Abstract : We now live in the Fourth Industrial Revolution characterized by cyber-physical systems and human-machine collaboration. Artificial Intelligence (especially machine learning algorithms) is ever increasingly used to augment and automate human decision-making. Many of such systems that operate in the real-world are typically black boxes with minimal explanatory capabilities. There are also concerns over fairness of decisions made by AI systems. This talk aims to elucidate the issue of AI fairness through a series of real-world examples. We will also look at some general methods for assessing and mitigating circumstances that may lead to unfair decisions.

Abstract : Ever-increasing qualities and quantities of clinical data are routinely collected concerning all aspects of patient care. Electronic health records offer a rich source of clinical information. Developments in wearable sensors, smart home technologies, and the Internet of Things provide the health industry with opportunities to monitor patients either in hospital or in home settings without significant disruption to their everyday activities. Innovations arising from machine learning approaches can facilitate rapid clinical intervention, transform a hospital-only treatment pathway into a cost-effective, home-based, combined alternative, and thus improve the overall quality of patient healthcare. However, analysing real-time collected data poses several challenges as the data can have substantial artefacts, might be highly imbalanced and incomplete, and might contain high variations. Moreover, data labelling can be expensive and time-consuming if there is an insufficient number of high-quality labels. This presentation will introduce several machine learning techniques to tackle these issues and to provide robust solutions to address such challenges.

Abstract :Within the literature, many specific conceptual learning mechanisms have been identified that play a role in organisational learning. Organisational learning emerges as an interplay of them. To be able to develop computational models for organisational learning, such conceptual mechanisms have to be formalised by computational mechanisms. Self-modeling networks provide powerful means to address the complexity of the interaction of different forms of learning, including the control over them. In this paper, recent developments are presented showing how this computational intelligence approach can be used to model the complex process of organisational learning and its underlying mechanisms covering both feed-forward learning to learn shared team or organisation mental models out of individually learned personal mental models and feedback learning to let individuals learn personal mental models from shared mental models. It is shown how by a self-modeling network, the different types of learning can be modeled by including a first-order self-model level and the control over them by including a second-order self-model level.

Abstract : Computing is being transformed to a model consisting of services that are delivered in a manner similar to utilities such as water, electricity, gas, and telephony. In such a model, users access services based on their requirements without regard to where the services are hosted or how they are delivered. Cloud computing paradigm has turned this vision of "computing utilities" into a reality. It offers infrastructure, platform, and software as services, which are made available to consumers as subscription-oriented services. Cloud application platforms need to offer (1) APIs and tools for rapid creation of elastic applications and (2) a runtime system for deployment of applications on geographically distributed computing infrastructure in a seamless manner.

Abstract : Modern industries are awash with a large amount of data. Extraction of information and knowledge discovery from data for process design, control and optimization, from day by day routine process operating data, is especially interesting but also challenging. Traditional automatic control theory has been developed mainly through model-based optimation techniques. The emerging data-driven learning technologies, especially reinforcement learning, being able to actively learn from control systems, have gained significant progress, attracting great interests from automation and control engineering communities. This presentation will introduce reinforcement learning and its applications in automation.

Abstract : With the rapid growth in the quantity and complexity of scientific knowledge available, the problems associated with dealing with this knowledge when applied to the design of modern intelligent systems are well-documented. Some of these problems are a result of the uncertainties and inconsistencies that arise in intelligent systems design and applications. Other problems arise from heterogeneous and informal formats for this knowledge. To address these problems, developments in the application of knowledge representation and reasoning technologies can allow scientific knowledge to be captured in logic-based formalisms. Using such formalisms, we can undertake reasoning under uncertainty and inconsistency to allow automated techniques to be used for querying and integrating scientific knowledge into system design. In this speech, we review some of the significant proposals for formalisms for representing and reasoning under uncertainty taking special attention into techniques that could be integrated in Data Science and Machine Learning Systems will be analyzed. The presentation will be focused on analyzing integration methodologies of interval analysis framework in intelligent data science and machine learning systems computations, which has been recently employed by the speaker in several algorithms.
This keynote lecture will, therefore, deal with all major tools for handling uncertainty in knowledge representation in the design and applications of data science and machine learning systems, but special attention will be given to interval analysis modelling as a new and very promising tool for efficiently managing uncertainty, emphasizing and analyzing its advantages and shortcomings.

