Tutorials
Tutorial 1Hands-on Deep Learning for Industrial Informatics Applications Brief description: Deep learning is gradually becoming a mature artificial intelligence paradigm in both research and practice. Supported by a substantial evidence base, it demonstrates increasing potential for applications in the Industrial Informatics of automation, energy, manufacturing, transport, communication and human engagement. This workshop aims to develop essential knowledge of deep learning with hands-on tutorials in Python, using Google Collaboratory, Jupyter Notebooks and Visual Studio Code. The workshop will begin by exploring the structural elements of deep learning models, hyper-parameters, and comparison to standard machine learning algorithms, followed by the theory and application of deep neural networks (classification), convolutional neural networks (image processing), and recurrent neural networks (time-series prediction). Participants will conduct hands-on experiments of each technique using benchmark and real datasets, for training, testing and evaluation. Each technique will be demonstrated in the context of real-world projects in Industrial Informatics. The learning outcomes of this workshop are; the theoretical foundations of deep learning - when to use and in which settings, the design and development of deep learning models, rapid prototyping, evaluation and deployment using Python. Participants will access Google Collaboratory using a Gmail account, a laptop and a stable Internet connection will be essential. Presenters: Daswin De Silva, Rashmika Nawaratne, Achini Adikari Centre for Data Analytics and Cognition (CDAC) Daswin is Artificial Intelligence and Technology Platforms specialist in the Research Centre for Data Analytics and Cognition (CDAC) at La Trobe University, Australia. Daswin’s research interests are incremental machine learning, information fusion, deep learning, auto ML, with applications in energy, smart cities, and human emotions. He’s an associate editor of the IEEE Transactions of Industrial Informatics. |
Tutorial 2Machine Intelligence for Human-oriented Computing: From Deep Learning-based Video Processing to System Optimization Brief description: Recent years have witnessed unprecedented development of technologies in Artificial Intelligence (AI), Internet of Things (IoT), and cloud computing under the globally networking environment. Advanced video processing techniques play a central role in many real-world industrial systems, such as traffic surveillance, robot navigation, Unmanned Aerial Vehicles (UAV) patrols, automated electricity system inspection, etc. Efficient video sensing and computing on industrial applications can not only greatly reduce the consumption of labor and resources, but also guarantee the safety of operations in certain traditional industries. With high popularity of some video processing techniques such as object detection, localization, classification, clustering, scene understanding methods, many industrial engineering problems can now be solved effectively using data-driven machine learning approaches, especially the use of deep learning methods. But industrial systems have also become increasingly complex and there is an increasing interest and need from researchers and industrial engineers to solve more challenging and complex problems using more advanced machine learning concept. The proposed tutorial is aimed at introducing the most important machine learning techniques that are particular useful to industrial informatics engineers and researchers. The techniques that will be covered in the tutorial are the state-of–the-art and are still ongoing hot research topics. This tutorial is designed in a way to bridge the areas of industrial informatics and machine learning. Presenters:
|
Tutorial 3Energy-Efficient Train Operations and Its Applications: Boosting Energy Efficiency of Future Urban Electrified Transportation in Smart Cities Brief Description: This tutorial aims to offer a broad introduction on important methods for improving train operations efficiency using speed trajectory optimizations and other integrated energy management strategies. Based on Pontryagin’s Maximum Principle, in speed trajectory optimization problem can be solved by generating a series of optimal train operations, namely maximum acceleration, maximum braking, cruising and coasting. Different methods have been proposed to locate the series of operations to achieve the minimum traction energy as shown in Fig. 1. These methods include graphic-based search methods (Dynamic programming and A* search), optimal train control methods including both direct methods and indirect methods, mathematical programming and other heuristics and enhancement learning methods. This tutorial will provide a thorough introduction on these methods with an emphasis on the characteristics of each methods and the application of each methods on train energy efficient operations. Merging development of the study on train energy-efficient operation for emerging traction systems such as maglev train, fuel-cell train, train operation integrated with on-board energy storage, complex network applications for network-wide timetabling will be reviewed and discussed. This tutorial will touch on the topic of future development for electrified urban transportations including the fast-developing electric vehicles and urban railway systems. A summary and conclusions will be drawn towards the end of the tutorial. Presenters:
|