificial intelligence in order to apply the new enhanced tools and validate their efficiency on realistic instances.On the application side of telecommunication networks, we will focus on the development of enhanced tools for network design and management under unicast and multicast traffic. Indeed, with the progressive disappearance of telecommunication monopols and the decreasing cost of the bandwidth (30% per year for several years), the telecommunication industry has encountered deep transformations that led to a much more rational use of the equipment both in the core and the access networks, hence the need of optimization tools for a more efficient management of those networks. Still for the next five years, it is forecast that the traffic in the optical networks will double every year. After a gel of the investments during the last four years, the telecommunication industry is ready to reinvest in the communication infrastructures. A new generation of equipment with new features and better performances is emerging. Both for residential and industrial users, it will allow a better integration of voice, data and video with a more dynamic physical layer. Therefore, the development of efficient tools for the dimensioning of networks becomes more and more important.On the application side in artificial intelligence, the first objective is to generalize the tols previously developed for propositional probabilistic logic to first order probabilistic logic for decidable instances. The second objective is to design efficient solution tools for automated mechanism design and apply them for the self-organization and self-management of overlay networks.