Canada Excellence Research Chair in Data Science for Real-time Decision Making

CERC的目标是开发利用“大数据”的新方法，并为当前(和未来)·决策时代实现优化战略。 这涉及到多线程计算、移动传感器、廉价数据存储和 实时通信和这些要素定义了全新的、令人兴奋的挑战，这些挑战强烈要求新的算法。 这些挑战将通过数学优化和机器学习之间的紧密合作来解决。该场景以数据为中心。这是因为： ·数据是动态的，也就是说，不再被视为简单地定义要求解的“实例”。该方法必须针对实时变化的数据量身定做，并且由数学优化开发的算法通常需要在交通网络、电网、供应链系统等的(轻微)变化的条件下进行重新优化。 ·数据是可用的，但很复杂，即信息量反映了传统上分开处理的问题之间的紧密结合。然后，数学“最优化”和机器学习必须能够提取“正确”类型和数量的信息，从而定义要解决的“正确”问题。 ·数据很大，即学习和优化方面的所有算法都必须规模化。这反映了涉及的数据量和决策过程的复杂性，这一过程前所未有地着眼于需要管理和优化的系统的更广泛的图景。 预计混合使用混合整数线性规划和非线性规划以及元启发式和约束编程技术将解决物流、供应链、收入管理、能源网络设计和运营等方面的应用。然而，往往受到应用的推动，主席的研究重点将放在这些技术的创新集成上，并开发利用机器学习和数学优化进行数据驱动决策的新范式。

The CERC objective is on developing new methodologies to exploit "Big Data" and achieve optimized strategies for the current (and future)· era of decision-making. That involves multi-thread computing, mobile sensors, cheap data storage and real-time communication and these ingredients define absolutely new and exciting challenges that strongly urge for new algorithms . Those challenges will be tackled by a tight cooperation between Mathematical Optimization and Machine Learning. The scenario is centered on data. This is because: • Data is dynamic, i.e., not anymore to be seen as simply defining the "instance" to be solved. The methodology must be tailored to data changing in real time, and often the algorithms to be developed by Mathematical Optimization need to re-optimize over (slightly) changed conditions of traffic networks, of power grids, of supply-chain systems, etc. • Data is available but complex, i.e., the amount of information reflects the tight integration between problems traditionally addressed separately. Then, Mathematical" Optimization together with Machine Learning must be able to extract the "right" type and quantity of information leading to the definition of the "right" problems to be solved. • Data is big, i.e., all algorithms both on the learning and optimization sides must go large scale. This reflects the amount of data involved and the complexity of a decision-making process that looks, as never done before, to a much broader picture of the system to be managed and optimized. It is expected that using a mixture of mixed-integer linear and nonlinear programming as well as metaheuristics and constraint programming techniques, will solve applications in logistics, supply chain, revenue management, design and operation of energy networks, etc. However, often motivated by applications, the chair research focus will be on the innovative integration of those techniques and on the development of new paradigms exploiting both Machine Learning and Mathematical Optimization for data-driven decision-making.