2023 IEEE Symposium Series on Computational Intelligence December 5-8, 2023 | Mexico City, Mexico |
Tutorials and Workshops |
T1: Evolutionary Multi-objective Feature Selection for Machine Learning |
Abstract: We are now in the era of big data, where vast amounts of high-dimensional data become ubiquitous in a variety of domains, such as social media, healthcare, and cybersecurity. When machine learning algorithms are applied to such high-dimensional data, they suffer from the curse of dimensionality, where the data becomes very sparse. Furthermore, the high- dimensional data might contain redundant and/or irrelevant features that blur useful information from relevant features. Feature selection can address the above issues by selecting a small subset of relevant features which can improve the performance of machine learning methods, reduce the dimensionality of data, reduce space storage, improve computational efficiency, and facilitate data visualization and understanding. Feature selection plays a critical role in data mining, computational intelligence, and machine learning. Compared with other dimensionality reduction techniques, such as feature construction and feature extraction, feature selection can preserve the original semantics of the data, making it an effective method with interpretability and facilitating human understanding of the results. Feature selection is inherently a multi-objective problem. The two main goals of feature selection are to maximize the classification performance and minimize the number of selected features. However, these two objectives are usually in conflict. For example, removing relevant and/or complementary features can deteriorate classification performance. There is no single best feature subset, but rather a set of non-dominated subsets showing trade-offs between the two objectives. Optimizing the two objectives can more accurately reflect the decision-making reality of feature selection problems in practical applications. In this tutorial, the essential components in multi-objective feature selection such as solution representation, evaluation function (wrapper/filter/embedded), population initialization, offspring generation, environmental selection, and decision making will be discussed extensively, and the strength and weakness of each category of methods will be summarized. In addition, this tutorial will introduce the applications of multi-objective feature selection in various fields, such as image and signal processing, biological and biomedical tasks, business and financial problems, network/web service, and engineering problems, and illustrate the necessity of multi-objective feature selection for these fields. While state-of-the-art techniques have made significant progress in solving multi-objective feature selection, this tutorial will also identify and summarize the major issues and challenges when using evolutionary multi-objective optimization methods for multi-objective feature selection, and suggest some possible future research directions. |
Speaker(s):
Dr. Bing Xue is currently a Professor of artificial intelligence and the Deputy Head of the School of Engineering and Computer Science, VUW. She has more than 300 articles published in fully refereed international journals and conferences. Her research focuses mainly on evolutionary computation, machine learning, classification, symbolic regression, feature selection, evolving deep NNs, image analysis, transfer learning, and multi-objective machine learning. Dr. Xue is currently the Chair of the IEEE Computational Intelligence Society (CIS) Evolutionary Computation Technical Committee and IEEE CIS Task Force on Evolutionary Deep Learning and Applications, and an Editor of IEEE CIS Newsletter. She has also served as an Associate Editor for several international journals, such as IEEE Computational Intelligence Magazine, IEEE Transactions on Evolutionary Computation, and ACM Transactions on Evolutionary Learning and Optimization. She is a fellow of Engineering New Zealand. Dr. Mengjie Zhang is currently a Professor of computer science, the Head of the Evolutionary Computation Research Group, and the Associate Dean (Research and Innovation) of the Faculty of Engineering, Victoria University of Wellington, Wellington, New Zealand. He has published more than 700 research articles in refereed international journals and conferences. His current research interests include machine learning, evolutionary computation, genetic programming, image analysis, multiobjective decision-making, feature selection and reduction, scheduling and combinatorial optimization, and evolutionary deep learning and transfer learning. Prof. Zhang is a fellow of IEEE, a fellow of the Royal Society of New Zealand, a fellow of Engineering New Zealand, and an IEEE Distinguished Lecturer. He was the Chair of the IEEE Computational Intelligence Society (CIS) Intelligent Systems and Applications Technical Committee, the IEEE CIS Emergent Technologies Technical Committee, and the IEEE CIS Evolutionary Computation Technical Committee. He is currently the Chair of the IEEE CIS PubsCom Strategic Planning Committee and the IEEE CIS Outstanding Ph.D. Dissertation Award Committee, and the Founding Chair of the IEEE Computational Intelligence Chapter in New Zealand. |
|
T2: A Tutorial on Evolutionary Bilevel Optimization_ Concepts, Algorithms, and Applications |