Abstract : This talk will introduce the issue of "fair work" in the platform economy, along with an overview of the global Fairwork action research project. The first two emerging countries in which the Fairwork project was undertaken were India and South Africa (4 years ago); and the Fairwork project is currently officially active in 26 countries. Particular emphasis will be given here to the experiences of South Africa, with some notes on the Indian and global findings.

Abstract : Biometrics represents one of the most robust and reliable forms of human identification in physical and cyber security. The last decade has witnessed tremendous advances in sensor technologies and data processing techniques and algorithms. This has led to the strengthening of traditional biometrics technologies (e.g., fingerprint, face, iris, retina, keystroke dynamics, mouse gestures/dynamics and voice) and the emergence of several new technologies, which are showing great promises. The confluence of the consumer markets and national security needs have led to a growing demand for biometrics products and services. For instance, the integration of biometric sensors in smartphones and the use of these technologies for online banking have boosted the adoption of biometric technologies for the masses.

Abstract : The primary objective of this talk will focus on the topic, the Machine Learning that, of course, concerns the research community now a days. The philosophy behind machine learning is how we make the machine to learn from data (preferably numerical data) as the human learns from experience (that is through visual data). From this observation we can say that in order to design a good and/or perfect model(relation) for the proper pattern recognition, the right properties (attributes) of the object of concern should be selected and used. In this respect, we will address here the logic behind the development of various models that are useful in recognizing patterns lying in a given data set.

Abstract : Forward Selection Component Analysis (FSCA) is an unsupervised variable selection technique which has been developed as a counterpart to Principal Component Analysis (PCA) for variable selection problems. FSCA seeks to identify a small number of key variables that are representative of the observed variance across all variables in a dataset, thereby offering enhanced transparency and utility over PCA in many applications, from determining key wavelengths in Optical Emission Spectroscopy data analysis of plasma etch processes to optimising measurement site selection for spatial wafer metrology in semiconductor manufacturing.
However, determining the optimum subset of variables from a candidate set is inherently an NP hard problem, and therefore intractable even for modestly sized problems. In this talk I will introduce Forward Selection Component Analysis (FSCA) and how the basic algorithm can deliver near optimal performance using a GREEDY selection approach. Then, leveraging results from the field of combinatorial optimisation on submodular selection functions, I will describe how a LAZY implementation can be obtained, that can achieve similar variance explained performance to standard greedy implementation, while being up to an order of magnitude faster to compute.
In addition, I will briefly touch on recent extensions of the basic FSCA algorithm and some practical industrial application examples from the field of semiconductor manufacturing.

Abstract : Quickest detection, which refers to real-time detection of abrupt changes in the behavior of an observed signal or time series as quickly as possible after they occur, is essential to enable safety, security, and dependability of cyber-physical systems (CPS). Real-Time Machine Learning (RTML) has the potential to achieve quickest detection. However, Machine learning lacks the necessary mathematical framework to provide guarantees on correctness. The integration of machine learning with quickest detection not only creates new research opportunities with major societal implications, but also poses new research challenges in safety, security, and dependability. In this lecture, I will present a comprehensive survey of existing literature in the emerging area of real-time machine learning for quickest detection, identify the challenges, and evaluate the trends. I will also introduce our research findings in this area.

Abstract : Recent robotics, automation and sensing endeavors are making it into the mainstream. This talk will highlight several recent research and development projects with in the areas of designing hybrid controllers for robotics and automation applications, service robots, prototyping and synthesis of controllers, simulators, and monitors for manipulators and autonomous mobile platforms, algorithms for uncertainty computation from sense data, and web-based prototyping, control synthesis, and simulation of robotic agents. Several innovations and breakthroughs in Robotics, Intelligent Sensing and Control at Lawrence Technological University will be detailed. Research projects on developing and implementing theoretical and experimental tools to aid performing adaptive goal-directed robotic sensing for modeling, observing and controlling interactive agents and electromechanical systems in unstructured environments will be discussed.

