Call for Participants

We are excited to invite undergraduate, postgraduate, and PhD students across the UK to participate in a pivotal two-day workshop event "Cyber-Physical Systems (CPS): Cyber Security in Smart Grid, Electric Vehicles, and Autonomous Vehicles", supported by Google Research.  This event is dedicated to exploring the increasingly crucial intersection between the domains of Cyber-Physical Systems. The event is scheduled to take place at Cardiff University on the 10th and 11th of June 2024 at Cardiff University. The workshop aims to serve as a nexus for academic discourse, professional development, and collaborative exploration. Our commitment to fostering diversity and inclusivity is at the forefront of this initiative. We strongly encourage applications from students of the UK marginalized groups, including women and those with disabilities, underscoring our dedication to cultivating a varied and dynamic academic environment.

The workshop is designed to offer attendees a comprehensive understanding of the current and future landscape of cyber-physical systems and cyber-attacks impact on smart infrastructure. It will feature insightful presentations from leading industry experts, panel discussions, and a dedicated session for PhDs to present their research, along with providing a unique platform for scholarly exchange and network expansion. 

The workshop is divided into two parts: the 10th of June will be in person mode, and the 11th of June will be in online mode. Follow the below information to apply for either in-person event (PART-1) or online event (PART-2), or both (if you can).

To acknowledge each participant's contribution, the event will award the 'Certification of Participation' from Google. 

We invite you to join us in this advanced exploration of cyber security and innovation, where academic and industry boundaries converge, fostering a rich environment for learning and future collaborations.

PART-1 (10th of June 2024) - in person 

Expenses Support

Furthermore, to ensure an accessible and equitable experience for the first part of the workshop, the event offers to reimburse expenses for outstation participants up to £160 (for travel up to £100 and accommodation up to £60) to attend the workshop in Cardiff. 

International participants and those who cannot join the in-person event are highly encouraged to attend the online part on the 11th of June 2024.

Eligibility Criteria

Ø  Academic Level: Open to undergraduate, postgraduate (Master's), PhD, and postdoctoral students of all disciplines from universities across the UK.

Ø  Focus on Marginalised Groups: We strongly encourage applications from students who identify themselves as a part of the UK marginalised groups. This includes but is not limited to, ethnic minorities, women, individuals with physical and mental disabilities, and members of the LGBTQIA+ community. We aim to provide a platform where diverse voices and perspectives are heard, valued, and respected.

Ø  Preference for Undergraduate Students: While the workshop is open to students at all levels of higher education, there is a particular emphasis on including a higher number of undergraduate students. This is to facilitate a broader range of discussions about academic and professional opportunities available at the early stages of one’s academic journey. Undergraduate students from Cardiff University (CU) are especially encouraged to apply, although the call is open to undergraduates from all UK universities.

How to Apply

To apply for this workshop, interested candidates are required to apply with the following, which will enable us to understand your background and interests in the context of this event:

Ø  CV: Please provide a concise CV (including your name and email ID), limited to a maximum of two pages, detailing your academic and professional background, including any volunteer work on the topic related to this event.

Ø  One-page write-up highlighting your:

§  Interest and Expertise: Describe your interest, expertise, and any relevant experience in one or more of cyber security, energy systems, power/smart grids, transportation systems, and electric and autonomous Vehicles.

§  Inclusivity Aspect: Share your perspective on how belonging to a marginalized group will shape your academic and professional journey by attending this workshop. This should be a reflective and personal insight, focusing on your experiences and viewpoints. 

The application document should be a combined ONE PDF file, including a CV along with a one-page write-up. Kindly submit this PDF file to Dr Neteesh Saxena at saxenan4@cardiff.ac.uk with the subject line "Google workshop application" by the 14th of May 2024 (midnight). Having a different subject line will not be considered and your application will not be processed. Spaces are limited, so we encourage to you submit your application ASAP.

Selection Process

The selection process is committed to fairness and meritocracy, aiming to recruit selective high-potential students. Here is an overview of the selection criteria:

Ø  Merit-Based Selection: Candidates will be selected based on the potential and merit demonstrated in their submitted documents. This includes academic achievements, relevant experiences, and the expressed interest in the workshop.