Abstract: Bilevel optimization is a challenging problem that arises in various fields, requiring solving an optimization problem with a nested optimization task. This tutorial aims to provide a comprehensive overview of evolutionary bilevel optimization, focusing on concepts, algorithms, and applications. The tutorial begins by introducing important concepts on evolutionary optimization. Subsequently, the fundamental concepts and mathematical formulation of bilevel optimization is given. After that, it delves into the principles and methodologies of evolutionary bilevel algorithms, highlighting their suitability for tackling bilevel optimization problems but also their limitations. Various evolutionary bilevel algorithms are discussed, along with their adaptations and enhancements for bilevel optimization. Moreover, the tutorial explores diverse real-world applications of evolutionary bilevel optimization across domains such as engineering, economics, transportation, and machine learning. Case studies and practical examples illustrate the efficacy of evolutionary approaches in addressing complex bilevel decision-making scenarios. |
Speaker(s):
He is currently a full-time professor at the research and postgraduate division in the Instituto Tecnol ́ogico de Tlalnepantla (ITTLA) of the Tecnolo ́gico Nacional de M ́exico (TecNM). His research interests are the design and implementation of AI techniques and bio-inspired metaheuristics for optimization and their application to engineering problems. |
|
T3: Real-World Robot Evolution |
Abstract: The main goal of the tutorial is to outline the WHAT, the WHY and the HOW of real-world robot evolution. It will review the current state of the art and the main research directions to advance that in the short, mid and long term. As for the WHAT, it is about having a system of robots that can evolve, not inside a simulator, but in the physical realm. By the definition, this means selection, reproduction and heredity working in populations of real robots. The similarities and differences between natural and artificial evolution will be discussed, based on [1]. Based on [2], I will briefly summarise the key concepts of EC and describe the main components of Evolutionary Algorithms (EA). This part will end with elaborating on the differences between artificial evolution in software and artificial evolution in hardware. As for the WHY, I will discuss two principal motivations, one with engineering and one with science as the main angle. I will also explain that robot evolution can be used in two modi. First, as an off-line optimiser/designer that stops when a good solution is achieved. After this, many copies of a good solution can be produced and deployed. Second, in an on-line mode, similar to natural evolution of living organisms that never stops. This version is not about optimisation, but about adaptation, offering the ability to adjust the robots if the circumstances change. The main body of the tutorial is about the HOW with technical detailes and lots of examples, borrowing from “all” literature about the state of the art, including but not limited to [3]. Special attention is paid to the morphology-controller (body-brain) dichotomy, the role of individual (lifetime) learning after birth and the reality gap or sim2real gap. In this part I will distinguish:
[1] A.E. Eiben and J. Smith, From evolutionary computation to the evolution of things, Nature, 521:476-482, doi:10.1038/nature14544, 2015. |
Speaker(s):
|
|
T4: Adversarial Attacks, GAN, TrojAI, and Defense Mechanisms |
Abstract: Adversarial attacks have become a significant concern in the field of deep learning, posing threats to the security and reliability of AI systems. This comprehensive 2‐hour tutorial delves into the realms of adversarial attacks, Generative Adversarial Networks (GANs), Trojan AI (TrojAI), and defense mechanisms. The tutorial begins with an introduction to adversarial attacks, highlighting the vulnerabilities of deep learning models and their real‐world implications. Various types of adversarial attacks, including gradient‐based attacks, generative attacks, and poisoning attacks, are explored through case studies and examples. Next, the tutorial delves into the world of GANs, elucidating their architecture, training process, and their applications in generating adversarial examples. The vulnerabilities and weaknesses of GANs are examined to provide a holistic understanding of their security implications. Trojan AI, another emerging threat, is then examined in detail. The tutorial uncovers the methodologies and techniques behind Trojan attacks on AI systems, highlighting the challenges in detection and mitigation, and the potential risks associated with these attacks. The tutorial also emphasizes defense mechanisms against adversarial attacks, GANs, and TrojAI. Adversarial training, robust optimization techniques, and specific defense strategies are explored. Detection and mitigation methods for GAN‐based attacks, as well as Trojan detection and prevention mechanisms, are discussed in‐depth. We will also discuss Generative AI advancements like GPTs, Bard, diffusion models, and LLMs offer great potential but also pose adversarial risks. Advanced defense mechanisms and ongoing research are also addressed. The tutorial concludes with a focus on evaluating and assessing model security, including metrics for evaluating model robustness, security considerations in model deployment, and future directions in this rapidly evolving field. Through hands‐on demonstrations, interactive sessions, and participant engagement, this tutorial equips attendees with a comprehensive understanding of adversarial attacks, GANs, TrojAI, and defense mechanisms. Participants will gain valuable insights into the challenges, techniques, and emerging trends in securing deep learning models against adversarial threats. We will also present details of the federated learning and its applications in several domains and the recent advances in the area of understanding threats in the federated learning environments and possible remedy available. |
Speaker(s):
