Cyber-Physical Systems Security: A Pacemaker Case study

Abstract: Cyber-Physical Systems (CPSs) use distributed embedded devices to control physical processes. In this talk I will explore emerging challenges related to CPS security. In particular, I will elaborate on challenges related to implantable medical devices such as pacemakers. It was reported recently by the US Food and Drug Administration (FDA) that hackers could either pace the devices rapidly inducing arrhythmia or could drain the battery. While methods exist for formal verification of pacemaker software, these are limited techniques to prevent security vulnerabilities. To this end we develop formal techniques based on both run-time verification and enforcement. Our approach proposes a wearable device that non-invasively senses the familiar ECG signals in order to determine if a pacemaker has been compromised. We develop a set of timed policies to be monitored at run-time. Towards the latter part of this talk, I will also discuss the potential for run-time enforcement in this setting.

Speaker Biography:

Partha S. Roop is a Professor in the Department of Electrical, Computer and Software Engineering at the University of Auckland. He is currently the Associate Dean (International) for the Faculty of Engineering. Partha's research interests are in Digital Health, Formal Methods for Safety-Critical applications of AI and Machine Learning, and Real-Time Systems. He is working with colleagues from the Medical School and the Auckland Bioengineering Institute (ABI) on new techniques developed by his group known as organ on a chip. He is also interested in heart rate variability and Biofeedback. Partha is an Associate Editor of IEEE Embedded Systems Letters and serves in the TPC of several top conferences including ACM International Conference on Embedded Software (EMSOFT) and ACM International Conference on Hybrid Systems: Computation and Control (HSCC).

Humans in Data Science – How to Engineer Features for Machine Learning

Abstract: Feature engineering is one of the most challenging aspects of a data science pipeline, and is considered to be an “arduous process for data scientists”, where the raw attributes often need to be transformed into derived attributes that can be more effective for building predictive models such as machine learning classifiers. The state-of-the-art fall into one of these two extremes: (a) fully manual, painstakingly slow and heavily reliant on domain expertise, often requiring data scientists to go through a repetitive trial-and-error exercise until the set of designed features are satisfactorily effective or (b) fully automated techniques (some notable systems are, Data Science Machine, ExploreKit, One Button Machine, or Featuretools4), which are usually not model agnostic, and require substantial time to engineer features that are opaque to the human analyst. In our work, we investigate a semi-automated “human-in-the-loop” framework for feature engineering that is developed around three fundamental aspects: (a) Model agnostic measure - our engineered features agnostic to the specifics of any underlying ML classification model; (b) Attributes algebra – we consider an algebra that dictates how raw are to be combined to produce the derived features; (c) Guiding human analysts – we investigation how to optimize the involvement of humans in the feature engineering process and ensure their efforts are successful. Rigorous evaluations are conducted over several real world datasets to demonstrate that our approach performs considerably better than fully automated systems (e.g., ExploreKit) as well as fully manual systems.

Speaker Biography:

Gautam Das is the Associate Dean for Research of the College of Engineering, a Distinguished University Chair Professor in the Computer Science and Engineering Department, and Director of the Database Exploration Laboratory (DBXLab) of the University of Texas at Arlington (UTA). Prior to UTA, Dr Das has held positions at Microsoft Research, Compaq Corporation and theUniversity of Memphis, as well as visiting positions at IBM Research and the Qatar Computing Research Institute. He graduated with a BTech in computer science from IIT Kanpur, India, and with a PhD in computer science from the University of Wisconsin-Madison. Dr. Das's research interests span data mining, databases, and machine learning, and have resulted in over 200+ papers, many of which have appeared in premier conferences and journals such as SIGMOD, VLDB, ICDE, KDD, TODS, and TKDE. He is a recipient of the IEEE ICDE “Test of Time” Influential Paper Award in 2012. Dr. Das has served in the Editorial Board of IEEE Transactions on Big Data, ACM TODS, and IEEE TKDE, has served as the General Chair of the flagship SIGMOD 2018 conference, as well as numerous other premier conferences. His research has been supported by grants from federal and state agencies such as NSF, ARO, ONR, DOE, THECB, as well as industry such as Microsoft, AT&T, Nokia, Cadence, and Apollo.

Data Science and Machine Learning: Recent Issues and Challenges

Speaker Biography:

Dr. Ujjwal Maulik is a full Professor in the Dept. of Comp. Sc. and Engg., Jadavpur University since 2004. He was also the former Head of the same Department. He held the position of the Principal in charge and the Head of the Dept. of Comp. Sc. and Engg., Kalayni Govt. Engg. College. Dr. Maulik worked in many universities and research laboratories around the world as visiting Professor/Scientist including Los Alamos National Lab., USA in 1997, Univ. of New South Wales, Australia in 1999, Univ. of Texas at Arlington, USA in 2001, Univ. of Maryland at Baltimore County, USA in 2004, Fraunhofer Institute for Autonome Intelligent Systems, St. Augustin, Germany in 2005, Tsinghua Univ., China in 2007, Sapienza Univ., Rome, Italy in 2008, Univ. of Heidelberg, Germany in 2009, German Cancer Research Center (DKFZ), Germany in 2010, 2011 and 2012, Grenoble INP, France in 2010, 2013 and 2016, University of Warsaw in 2013 and 2019, University of Padova, Italy in 2014 and 2016, Corvinus University, Budapest, 2015 and 2016, University of Slovenia in 2015 and 2017, International Center for Theoretical Physics (ICTP) in 2014, 2017 and 2018. He is the recipient of Alexander von Humboldt Fellowship during 2010, 2011 and 2012 and Senior Associate of ICTP, Italy during 2012-2018. He is the Fellow of International Association for Pattern Recognition (IAPR) and Indian Academy of Engineers (INAE). His research interest include Computational Intelligence and Machine Learning, Pattern Recognition and Data Mining, Bioinformatics, Multi-objective Optimization and Social Networking.