Ø  Inclusivity and Diversity: In line with our commitment to inclusivity, preference will be given to female candidates and individuals with declared disabilities who meet the merit criteria.

Ø  Open to All: While we prioritise marginalised groups, the call for participation is open to all students, including male and non-disabled candidates, from across the UK.

Ø  The selection process is designed to ensure a diverse and balanced representation of participants, contributing to an enriching workshop environment.

Only the selected participants will be contacted through e-mail.

Registration

Registration to the part-1 event is invite-only based on the selection of the candidate. If you are selected, you will be contacted via email to register for the in-person event.

PART-2 (11th of June 2024) - online

International participants and UK individuals are highly encouraged to attend the online part on the 11th of June 2024. There is no selection process for this online event. There are still limited seats for participants to attend the event. However, you are required to register for the event (as mentioned below). An online link to join the event will be emailed later close to the event date.

Registration

To register yourself for the online event, click here

Benefits

Ø  Gain valuable insights from industry and academic leaders through focused discussions, expert talks, and hands-on knowledge.

Ø  Enjoy complimentary expense coverage through the workshop, including morning and afternoon tea/coffee and a lunch meal.

Ø  Network with distinguished scholars and industry experts in the relevant field.

Ø  Receive a Certificate of Participation endorsed by Google Research, recognising your involvement and learning.

Event Summary

Workshop Duration: 10th and 11th June 2024

Venue: Cardiff University, Abacws Building, Cathays, CF24 4AG

Important Dates

Application Submission Date: 14th of May 2024 21th of May 2024 (for part-1 in-person)

Selected Participant Notification: On a rolling basis (for part-1 in-person)

We reserve the right to close the acceptance of applications sooner than the deadline for part-1 in-person, if the seats are full well before time.

Contact Information

Organiser: Dr Neetesh Saxena (saxenan4@cardiff.ac.uk)

Follow us on Social Media to be updated: 

X (Twitter): @CYCISlab; 

https://www.linkedin.com/company/65714244/admin/feed/posts/

 https://www.facebook.com/CyCISlab/

We warmly invite you to seize this unique opportunity to expand your horizons in the world of Cyber-Physical Systems. Don't miss your chance to be part of this enlightening experience – register now to secure your spot! Thank you for considering this transformative workshop. We eagerly anticipate your participation and are excited to witness the diverse ideas and perspectives you will bring to this dynamic event.

Workshop Schedule:

10th June 2024 [In-person]

Venue: 

Abacws Building 

Cardiff University

Senghennydd Rd, Cardiff CF24 4AG

All participants will be provided with breakfast, lunch and evening tea/coffee - (Abacws/0.04)

Workshop Registration (with refreshment) - 0900 - 1000 hrs

Morning Session: 1000 - 1200 hrs (Abacws/0.01)

Welcome by Neetesh Saxena 1000 - 1010 hrs

Speaker: Nick Reed (Founder at Reed Mobility) - 1010 - 1040 hrs

Title: Automated Vehicles - quo vadis?

The Automated Vehicles bill is about to become law following extensive work by the GB Law Commissions to review the regulatory framework for self-driving vehicle operation. This push to legislate reflects a desire in the UK to be at the forefront of this technology but what are the implications of this legislation and what will it mean to have automated vehicles that society considers acceptably safe? In this talk, Nick will discuss the past, present and potential future of self-driving vehicles with reference to the Automated Vehicles bill and some of the incidents involving self-driving vehicles that have triggered concern about this technology.

Bio: Dr Nick Reed is the founder of Reed Mobility, an independent research consultancy on the future of mobility, working across public, private and academic sectors to deliver transport systems that are safe, clean, efficient, ethical and equitable. Nick is an internationally recognised expert on the safety and ethics of self-driving vehicle operation. In November 2021, he was appointed as the first ever Chief Road Safety Adviser to National Highways, providing review and challenge to the organisation in its aim to deliver Vision Zero on the national strategic road network.