Dr. Dasgupta has more than 300 publications with 19000+ citations and having h‐index of 62 as per Google scholar. He received four Best Paper Awards at international conferences (1996, 2006, 2009, and 2012) and two Best Runner‐Up Paper Awards (2013 and 2014). Among other awards, he is the recipient of 2012 Willard R. Sparks Eminent Faculty Award, the highest distinction and most prestigious honor given to a faculty member by the University of Memphis. Prof. Dasgupta received the 2014 ACM SIGEVO Impact Award, an ACM Distinguished Speaker from 2015‐2020, and currently IEEE Distinguished Lecturer until 2024. Dr. Arunava Roy is currently a Research Assistant Professor of Computer Science at The University of Memphis. His area of interests includes Machine Learning, Security, Statistical Modeling, and Reliability. Dr. Roy obtained his Ph.D. from the Dept. of Applied Mathematics, IIT ISM Dhanbad, India in 2015. Dr. Roy then continued his research in the development of the Machine Learning and Statistical methods for mitigating cyber‐attacks, Big Data and Artificial Intelligence in the Dept. of Computer Science at The University of Memphis, USA as a Post‐ Doctoral Research fellow until 2016. Then joined the National University of Singapore (NUS) as a Research Fellow in the Dept. of Industrial and Systems Engineering in 2016 and later, he joined the Singapore University of Technology and Design (SUTD) in 2017 as a Post‐Doctoral Fellow in Computational Statistics for mitigating cyber issues in the CorpLab. Dr. Roy worked as a Research Assistant Professor of Computer Science in the Dept. of Computer Science & Engineering at the SRM Institute of Science and Technology (SRM IST), Chennai, India in 2017, where Artificial Intelligence, Machine Learning, and Cyber Security were his subjects of interests. He then worked in the Research Faculty of Computer and Information Science (CIS) at the Universiti Technologi Petronas (UTP) Malaysia and the School of IT at Monash University between 2019‐2021. Currently, he is a Research Assistant Prof. of Computer Science in the University of Memphis, TN, USA. Dr. Roy has several publications in various Q1 journals including IEEE, Elsevier, Springer, Taylor & Francis, and Wiley. He also co‐authored a book entitled “Advances in User Authentication” published by Springer USA in 2017. He has four US Patents, one of which is recently licensed by a Silicon Valley security startup called i2chain. He also filed another US patent on 2020 jointly with The University of Memphis, TN, USA. Presently, he is authoring a book entitled “Emerging Trends Techniques in Reliability Engineering & Security”, which will be published by Springer‐Nature, Switzerland. Dr Kishor Datta Gupta is an assistant professor of computer and information Science at Clark Atlanta University, GA. He completed his Ph.D. in computer science from the University of Memphis in 2021. He is presently researching Autonomous system vulnerabilities. His research interest includes evolutionary computation, Adversarial machine learning, algorithm bias. He is co‐inventor of adversarial defense system patent. |
|
T5: Computational Intelligence: Applications in system identification, control, and optimization |
Abstract: The objective of this tutorial is to expose researchers from the academia and industry to field of computational intelligence (CI) and learning methods and their applications for solving numerous engineering problems including system identification/modeling, nonlinear control, and optimization in uncertain and dynamic environments. This four-hour tutorial will focus on the following major topics, starting with introduction to the field of computational intelligence, the different CI paradigms, and their applications in system identification, control and optimization. Topics to be covered will include:
This tutorial focuses on the dominant paradigms of CI. Concepts, models, algorithms and tools for development of fuzzy logic, artificial neural networks, evolutionary computing, swarm intelligence and artificial immune systems and their applications will be covered. Implementations of these algorithms will be demonstrated, and a comparative performance analysis will be carried out. Some reported applications of these algorithms will be discussed in detail with an emphasis on their pros and cons. Theoretical background, mathematical formalism, implementation considerations, case studies on applications of each of these paradigms will be provided. Nonlinear modeling and control have been proposed using intelligent techniques such as neural networks, fuzzy, reinforcement learning and many others using inverse models, direct/indirect adaptive, or cloning a linear controller. There are merits for each approach adopted. There is a wide gap between applications of these methods in real time and in simulation. Issues such as stability, processor speeds, learning time, types of training algorithms etc. arise when it comes to real-time implementations. Static and dynamic optimizations using CI methods will be covered with emphasis on evolutionary computation and swarm intelligence. Adaptive Critic designs are neural networks capable of optimization over time under conditions of noise and uncertainty. The optimization technique is based on a combination of the concept of reinforcement learning and approximate dynamic programming. The Adaptive Critic method determines an optimal control law for a system by successively adapting two neural networks, an Action network (which dispenses the control signals) and a Critic network (which ‘learns’ the desired performance index for some function associated with the performance index). The primary aim of this tutorial is to provide engineers/researchers from industry/academia, new to the field of computational intelligence and learning methods with the fundamentals required to benefit from and contribute to the rapidly growing field of intelligent systems applications in uncertain and dynamic environments. |
Speaker(s):