Abstract : Graph analytics plays a critical role in the analysis of data from diverse sources such as the Internet, social networks, computational biology, scientific simulations, and finance. Graph-based analysis of data offers valuable insights and can lead to the discoveries of hidden patterns in massive datasets. Traditional algorithms do not work well for big data. Moreover, such huge data may not fit in the memory of a single processing unit, and thus, require distributed systems where the data is distributed among multiple compute nodes. High performance computing platforms such as MPI, Spark and Hadoop, are now essential parts of big data analytics as they provide frameworks for parallel and distributed computing with thousands of processing cores. In this talk, I will discuss some unique challenges in massive-scale graph analytics posed by big data and present some of my work on developing scalable distributed-memory parallel algorithms for generating random graphs, counting triangles, and subgraph analysis. These algorithms scale very well to a large number of cores (more than a thousand cores) and can work on graphs with billions of edges and vertices efficiently.

Abstract : AI is incorporated into recommendation services, driverless vehicles, surveillance and security, and many other types of systems. Yet people who interact with such systems often do not understand what the AI does or how it works. This lack of transparency can evoke confusion, frustration, and mistrust. And indeed, specific socially untoward consequences of algorithmic interactions have been widely identified. A popular way to think about evoking or enhancing trust between humans and AI is “explanatory AI”, the notion that AI should be able to explain itself, and thereby to help people understand what it is doing and how. This is similar to traditional conceptions of design rationale, but more dynamic and episodic; the rationale or explanation is generated in the course of operation and sharing it with human participants is incorporated into the basic interaction cycle. In this talk we contrast explanation and rationale with that of pragmatics. Explanation, understood logically, is both more and less than people want, need, and can make use of. An explanation that is not useful or usable by people in a situation cannot ameliorate the confusion, frustration and mistrust entrained by a lack of transparency in AI interactions. Rather than asking how AI can provide explanations, we need to ask how AI can be engaging and transparent to people. These are the original and abiding objectives of human-centered informatics. They apply to AI today, as they have always applied to information technology.

Abstract : Brain tumor segmentation is an important task to be performed in analyzing the brain Magnetic Resonance Images (MRI). These brain magnetic resonance (MR) image segmentation techniques help in making a biomedical image processing application for supporting neurologists and brain disease patients. There is need for automatic brain tumor segmentation methods to facilitate neurologists. This work implements a convolutional neural network based variational autoencoder model, modified with an additional flattening layer, on four different scenarios. In the first scenario, the model was trained on the BraTS-2019 dataset with a batch size of 1 over 10 epochs, and then tested on the BraTS-2018 dataset. In the second scenario, the model was trained on the BraTS-2018 dataset with a batch size of 1 over 10 epochs, and then tested on the BraTS-2019 dataset. In the third scenario, the model was trained on the BraTS-2019 dataset with a batch size of 2 over 20 epochs, and then tested on the BraTS-2018 dataset. In the fourth scenario, the model was trained on the BraTS-2018 dataset with a batch size of 2 over 20 epochs, and then tested on the BraTS-2019 dataset. The average Dice coefficient values of the whole tumor segmentation were 0.89003, 0.88551, 0.88569, and 0.9089, for the four scenarios, respectively. This proposed method’s accuracy on average was 0.893, which was shown to be a higher accuracy for whole tumor segmentation than other previous methods that have been compared in this research.

Abstract : By 2023 more than 30% of large companies will have invested, at least, in an Automated Decision System (ADS) and Artificial Intelligence will have a transformative impact in 2029. Problems of decision and optimization are of an importance beyond all doubt and hence to build ADS that act in daily life like people do, it is very important to reproduce as exact as possible the mechanisms that govern the decision processes that human beings follow. Information and framework in which a decision process is developed are key aspects that need to be known. In this talk, both concepts shall be introduced, described and analyzed. Accordingly the classical definition of General Decision Problem will be modified and, by means of a simple example, it will be shown how the solutions of decision and optimization problems can vary depending on the framework, thus showing that these ADS can work in a way even closer to how we humans do.