Speaker: Emil C Lupu (Imperial College London) - 1050 - 1120 hrs

Title: Why CPS Security is Different? And What can We do About it?

The talk will examine the intrinsic characteristics of cyber-physical systems and, in particular, how the link between the human, cyber and physical aspects of the system impacts cyber-security. It will show how due to their very nature resilience of cyber-physical systems becomes an objective to achieve and why to cyber-attacks is difficult to evaluate. Finally, we will take a new look at the intersection of security and safety.

Bio: Emil Lupu is a Professor in the Department of Computing at Imperial College, Security Science Fellow of Imperial's Security Institute and interim co-director of the Research Institute in Trustworthy Inter-Connected Cyber-Physical Systems. At Imperial, he leads the Resilient Information Systems Security Group (RISS) (www.rissgroup.org). His research interests include cyber-physical systems resilience, the intersection of security and safety, the use of AI for cyber security and cyber security of AI (aka adversarial machine learning).

Speaker: Alex Tarter (AlixPartners) - 1130 - 1200 hrs

Title: The Business Challenge of Secure-by-Design

Even if there is agreement within a company that their products & solutions must be built ‘Secure-by-Design’, there are practical business-related challenges to accomplishing this around budgeting, requirements, process & methodology, and capability. These must be appropriately considered if the right outcome is to be achieved. In this talk Alex will draw on real-world experiences from both establishing & assuring product security practices within global engineering organisations operating within the rail, aviation, space and defence markets.

Bio: For more than 16 years, Alex has been engaged in the cybersecurity sector of the critical infrastructure, defense, and manufacturing areas both in the UK and the USA. He is currently a Senior Vice President at AlixPartners helping clients around the world build resiliency and cybersecurity into their products and critical systems. He previously served as the inaugural chief product security officer at a large UK systems-integration company working in the transportation, nuclear and defence sectors, and co-founded a cyber-startup providing Digital Risk Management services ultimately taking it through to acquisition. Alex has a PhD in the security of aviation systems from Lancaster University and was a Civil Expert on communications & cyber security to NATO.

Lunch - 1200-1330 hrs

Afternoon Session: 1330 - 1520 hrs

Speaker: Anurag Srivastava (West Virginia University) - 1330 - 1400 hrs

Title: Enabling Cyber-Resilient Smart Grid with Edge Devices

The integration of Electric Vehicles (EVs) and distributed energy resources (DERs) provides solutions for sustainable smart grid operation but also comes with multiple challenges. EVs and DERs connected as internet-of-things (IoT) based edge devices may help improve the resiliency of the critical loads during adverse events, but also bring additional cyber vulnerabilities. Minimizing the impact of cyber-attack on the grid depends on one of these key factors: a) planning for cyber defense, b) training of human operators and employees, and c) control and operational decisions driven by metrics. Formal metrics are needed to quantify the cyber resiliency of the smart grid and how to use that for operational cyber-resilience considering edge devices. This talk will cover the basics of cyber-power security and resiliency, tools for operational decisions, resiliency metrics for the power grid with IoT-based DERs/EV, and a testbed to validate cyber-resilience tools for the cyber-physical smart grid.

Bio: Anurag K. Srivastava is a Raymond J. Lane Professor and Chairperson of the Computer Science and Electrical Engineering Department at the West Virginia University (USA). His research interest includes data-driven algorithms for power system operation and control including cyber-resiliency analysis. Dr. Srivastava's high impact research projects resulted in tools installed at the utility control center supported for more than $60M by funding agencies. He is an IEEE Fellow, a member of several CIGRE WG and the author of more than 375 technical publications including 2 books and 3 patents.

Speaker: Dene  Yandle (Thales) - 1410-1440 hrs

Title: CPS in Industrial and Critical National Infrastructure Environments

This talk is all about how cyber security risks impact UK essential infrastructure and manufacturing.

