Title
Learning the macroscopic traffic dynamics for adaptive optimal perimeter control
with integral reinforcement learning

Abstract
Conventional optimal perimeter control schemes require exact knowledge of the system dynamics and thus they would be fragile to endogenous uncertainties. Moreover, the prerequisite of accurate model calibration reduces their real-time applicability. To handle these challenges, this work proposes an integral reinforcement learning (IRL) based approach to learning the macroscopic traffic dynamics for adaptive optimal perimeter control. The proposed method relaxes the requirement on model calibration in a “model-free” manner that enables the robustness against the modeling uncertainty and enhances the real-time performance via a data-driven reinforcement learning (RL) algorithm. Approximate optimization methods are carried out to address the curse of dimensionality of the optimal control problem with consideration on the resolution of data measurement. A continuous-time control is developed with discrete gain updates to adapt to the discrete-time sensor data. The optimal control law is parameterized and then approximated by neural networks (NN), which moderate the computational complexity. Traditional RL methods usually converge slowly for lacking data efficiency, which is a major hurdle to real-time applications. To reduce the sampling complexity and use the available data more efficiently, the experience replay (ER) technique is introduced to the IRL algorithm. Different from the conventional RL approaches, the reinforcement interval can be varying with respect to the real-time resolution of data measurements, namely, the reinforcement interval can be selected online to ensure the richness of data. The convergence of the IRL based algorithms and the stability of the controlled traffic dynamics are proven. Both state and input constraints are considered while no model linearization is required. Finally, numerical examples are presented to verify the effectiveness of the proposed method.

Biodata
Professor Agachai Sumalee is currently a Professor at the Chulalongkorn School of Integrated Innovation with the specialization in Smart City. He received BEng in Civil Engineering with honor from King Mongkut’s Institute of Technology Ladkrabang, MSc(Eng) with Distinction in Transportation Planning and Engineering and also PhD from Institute for Transport Studies, University of Leeds. He has background in operation research and optimization and transportation engineering/economics. He has been involved in developing several successful technology solutions for smart mobility and smart city systems including the platform for the nationwide GPS management system, smart-taxi backend system, Highway Traffic Operational Control centre (HTOC), and several other systems for private sectors. In 2015 he founded Transcode company limited to spin-off his R&D technology to industry. Currently the company has been contributing to the deployment of smart city and smart mobility technology in Thailand with the current annual turnover of around twenty million USD. Prior to his current appointment at Chulalongkorn University Prof. Agachai is a full Professor at Department of Civil and Environmental Engineering, The Hong Kong Polytechnic University, visiting professor at Institute of Industrial Science University of Tokyo, Vice President at King Mongkut’s Institute of Technology Ladkrabang, and Senior Research Fellow at Institute for Transport Studies, University of Leeds. Prof. Sumalee has also published his research works on the topics of transport modelling, planning, and smart city extensively with over 100 journal publications. He is a Founding Editor and currently Editor in Chief of Transportmetrica B: Transport Dynamics which is an SCI-journal dedicated to the dynamical element in transport planning and management.

Title
Adversarial attacks on deep learning for computer vision

Abstract
Deep learning is at the heart of the current rise of artificial intelligence. However, deep models are vulnerable to adversarial attacks in the form of subtle perturbations to inputs leading to incorrect decisions, often with high confidence. In this talk, I will give a brief introduction to the methods for generating adversarial perturbations. I will discuss early defence mechanisms, including our work, for defence against such attacks. I will then discuss our method for generating the first ever attack on skeleton based human action recognition that also translates to the physical world. Following this, I will explain our Label Universal Targeted Attack (LUTA) that makes a deep model predict a specific target label for any sample of only a given source class with high probability. This is achieved by stochastically maximizing the log-probability of the target label for only the source class while suppressing leakage to the non-source classes. Finally, I will demonstrate the use of LUTA as a tool for deep model autopsy. LUTA results in interesting perturbation patterns revealing the inner working of the deep models and the training process itself exposes the feature embedding space.

Biodata
Ajmal Mian is a Professor of Computer Science at The University of Western Australia. He has received two prestigious fellowships and several research grants from the Australian Research Council, the National Health & Medical Research Council of Australia and US Department of Defence DARPA. He was the West Australian Early Career Scientist of the Year 2012 and has received several awards including the Excellence in Research Supervision Award, EH Thompson Award, ASPIRE Professional Development Award, Vice-chancellors Mid-career Research Award, Outstanding Young Investigator Award, and the Australasian Distinguished Doctoral Dissertation Award. He is an Associate Editor of IEEE Trans on Neural Networks & Learning Systems, IEEE Trans on Image Processing and the Pattern Recognition journal. He was the General Chair for the Int Conf on Digital Image Computing Techniques & Applications (DICTA) 2019, General Chair of the Asian Conference on Computer Vision 2018, Program Chair of DICTA 2012 and Area Chair of WACV 2019, WACV 2018, ICPR 2016, ACCV 2014. Ajmal Mian has supervised 16 PhD students to completion and has published over 200 scientific papers. His research interests are in computer vision, deep learning, 3D point cloud analysis, facial recognition, human action recognition and video analysis.

Title
Using Particle Swarm Optimization in testing

Abstract
In this talk, I will present recent work on the application of the Particle Swarm Optimization algorithm (PSO) to software testing. Initially, I will review the main concepts underlying the definition of a PSO. Our first line of work considers the use of PSO in search spaces whose elements are trees instead of, as usual, vectors. We apply this framework to generate test cases by evolving a population of trees. Our second line of work uses swarms in the scope of mutation testing. Specifically, given a set of mutants, we use a swarm to select hard-to-kill mutants.