Abstract : Within the literature, many specific conceptual learning mechanisms have been identified that play a role in organisational learning. Organisational learning emerges as an interplay of them. To be able to develop computational models for organisational learning, such conceptual mechanisms have to be formalised by computational mechanisms. Self-modeling networks provide powerful means to address the complexity of the interaction of different forms of learning, including the control over them. In this paper, recent developments are presented showing how this computational intelligence approach can be used to model the complex process of organisational learning and its underlying mechanisms covering both feed-forward learning to learn shared team or organisation mental models out of individually learned personal mental models and feedback learning to let individuals learn personal mental models from shared mental models. It is shown how by a self-modeling network, the different types of learning can be modeled by including a first-order self-model level and the control over them by including a second-order self-model level.

Abstract : Adaptability and advanced services for ambient intelligence require an intelligent technological support for understanding the current needs and the desires of users in the interactions with the environment for their daily use, as well as for understanding the current status of the environment also in complex situations. This infrastructure constitutes an essential base for smart living. Various technologies are nowadays converging to support the creation of efficient and effective infrastructures for ambient intelligence. Artificial intelligence can provide flexible techniques for designing and implementing monitoring and control systems, which can be configured from behavioral examples or by mimicking approximate reasoning processes to achieve adaptable systems. Machine learning can be effective in extracting knowledge form data and learn the actual and desired behaviors and needs of individuals as well as the environment to support informed decisions in managing the environment itself and its adaptation to the people’s needs. Biometrics can help in identifying individuals or groups: their profiles can be used for adjusting the behavior of the environment. Machine learning can be exploited for dynamically learning the preferences and needs of individuals and enrich/update the profile associated either to such individual or to the group. Biometrics can also be used to create advanced human-computer interaction frameworks. Cloud computing environments will be instrumental in allowing for world-wide availability of knowledge about the preferences and needs of individuals as well as services for ambient intelligence to build applications easily. This talk will analyze the opportunities offered by these technologies to support the realization of adaptable operations and intelligent services for smart living in an ambient intelligent infrastructures.

Abstract :The global economy is transforming rapidly due to Covid-19 pandemic and growing use of digital technologies. The size of the digital economy in 2017 was worth approximately US$12.3 trillion or 15.5% of the global GDP and by 2025 it is expected to grow to 24.3% or US$23 trillion of the global GDP. The economy continues to grow driven by the ability to collect, use, analyze and monetize massive volume of machine-readable data. Rapid digitalization of both economics and societies are creating new means of tackling global development challenges. The evolving digital economy is driven by several frontier technologies, including big data & data analytics, artificial intelligence, internet of things, cloud computing, blockchain, automation & robotics, immersive technologies, 5G and IR4.0. The keynote will address the importance of research and innovation in frontier technologies and having the integrated research & innovation ecosystem to drive economic growth. The presentation will focus on research and innovation opportunities in intelligent technologies like AI, blockchain, 5G/6G and IR4.0/5.0. in driving more sustainable, human centric and resilient economic growth. This will be highlighted with number of use cases in different economic sectors such as agriculture, manufacturing, healthcare, etc.

Abstract : There has been many research development in fall detection systems. Machine learning approaches for fall detection are the most employed method to detect human falls. An assessment undertaken by the Australian Institute of Health and Welfare confirmed that nearly 2 in 5 of the Australians were living alone and the likelihood of doing so increases with age. Elderly falls is the second leading cause of death among eldely according to World Health Organization (WHO). It is reported that 39% of people aged 65-74, 40% of 75-84, and 51% of elders aged 85 and over live by themselves. Falls are considered common among the elderly especially with those aged 65 and over. The fall detection systems are very important systems as they can provide alarms for immediate medical attention when a fall is detected. Fall detection systems can provide relevant personnel with alarms to attend and reduce potential damage and life-threatening injuries after a human fall. Human falls has a great impact on the physical and mental health of the person who had a fall. It is reported that the psychological effects include depression, anxiety, fear of falling, and restricting activities of daily living (ADL). Among all effects of fall the fear of falling causes an estimated 60% of the elderly people to restrict their ADLs. This effect creates a negative cycle of not performing ADLs that causes isolation, impact gait imbalance and deteriorating muscles leading to anxiety depression and other falls in future. Fall detection systems are normally classified into wearable-based systems and context-aware systems. Approximately 50% of the falls observed in independent elderly people occurs out their home. Hence wearable fall detection systems are most useful as they are portable and easy to use of use in many locations. Using wearable sensors, it is possible to use Machine Learning systems by using sensor data to detect human falls. Issues related to data sets selection, data processing, and result of several machine learning fall detection algorithms and their accuracy of fall prediction will be discussed in this talk.