Bio: Dene is a Systems Engineering Manager ICS/OT Cyber at Thales Ebbw Vale. He is an engineering professional specialising in Operational Technology cyber, automation and control systems design, commissioning and engineering management. Experienced in system design including PLC, HMI, electrical, pneumatic and refrigeration design. Extensive experience within a manufacturing industry covering design through to managerial positions.

Speaker: David Rogers (Copper Horse) - 1450- 1520 hrs

Title: TBA

Tea/Coffee Break 1520 - 1540 hrs

Evening Session: 1540 - 1650 hrs

Speaker: Ze Ji (Cardiff University) - 1540 - 1610 hrs

Title: Robot Learning for Autonomous Manipulation and Navigation

This talk explores the fusion of robotics and reinforcement learning, for autonomous robot manipulation and navigation. Specifics of employing reinforcement learning for manipulation tasks and navigation challenges are detailed. This paradigm enables robots to acquire complex skills, navigate intricate environments, and execute multifaceted manipulation tasks effectively. In particular, a hierarchical reinforcement learning framework is introduced for solving long-horizon multi-step manipulation tasks. This enables efficient parallel training of robot manipulation. In addition, the framework is also applied to robot navigation tasks that effectively improve the performance of local minimum. Affordance-based robot learning is also employed in our framework for enhanced efficiency; this has been demonstrated in tasks of computer games, and manipulating deformable and cluttered objects. This talk aims to ignite some discussions about the advancements in merging robot perception and learning to bolster higher-level robot autonomy. 

Bio: Dr Ze Ji is a Senior Lecturer with the School of Engineering, Cardiff University, and the recipient of the Royal Academy of Engineering Industrial Fellowship. Dr Ji is leading the research group of Robotics and Autonomous Intelligent Machines (RAIM). He worked in industry (Dyson, ASV, Lenovo) in the areas of robotics and autonomous systems. As PI and Co-I, he has secured funding from EPSRC, BBSRC, Royal Academy of Engineering, EU H2020, ERDF/WEFO, Royal Society, Industry, etc. His research is currently focused on robot vision, reinforcement learning, machine learning, simultaneous localisation and mapping (SLAM), tactile sensing, and their applications on autonomous robot navigation, manipulation and smart manufacturing. 

Speaker: Amir Javed (Cardiff University) - 1620 - 1650 hrs

Title: Adversarial Attacks on Intrusion Detection Systems in In-Vehicle Networks of Connected and Autonomous Vehicles

Rapid advancements in connected and autonomous vehicles (CAVs) are driven by breakthroughs in machine learning, yet these vehicles face significant risks from adversarial attacks. This paper examines the vulnerabilities of machine learning-based intrusion detection systems (IDSs) in in-vehicle networks (IVNs) to such attacks, shifting focus from the common research area of manipulating CAV perception models. Given the relatively simple nature of IVN data, we assess how susceptible IVN-based IDSs are to manipulation—a crucial inquiry since adversarial attacks often exploit complexity. We propose an adversarial attack strategy using a surrogate IDS trained with data from the onboard diagnostic port. By conducting these attacks under black-box conditions and adhering to realistic IVN traffic constraints, our method aims to deceive the IDS into misclassifying both normal-to-malicious and malicious-to-normal cases. Evaluations on two IDS models—a baseline IDS and a state-of-the-art model, MTH-IDS—revealed significant vulnerability, with F1 scores dropping from 95% to 38% and from 97% to 79%, respectively. Notably, inducing false alarms proved especially effective as an adversarial tactic, undermining user trust in the defence mechanism.

11th June 2024 [Online]

Welcome by Neetesh Saxena 1015 - 1025 hrs

Morning Session: 1025 - 1205 hrs

Speaker: David Choi (Soongsil University) - 1025 - 1055 hrs (1825-1855 hrs South Korea)

Title: Privacy Preserving Machine Learning for Energy Systems - Challengies and Opportunities