Dr. Venayagamoorthy’s research interests are in the development and innovation of smart grid and artificial intelligence technologies. Dr. Venayagamoorthy is an inventor of technologies for scalable computational intelligence for complex systems and dynamic stochastic optimal power flow. He has published over 550 refereed technical articles which are cited over 23,000 times with a h-index of 70 and i10-index of > 300. Dr. Venayagamoorthy has given over 500 invited technical presentations including keynotes and plenaries in over 40 countries to date. Dr. Venayagamoorthy is the Chair and Founder of the IEEE PES Working Group on Intelligent Control Systems and IEEE Computational Intelligence Society (CIS) Task Force on Smart Grid. He has served/serves as Editor/Associate Editor/Guest Editor of several IEEE Transactions and Elsevier Journals. He is the Editor for the IEEE Press Series on Power and Energy Systems. Dr. Venayagamoorthy is a Fellow of the IEEE, IET (UK), the South African Institute of Electrical Engineers (SAIEE) and Asia-Pacific Artificial Intelligence Association (AAIA), and a Senior Member of the International Neural Network Society (INNS). He is an IEEE CIS and IES Distinguished Lecturer and a Member of the Board of Governors and Vice-President for Industry Relations of the INNS. |
|
W1: Trustworthy AI through Model risk Management |
Abstract: The workshop will focus on model risk management (MRM) for trustworthy AI which is an open and emerging area of research in data science, mathematics, and statistics. In particular, development of AI/ML models without understanding the underlying risk and uncertainty, particularly where pathological bias exists, can be detrimental to our society. As more and more complex and critical systems decision making relies on ML for applications ranging from financial to biological to defense. It is crucial to develop rigorous scientific techniques for decision making under risk and uncertainty using ML. The workshop will introduce the new center established at UNC Charlotte called TAIM^2 and invite speakers to provide overview of the current state as well as help identify future directions of the emerging area of identification and management of risks when adopting AI. |
Speaker(s):
Dr. Minwoo Lee is the Co-PI of the TAIM2 proposal and Assistan Professor in the Department of Computer Science and School of Data Science Dr. Gunay Dogan is a staff member at NIST working on mathematical software. He works on model validation approaches and recently is interested in AI relevant software. Dr. Michael Pokojovy is an Assistant Professor at the University of Texas at El Paso. He works on data science and statistical machine learning theory. His research is in the intersection of data science, applied mathematics and he also worked in the industry in the area of model risk management. Dr. Robert Thomas is Chief Scientist at the Air Force Research Lab in San Antonio Dr. Robert Thomas, lead for the Bioeffects core technical competency, Bioeffects Division, Airman Systems Directorate, 711th Human Performance Wing, Joint Base San Antonio, Texas. |
|
W2: Computational Intelligence Techniques for Solving Equity, Diversity, and Inclusivity Problem |
Abstract: In recent years, government organizations, universities, granting agencies, and industries have been committed to addressing equity, diversity and inclusivity (EDI) in their policies and strategies. Although this is the first step in promoting awareness of EDI issues, the research in this field is still limited. The focus of this workshop, therefore, is to explore research questions in EDI and their solutions through computational intelligence techniques, namely, evolutionary computation, neural networks, fuzzy logic and probabilistic modelling. The workshop aims to create a forum to discuss: (i) how to design algorithmic and data-driven approaches to detect EDI parameters such as bias and fairness in models and data using computational intelligence; (ii) how to develop computational intelligence system tools to study EDI parameters; (iii) what metrics and evaluation criteria are required to measure and assess the computational intelligence system; (iv) how to use social media dialogues and large language models to identify the general well-being of equity deserving groups, amongst others. Topics of Interest with respect to computational intelligence research in EDI Meta-heuristic Techniques Probabilistic Models Evolutionary Computation Fuzzy Logic Techniques Agent Based Techniques Transfer Learning Automated Design
Neural Networks and Deep Learning Explainable AI
Multi objective optimization |
Organizers:
|
|
W3: Quantum Machine Learning |
Abstract: The objective of this workshop is to present the fundamental ideas, concepts and elements used in quantum machine learning through several examples executed in the Qiskit environment based on Python. Among those examples, the participants will study the quantum implementation of supervised machine learning algorithms oriented for regression and classification. Quantum computing and quantum information theory are currently two research areas of great interest in the international academic community of electronic engineering, computer science, telecommunications and other related fields. On the other hand, quantum machine learning (QML) explores the interplay of ideas from quantum computing and machine learning, such that QML extends the set of hardware available for machine learning, through a new type of computing device based on quantum mechanics. Given the above, this workshop offers a basic understanding of QML, first introducing quantum phenomenology applied to computing, through concepts such as the “qubit”, superposition and quantum entanglement. Subsequently, the Parameterized Quantum Circuit (PQC) will be studied in order to establish the necessary bases to implement various models of Quantum Neural Networks defined in the IBM-Qiskit environment for classification and regression tasks. |
Speaker(s):
|