SIMTech-NTU Joint Lab on Complex Systems
Block N4, B1a-01 | School of Computer Engineering, Nanyang Technological University | Singapore 639798


What is exciting about this research? Milestones/progress?
28 January 2015 - 1:45pm

Research Direction / Sustainable Plan

Complex systems modeling, simulation, data analytics and machine learning and optimization sciences are becoming indispensible tools for identifying customer demand patterns, forecasting future demand, predicting auction closing prices, approximating and optimizing supply availability, performing what-if analysis and handling risks due to uncertainties and disruptions. Such technologies and tools build can significantly alleviate or bring new insights into the complexities of today's rapidly changing supply chain dynamics and globalized economy. In particular, to cope with the rapid advances in e-Commerce, today's supply chains need to be reconfigured swiftly to re-shape the companies' business models, deliver more orders and cater to the new customer preferences instantaneous, so as to support the shift towards more services-oriented activities in manufacturing, so as to extend Singapore's leadership as master facilitative control hubs of supply chain in the Asia-Pacific.

In contrast to using a top-down hierarchical thinking, the joint lab considers a radical shift of paradigm in the system and software engineering of decision-support systems by taking into account of individuals belong to different formal and informal social systems, and makes intelligent integration of knowledge mined from large amount of data that could be churned out in seconds, together with modeling and optimization tools, to manage complexities effectively. The manufacturing logistics and supply chain operations sector are the focus with research directions summarized as follows.

Computational Intelligence for developing competencies to cope with Complexity Management for Future of Manufacturing and Supply Chain Network Management

One focus is on exploring a data-driven computational modeling and simulation process that is known as rapid memetic imitation. The rapid memetic imitation process is a memetic evolution and learning process that is consistent with the world of intentional, conscious and responsible free agents, commonly modeled in a supply chain network. This self-organizing process allows supply chain tools built with it to perform independent experiments to test hypotheses and interpret findings without human guidance. It essentially transforms the informational supply chain representational contents (such as demand data, policy and events) which has a low level of complexity), into high-level co-adaptive meme complexes, which are usually highly unpredicted and unanticipated complex system knowledge and decision making (such as innovative policy to reduce bullwhip effects, etc).

Meta-Learning, Multi-Co-objective Optimization and Domain Adaptation for Mathematically Intractable Factors

It is expected that there will be more and more cases where mathematically intractable factors such as flexibility, robustness and resiliency of supply chain are critical make-or-break decisions aside cost reduction and profit maximization. We will be looking into ways of extracting, modeling and transferring the amounts of knowledge associated with supply chain risks, such as cause-effect relations, effects of risk events, and strategies of handling risk events. The search for transferrable structural patterns can happen at different levels of abstraction as the pattern itself may be organized hierarchically and composed of clusters of patterns. With the increased usage of the knowledge, the data mining tools traverse the hierarchies and produce more useful meta-knowledge for effective and efficient optimization.

Computational Intelligence, Data Analytics & Machine Learning for Supply Chain Visibility and Smarter Sense

It should be emphasized that supply chain software are tightly linked with operational modules including supply chain planning, warehouse management, distribution planning and even accounting systems in two aspects of consistency; data consistency and decision consistency. Keeping high data quality is inevitable for any decision support tool to generate useful plan. Especially for supply chain design, correctness of fundamental input parameters such as lead-times, capacity levels and yield rates directly connects with reliability of the output plan. On the other hand, the high dimensionality and the non-determinism of the dynamic supply chain environment, as well as the combinatorial nature of the problem, makes this task extremely difficult to be realized in real-world supply chain.

With the elevating amount of daily transactions in the supply chain, large-scale data analytics and machine learning techniques for high dimensional data are vital to forecasting future demand, predicting capacity levels, and approximating supply availability. Data-centric supply chain analytics become an indispensable tool to significantly alleviate the extremely high complexities in modeling the dynamic supply chain environment, assist analyzing the customer demand patterns, and also accelerate the overall optimization process. To this end, we shall also perform investigations on efficient large-scale machine learning technologies for data-centric supply chain and analytics. Specifically, we will explore data analytics techniques, such as dependence estimation, anomaly detection and trend detection, to identify useful patterns for customer behaviors and discover the trend of capacity levels from daily transactions.