Recently, there has been an increase in distributed intelligence applications (e.g., 5G, Autonomous Driving, IoT, Crowdsourcing, Edge/Fog Computing, and Smart Grid) and an increase in the importance of privacy preservation of AI services and applications. Federated learning has emerged as a promising approach to accommodate these requirements. Clients in FL collaboratively train a model under the orchestration of a central server while keeping the training data decentralized. Compared to the traditional centralized learning architecture, federated learning can mitigate many systemic privacy risks by leaving the raw data distributed to the client. Also, the Smart grid has emerged as a key solution to provide clean and sustainable energy resources for the future. Smart grids have a distributed AI architecture that involves many different entities and private information. Therefore, there have been some efforts to use federated learning in smart grids. However, in addition to the inherent limitations of federated learning, there are unique challenges for the smart grid due to its unique features and requirements. This talk will present challenges in federated learning and federated optimization for distributed data protection and model optimization in smart grids. We believe that this convergence research is crucial to providing a secure and efficient AI environment needed for the continued growth of the smart grid for future generations.

Bio: Bong Jun Choi is an associate professor at the School of Computer Science & Engineering and jointly at the School of Electronic Engineering, Soongsil University, Seoul, Korea. Previously, he was an assistant professor at the Department of Computer Science, State University of New York Korea, Korea, and concurrently a research assistant professor at the Department of Computer Science, Stony Brook University, USA. He received his B.Sc. and M.Sc. degrees from Yonsei University, Korea, both in Electrical and Electronics Engineering, and his Ph.D. degree from the University of Waterloo, Canada, in Electrical and Computer Engineering. His current research focuses on distributed artificial intelligence, distributed intelligent energy networks, federated learning, and security. He is a senior member of IEEE and a member of ACM.

Speaker: Daisuke Mashima (Illinois ARCS and NUS) - 1100 - 1130 hrs (1800-1830 hrs Singapore)

Title: Smart Grid Cyber Range for Cybersecurity Research and Training

Smart grid systems, along with other modernized critical infrastructures, have been targeted by cyber attackers in recent years, which has surged the demand for research and development of cybersecurity technologies to defend them as well as for training of cybersecurity experts. Cyber range, which is a virtual environment imitating the configuration and behaviours of real system infrastructure, is a solution to provide a venue for such purposes. However, it is still a challenge for users to own/have access to a smart grid cyber range that offers system configuration and scale of interest. In this talk, we discuss an automated framework for generating smart grid cyber range to address this issue. We further introduce its cloud-based deployment for enhanced scalability and accessibility.

Bio: Dr. Daisuke MASHIMA is a Principal Research Scientist at Illinois Advanced Research Center at Singapore (Illinois ARCS) where he leads a cyber-physical systems security research team under multiple government-funded projects. He also serves as Adjunct Associate Professor at National University of Singapore. Previously, he was a member of the research staff at Fujitsu Laboratories of America, working on security and privacy in smart energy systems. He received his PhD degree in Computer Science from the College of Computing, Georgia Tech in 2012. His research interest spans over critical infrastructure cybersecurity, industrial control systems security, network security, smart grid, smart nation, and application of digital twins for cybersecurity. He was/is involved in the organizing committee of multiple international academic conferences, including IEEE SmartGridComm 2022, IEEE PES ISGT Asia 2022, ACM CPSS 2022, and ACM AsiaCCS 2024.

Speaker: Nouman Nafees (Cardiff University) - 1135 - 1205 hrs

Title: Validating Cyber Security Methodologies for Cyber-Physical Systems: A PowerWorld Simulation Approach 

In the evolving landscape of cyber-physical systems (CPS), ensuring robust security measures is paramount. Validating these security methodologies presents unique challenges due to the interplay between physical processes and cyber components. The complexity makes traditional validation techniques insufficient, necessitating advanced tools and approaches. In this talk, we focus on the critical process of validating cyber security methodologies for CPS, with a particular emphasis on smart grid applications. The importance of accurate validation lies in the potential consequences of cyber-attacks, which can disrupt physical processes and lead to significant operational and economic impacts. Therefore, robust validation is essential to ensure the resilience and reliability of these systems. Simulation tools like PowerWorld play a crucial role in this validation process. PowerWorld, an industrial and comprehensive power system simulation tool, allows us to model and simulate the physical aspects of CPS. This enables us to create realistic attack scenarios and test the effectiveness of various security measures in a controlled environment. Through this simulation, we can gain insights into the behaviour of the system under attack, identify vulnerabilities, and evaluate the performance of defence mechanisms. During the presentation, we will investigate modelling and simulating cyber-physical attacks within PowerWorld. By demonstrating how PowerWorld can accurately validate our methodologies, we aim to show its potential to enhance the security of interconnected infrastructures.