Data Driven Modelling: New Paradigms, Challenges & Opportunities

Abstract: The traditional paradigm of data driven modelling and system identification employs apriori information on system structures and environments to derive system models. This has beenan established research area with rich literature and substantial benchmark applications due to extensive research and development over the past half century on its methodologies, theoretical foundation, algorithms, verifications, and applications. However, rapid advancement in science, technology, engineering, and social medias has ushered a new era of systems science and control where challenges and opportunities are abundant for system identification. From this perspective, data driven modelling and system identification remains an exciting, young, viable, and critical field which warrants new paradigms to meet such challenges. This talk will highlight some important aspects of data driven modelling and system identification in these new paradigms, suggest some worthy areas of research focus, and most importantly will open the forum for further discussions.

Speaker Biography:

Akshya Kumar Swain graduated from Veer Surendra Sai University of Technology,Burla, India with a Gold Medal and received Masters of Electronic Systems and Communication from National Institute of Technology, Rourkela, India. He was a Commonwealth Scholar in the United Kingdom from 1994 to 1996 and received Ph.D. degree from the Department of Automatic Control and Systems Engineering, University of Sheffield, U.K. Currently he is working in Electrical, Computer and Software Engineering in University ofAuckland, New Zealand. He holds adjunct professor positions in IIT, Jammu and NIT, Meghalaya. His research interests include nonlinear system identification, fault tolerant control, machine learning, biomedical signal processing, and smart grid. He has authored/co-authored over 200 papers in different international journals and conferences. Dr. Swain is an Associate Editor of IEEE Sensors Journal and Member of the Editorial Board of International Journal of Automation and Control and International Journal of Sensors, Wireless Communications and Control.

Drones and Data for Disasters

Abstract: Drones and drone-based data products are about to be the next big thing. We are witnessing the Fourth Industrial Revolution and it's time to gear up to not just be a part of the experience, but to lead it. While mega Industries are all set to innovatively use drones for their use, let us also give a thought to utilising the technologies of the future to solve existing complex social problems that still challenge us. How do we bring technologies such as drones where they are needed the most? How do we empower the local youth to implement drone based solutions? How do we collaborate and crowd source efforts to solve large scale issues such as disasters? This talk will engage students in catching a glimpse of the drone-based social good projects globally.

Speaker Biography:

Dr Ruchi Saxena is a Global Health Systems Consultant with over 15 years of experience in working with start-ups, healthcare corporates as well as social and humanitarian non-profit organisations for auditing, training, consulting and implementing projects on healthcare ecosystem for quality and safety systems and protocols, disaster risk reduction and designing and implementing technologies such as IT systems and drones. She has worked on core process development, treatment and service quality, lean six sigma, patient safety protocols, hazards and risks assessments, disaster risk reduction, data analytics, aerial robotics, business intelligence and measuring and designing patient experience. She has her base projects in Kathmandu, Mumbai and New Delhi. She is dedicated to the cause of making healthcare excellent for all.

Deep Learning with Weakly Labelled Datasets using Adversarial Agents

Abstract: A prime challenge in building data driven inference models is the unavailability of statistically significant amount of labelled data. Datasets are typically designed for a specific purpose, and accordingly are weakly labelled for only a single class of tasks instead of being exhaustively annotated. Despite there being multiple datasets which cumulatively represent a large corpus, their weak labelling poses challenge for direct use in knowledge integration. In case of retinal images, specific datasets exist for development of data driven machine learning based algorithms for segmenting anatomical landmarks like vessels and optic disc as well as pathology like microaneurysms, hemorrhages, hard exudates and soft exudates. The aspiration is to learn to semantically segment all such classes using only a single fully convolutional neural network (FCN), while the challenge being that there is no single training dataset with all classes annotated. We solve this problem by training a single network using separate weakly labelled datasets. Essentially we use multi-task and adversarial learning approaches in addition to the classically employed objective of distortion loss minimization for semantic segmentation using FCN. This talk would focus on a general introduction to learning theory, the objectives of performance optimization in deep neural networks through learning, and the art of crafting new learning rules in view of solving such classes of critical problems. This talk would also introduce the "Turing Test Loss" which has been the driver in solving such optimization problems which require perception loss minimization rather than the classical distortion loss minimization.

Speaker Biography:

Debdoot Sheet is an Assistant Professor of Electrical Engineering and Artificial Intelligence at the Indian Institute of Technology Kharagpur and founder of SkinCurate Research, a Kharagpur, India based medical imaging AI company. He received the BTech degree in electronics and communication engineering in 2008 from the West Bengal University of Technology, Kolkata; MS and PhD degrees from the Indian Institute of Technology Kharagpur in 2010 and 2014 respectively. His current research interests include computational medical imaging, machine learning, deep learning, high-density tensor computation, fairness-accountability-trust-explainability (FATE) of artificial intelligence (AI). He is also a DAAD alumni and was a visiting scholar at the Technical University of Munich during 2011-12. He is also recipient of the IEEE Computer Society Richard E. Merwin Student Scholarship in 2012, the Fraunhofer Applications Award at the Indo-German Grand Science Slam in 2012, Won the GE Edison Challenge 2013, IEM Kolkata Distinguished Young Alumnus Award 2016. He is a senior member of IEEE, member of SPIE and ACM, life members of BMESI and IUPRAI, and serves as an Editor of IEEE Pulse since 2014.