Biodata
Manuel Núñez is a Professor in the Department of Computer Systems and Computation of the Complutense University of Madrid, Spain. He holds a Doctorate degree in Mathematics & Computer Science, obtained in 1996. Additionally, he holds a Master degree in Economics, obtained in 2002. Professor Núñez has done research in the broad field of formal methods. Currently, he is interested in testing complex systems using both formal and heuristic approaches.
Professor Núñez is a member of the IEEE SMC Technical Committee on Computational Collective Intelligence, the Board of Directors of the Tarot Summer School on Software Testing and the A-MOST, DISCOTEC and ICCCI Steering Committees. He is a member of several Editorial Boards of scientific journals. He has published more than 150 research papers in international journals and meetings.

Title
Dynamic Predictive Maintenance with Self-Adaptive Evolving Forecast Models

Abstract
Predictive maintenance relies on real-time monitoring and diagnosis of system components, and process and production chains. The primary strategy is to take action when items or parts show certain behaviors that usually result in machine failure, reduced performance or a downtrend in product quality.
In the first stage, it is thus of utmost importance to recognize potentially arising problems as early as possible. Therefore, a core component in predictive maintenance systems is the usage of techniques from the fields of forecasting and prognostics, which can either rely on process parameter settings (static case) or process values recorded over time (dynamic case). We will focus on the latter and demonstrate a robust learning procedure of time-series based forecast models, which can deal with very high-dimensional batch process modeling settings. Furthermore, our approach allows the forecast models to be on-line updated over time and on the fly whenever required due to intrinsic system dynamics (such as, e.g. varying product types, charges, settings, environmental influences) => leading to the paradigm of self-adaptive forecast models. This is achieved i) by recursive adaptation of model parameters to permanent changes and to increase model significance and accuracy and ii) by evolution of new model components (rules) on the fly in order to account for variations in the process, which require a change in the model’s ‘non-linearity degree’. We will also present some enhanced methods in model adaptation for increased flexibility to properly compensate system drift and shifts, such as dynamic forgetting, rule merging and splitting as well as an incremental update of the latent variable sub-space as a variant of incremental feature space transformation.
In the second stage, the evolved and incrementally adaptive forecast models can be used as surrogates in a fully automatized optimization procedure in order to prevent operator’s intervention and time-intensive manual reactions to predicted downtrends. Often, there are some “control wheels”, usually machine parameter settings which are able to change the behavior of the production process in order to meet the quality standards. Such settings may have indeed been optimized before (due to expert knowledge or in a static optimization process), but may not take into account dynamically changing factors during production. In other cases, such settings could not be optimized before at all (as requiring time-intensive design of experiments cycles), such that often a default parametrization is used which is suboptimal for the final product quality. We will define the optimization problem, which typically leads to a many-valued problem with very-high dimensional input parameter space; thus, we will demonstrate possibilities how a reduction to smaller problems can be achieved, and how the reduced problems can then be more quickly and robustly solved with multi-objective evolutionary algorithms.
The talk will be concluded with a real-world application scenario from a (micro-fluidic) chip production site, where the self-adaptive time-series based forecast models together with the process optimization component have been successfully applied in order to detect product quality downtrends at an earlier stage and to even suggest modified process values trends (and associated) machine parameter settings to improve and assure high-level product quality anytime.

Biodata
Edwin Lughofer received his PhD-degree from the Johannes Kepler University Linz (JKU) in 2005. He is currently Key Researcher with the Fuzzy Logic Laboratorium Linz / Department of Knowledge-Based Mathematical Systems (JKU) in the Softwarepark Hagenberg. He has participated in several basic and applied research projects on European and national level, with a specific focus on topics of Industry 4.0 and FoF (Factories of the Future). He has published more than 200 publications in the fields of evolving fuzzy systems, machine learning and vision, data stream mining, chemometrics, active learning, classification and clustering, fault detection and diagnosis, quality control and predictive maintenance, including 80 journals papers in SCI-expanded impact journals, a monograph on ’Evolving Fuzzy Systems’ (Springer, Heidelberg Berlin), an edited book on ’Learning in Non-stationary Environments’ (Springer, New York) and an edited book on ‘Predictive Maintenance in Dynamic Systems’ (Springer, New York). In sum, his publications received 6280 references achieving an h-index of 42. He is associate editor of the international journals Information Sciences, IEEE Transactions on Fuzzy Systems, Evolving Systems, Information Fusion, International Journal of Big Data and Analytics in Healthcare, Soft Computing and Complex and Intelligent Systems, the general chair of the IEEE Conference on EAIS 2014 in Linz, the publication chair of IEEE EAIS 2015, 2016, 2017, 2018 and 2020, the program co-chair of the International Conference on Machine Learning and Applications (ICMLA) 2018, the tutorial chair of IEEE SSCI Conference 2018, the publication chair of the 3rd INNS Conference on Big Data and Deep Learning 2018, and the Area chair of the FUZZ-IEEE 2015 conference in Istanbul. He co-organized around 12 special issues and more than 20 special sessions in international journals and conferences. In 2006 he received the best paper award at the International Symposium on Evolving Fuzzy Systems, in 2013 the best paper award at the IFAC conference in Manufacturing Modeling, Management and Control (800 participants) and in 2016 the best paper award at the IEEE Intelligent Systems Conference