Bio: Nouman Nafees is a Lecturer in Cyber Security and Forensic Computing at Anglia Ruskin University, UK. He is currently completing his PhD in Cyber Security at Cardiff University, focusing on the security of cyber-physical systems, particularly within smart grids. Nouman has a robust background in developing cybersecurity solutions, including anomaly detection, moving target defense, and the application of deep learning models. 

Lunch – 1200 – 1330 hrs

Afternoon Session: 1330 - 1510 hrs

Speaker: Fei Teng (Imperial College London) - 1330 - 1400 hrs

Title: TBA

Speaker: Mohammed Asiri (Cardiff University) - 1405 - 1435 hrs

Title: Co-Simulation for Enhancing Cyber-Physical Security in Smart Grids

The increasing complexity of smart grids (SGs) necessitates advanced methods to ensure their security and resilience. Co-simulation, a technique that integrates both cyber and physical components of a system, offers a promising approach to address these challenges. In this talk, we will explore how co-simulation can be utilized to enhance the detection and investigation of Indicators of Compromise (IOCs) within SG environments. By combining communication network emulation with power system simulation, co-simulation provides a comprehensive framework for monitoring, analyzing, and responding to cyber threats. We will discuss the general principles of co-simulation, its relevance in supporting IOC identification, and its potential impact on improving the overall security posture of SGs.

Bio: Mohammed is a 4th-year Ph.D. student at Cardiff University, specializing in cyber-physical systems within the Cyber and Critical Infrastructure Security (CyCIS) Lab. His research focuses on Indicators of Compromise (IOCs) in cyber-physical systems, with a particular emphasis on power systems. His primary research goal is to understand the implications and challenges of implementing IOCs, identify potential IOCs for detecting targeted attacks, and develop capabilities to monitor these IOCs to enhance situational awareness. This involves a multidisciplinary approach to improve the resilience and security of critical infrastructure systems against cyber threats.

Speaker: Yazeed Alrowaili (Cardiff University) - 1440 - 1510 hrs

Title: Identifying Critical Assets in Smart Grids: A Network Theory Approach Integrating Power Measurements and Cyber Vulnerabilities

This talk focuses on identifying critical assets within smart grids by leveraging network theory and combining diverse data sources, such as power measurements and cyber vulnerabilities of Operational Technology (OT) and Information Technology (IT) assets. Smart grids, which integrate advanced digital communication technologies with traditional power grids, are essential for enhancing the efficiency, reliability, and sustainability of electricity distribution. However, their complexity and interconnectedness also introduce significant vulnerabilities. The study proposes a novel methodology to pinpoint key assets crucial for the stability and security of smart grids. By employing network theory, the research maps the interdependencies among various assets and evaluates their significance based on their roles within the grid. The approach incorporates both physical data, such as power flow measurements, and cyber data, including known vulnerabilities and potential attack vectors of OT and IT systems. Through a comprehensive analysis, the research aims to identify assets that, if compromised, could substantially impact the grid's performance and security. This identification process is vital for developing targeted strategies to protect and reinforce critical components of the smart grid infrastructure, ultimately contributing to a more resilient and secure energy system.

Bio: Yazeed is a PhD. student at Cardiff University, studying cyber-physical systems within the Cyber and Critical Infrastructure Security (CyCIS) Lab. His research focuses on the identification and risk analysis of critical assets in smart grids. His work uses network theory to analyze power measurements and cyber vulnerabilities of Operational Technology (OT) and Information Technology (IT) assets. By mapping interdependencies and assessing the importance of various assets, he aims to enhance the stability and security of smart grid infrastructure. His multidisciplinary approach seeks to improve the resilience and protection of critical infrastructure systems against cyber threats.