Abstract : Learning with knowledge distillation is a hot topic in today’s machine learning literature, with the motivation of equipping a student, much lighter-weight network with the knowledge from a teacher, typically much larger network. This research has a wide spectrum of real-world applications of developing powerful but light-weight and resource-limited front end devices. The idea of learning with knowledge distillation takes the metaphor of human learning as a student from a teacher. In this talk, I will address several interesting and important strategies and observations from the recent studies on the topic of learning with knowledge distillation, with the ultimate goal of substantially improving the learning effectiveness. I will discuss the applications of these new theories in scenarios with different modalities of data including imagery, bioinformatics data, and social network data.

Abstract : Music has been defined as "organized sound." Conventional music is highly organized, consisting of multiple hierarchical levels. In natural language, tokens (words) refer to external concepts, for example, the word "sound" means something outside of this sentence. But in music, we do not generally have anything like "words" and external reference. Therefore, in music, organization is achieved largely through repetition. When recognized, music repetitions serve as a kind of internal reference, so repetition itself is the basis for structure. Structure, in turn, plays a strong role in listener expectations and surprise. If structure and repetition are important in music, then any successful AI music generation system should be capable of creating musical repetition and perhaps learning hierarchical musical structures. A great hope, and even claim, of deep learning is that such abstract representations can be learned directly from data. In practice, we see little evidence that state-of-the-art sequence learners can form models of music structure or generate convincing musical repetition structures. This talk will explain some sources of this difficulty. Recent work has studied the relationship between repetition structure and the statistical distribution of rhythm, pitch and harmony. It is well-known that phrases often end with longer notes or tonic chords, for example. One must ask, however, are phrases defined by these features or do phrases give rise to these features? I will describe how phrases can be segmented purely on the basis of repetition and without regard to any other indicators of phrase boundaries. These phrases do in fact exert a significant influence on rhythm, pitch and harmony. Furthermore, multiple levels of hierarchy exert different influences on lower-level musical features, giving evidence that complex hierarchies are essential for music composition. I will describe some current work that uses these ideas of structure in music generation and music imitation. In evaluations with human listeners, we find that on average, successful pop songs are better than computer-generated imitations, but in random pairings, where neither song is familiar, our machine-generated songs are often rated equally good or even superior to human-composed songs. We also find that structure plays an important role in perceived song quality.

Abstract : Many real life computer science environments are characterized by the existence of intelligent multiagent systems where intelligent agents interact, formulate strategies, cooperate, coordinate, design systems, and plan actions autonomously for achieving their goals. Game theory analyses, formulates strategies and designs rules or mechanisms for these multiagent systems on the basis of artificial intelligence. For specifying, characterising and modelling and designing these intelligent multiagent systems, mathematical game theory models of different forms can be developed such as static, dynamic, evolutionary, differential and stochastic game theory models. Different algorithms such as Nash equilibrium, joint optimization, evolutionary algorithms, neural networks, genetic algorithms, and other machine learning algorithms can be applied to different game theory models for analysing, solving, and computing these models. Findings from these models are used to formulate strategies, cooperate, coordinate, design systems, and plan actions by different intelligent agents in different environments such as cyber security, internet trafficking, designing search engines, robot interactions, etc. Game theory applications in artificial intelligence and computer science is an important area for doing highly useful academic and practical activities and for academics and practitioners for building their careers. Therefore, it is necessary to give priority to this area in computer science for research and development.