Tea/Coffee Break 1510 - 1530 hrs

Evening Session: 1530 - 1640 hrs

Speaker: Z. Berkay Celik (Purdue University) - 1530 - 1600 hrs (1030-1100 hrs Indiana USA)

Title: From Perception to Control: Compositional Security Analysis of Autonomous Systems

Autonomous systems, such as self-driving cars, drones, and mobile robots, are rapidly becoming ubiquitous in our society. These systems are composed of multiple individual software components for perception, prediction, planning, and control. While these systems are now blurring the lines between traditional computing systems and human intelligence and revolutionizing markets, a significant gap exists in developing theory and practice that indicates how the behavior of each component can be unified to reason about their system-wide security. This gap is exacerbated by the increasing use of learning-enabled components with inputs from diverse sensors and actuators that operate in open and uncontrolled physical environments. In this talk, I present the challenges in compositional secure autonomy and principles from our recent efforts on vulnerability discovery and security enforcement to address these challenges. I illustrate these challenges and principles with examples and sample results by focusing on autonomous driving. I conclude with a discussion of the open problems and opportunities, and outline areas for defensive research in the future.

Bio: Z. Berkay Celik is an Assistant Professor of Computer Science at Purdue University, where he is the co-director of the Purdue Security (PurSec) laboratory and a member of the Center for Education and Research in Information Assurance and Security (CERIAS). His research investigates the design and evaluation of security for software and systems, specifically on emerging computing platforms and the complex physical environments in which they operate. Through systems design, program analysis, and formal methods, his research seeks to improve security and privacy guarantees in commodity computer systems. His research approach is best illustrated by his extensive work on the Internet of Things (IoT) and Cyber-Physical Systems (CPS), including robotic vehicles, automobiles, and autonomous vehicles. He received the National Science Foundation CAREER Award in 2022 and Google's ASPIRE Research Award in 2021-2023. His research has been supported by grants from the National Science Foundation (NSF), Office of Naval Research (ONR), Defense Advanced Research Projects Agency (DARPA), Department of Transportation (USDOT), Department of Energy (DOE), Denso North America Foundation, Google, Apple, Cisco, Rolls Royce, and Sandia National Laboratories. More information about his research group and publication record is available at https://beerkay.github.io.

Speaker: Zheng Xu (Google Research) - 1605 - 1635 hrs (0805 - 0835 hrs San Frans. USA)

Title: Advances in Private Training for Production on-Device Language Models

We discuss how years of research advances now power the private training of Gboard LMs, since the proof-of-concept development of federated learning (FL) in 2017 and formal differential privacy (DP) guarantees in 2022. FL enables mobile phones to collaboratively learn a model while keeping all the training data on the device, and DP provides a quantifiable measure of data anonymization. Formally, DP is often characterized by (ε, δ) with smaller values representing stronger guarantees. Machine learning (ML) models are considered to have reasonable DP guarantees for ε=10 and strong DP guarantees for ε=1 when δ is small. As of today, all NWP neural network LMs in Gboard are trained with FL with formal DP guarantees, and all future launches of Gboard LMs trained on user data require DP. These 30+ Gboard on-device LMs are launched in 7+ languages and 15+ countries, and satisfy (ɛ, δ)-DP guarantees of small δ of 10-10 and ɛ between 0.994 and 13.69. To the best of our knowledge, this is the largest known deployment of user-level DP in production at Google or anywhere, and the first time a strong DP guarantee of ɛ < 1 is announced for models trained directly on user data. 

Bio: Zheng Xu is a research scientist working on federated learning and privacy at Google. He earned his Ph.D. in optimization and machine learning from the University of Maryland, College Park, in 2019. Before that, he got his master's and bachelor's degree from the University of Science and Technology of China. 

Speaker: Jingwei Ji (Waymo LLC) - 1640 - 1710 hrs (0840 - 0910 hrs San Frans. USA)

Title: TBA