Panel CODASPY '21, April 26–28, 2021, Virtual Event, USA

AI for Security and Security for AI

Elisa Bertino Murat Kantarcioglu Cuneyt Gurcan Akcora
[email protected] [email protected] [email protected]
Purdue University The University of Texas at Dallas University of Manitoba, Canada

Sagar Samtani Sudip Mittal Maanak Gupta

[email protected] (Moderator) (Moderator)
Indiana University [email protected] [email protected]
University of North Carolina Tennessee Technological University
Wilmington

ABSTRACT With respect to ethics, we need to develop technical approaches to

On one side, the security industry has successfully adopted some embed ethics principles in intelligent systems.
AI-based techniques. Use varies from mitigating denial of service Elisa Bertino is the Samuel D. Conte Professor of Computer
attacks, forensics, intrusion detection systems, homeland security, Science at Purdue University. She serves as Director of the Purdue
critical infrastructures protection, sensitive information leakage, Cyberspace Security Lab (Cyber2Slab). In her role as Director of Cy-
access control, and malware detection. On the other side, we see the ber2SLab she leads multi-disciplinary research in data security and
rise of Adversarial AI. Here the core idea is to subvert AI systems privacy. Prior to joining Purdue, she was a professor and department
for fun and profit. The methods utilized for the production of AI head at the Department of Computer Science and Communication
systems are systematically vulnerable to a new class of vulnera- of the University of Milan. She has been a visiting researcher at the
bilities. Adversaries are exploiting these vulnerabilities to alter AI IBM Research Laboratory (now Almaden) in San Jose, at the Mi-
system behavior to serve a malicious end goal. This panel discusses croelectronics and Computer Technology Corporation, at Telcordia
some of these aspects. Technologies, and visiting professor at the Singapore Management
University and the National University of Singapore. Her recent
ACM Reference Format: research focuses on cybersecurity and privacy of cellular networks
Elisa Bertino, Murat Kantarcioglu, Cuneyt Gurcan Akcora, Sagar Samtani, and IoT systems, and edge analytics and machine learning for cy-
Sudip Mittal, and Maanak Gupta. 2021. AI for Security and Security for bersecurity. Elisa Bertino is a Fellow member of IEEE, ACM, and
AI. In Proceedings of the Eleventh ACM Conference on Data and Application
AAAS. She received the 2002 IEEE Computer Society Technical
Security and Privacy (CODASPY ’21), April 26–28, 2021, Virtual Event, USA.
ACM, New York, NY, USA, 2 pages. https://doi.org/10.1145/3422337.3450357
Achievement Award for “For outstanding contributions to database
systems and database security and advanced data management
systems”, the 2005 IEEE Computer Society Tsutomu Kanai Award
Statement of Elisa Bertino for “Pioneering and innovative research contributions to secure
Cyberattacks with different goals, such as data ransoms, denial distributed systems”, the 2014 ACM SIGSAC Outstanding Contribu-
of service, sabotage, data theft, are on a dramatic increase. To make tions Award with citation “For her seminal research contributions
defenses more effective, recent solutions leverage AI techniques. and outstanding leadership to data security and privacy for the past
However, a major problem is that the application of AI techniques to 25 years”, and the 2019-2020 ACM Athena Lecturer Award.
cybersecurity is not trivial. For example, if one would like to apply Statement of Murat Kantarcioglu
reinforcement learning, one has to understand how to properly Direct application of AI techniques to cyber security domain
design reward functions. In addition different security tasks may may be misguided. Unlike most other application domains, cyber
need different AI techniques. Thus an effective AI-enhanced defense security applications often face adversaries who actively modify
in depth must be based on a comprehensive cybersecurity lifecycle their strategies to launch new and unexpected attacks. Thus AI tech-
and an analysis of adversarial tactics and techniques. Also, AI- niques for cyber security need to be resilient against the adaptive
based cybersecurity approaches have been intensely scrutinized behaviors of the adversaries, and are able to quickly detect previ-
with respect to ethics. Thus, for AI to be effectively and quickly ously unknown new attack instances. Recently, various adversarial
adopted in cybersecurity, AI security and ethics need also to be AI techniques (including our proposed techniques developed using
assured. AI security requires assurance processes for data used in Army Research Office funding) have been developed to counter
AI training as well as systematic approaches to AI security testing. adversaries’ adaptive behaviors. For example, in our earlier work,
we developed a game theoretic framework to discover an optimal
Permission to make digital or hard copies of part or all of this work for personal or set of attributes to build machine learning models against active
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
adversaries. In another work, we modified an existing, and popular
on the first page. Copyrights for third-party components of this work must be honored. machine learning tool named Support Vector Machine to be more
For all other uses, contact the owner/author(s). resistant against adversarial attacks. The attack models are defined
CODASPY ’21, April 26–28, 2021, Virtual Event, USA
in terms of the adversaries’ capabilities of modifying data. Our
© 2021 Copyright held by the owner/author(s).
ACM ISBN 978-1-4503-8143-7/21/04. solutions minimize the worst-case loss corresponding to the attack
https://doi.org/10.1145/3422337.3450357

333
Panel CODASPY '21, April 26–28, 2021, Virtual Event, USA

CODASPY ’21, April 26–28, 2021, Virtual Event, USA

models, and show that such tailored tools could be more resistant and Blockchain networks. He has worked at and collaborated with
to adversarial behavior compared to existing SVM alternatives. Qatar Computing Research Institute, Yahoo Research Barcelona,
Murat Kantarcioglu is a Professor in the Computer Science and Huawei Research in Istanbul. He is a Fulbright Scholarship
Department and Director of the Data Security and Privacy Lab recipient, and his research works have been published in leading
at The University of Texas at Dallas (UTD). He received a PhD conferences and journals such as IEEEtran, VLDB, ICDM, SDM,
in Computer Science from Purdue University in 2005 where he IJCAI, and ICDE.
received the Purdue CERIAS Diamond Award for Academic excel- Statement of Sagar Samtani
lence. He is also a visiting scholar at Harvard Data Privacy Lab. Dr. Modern society’s irreversible dependence on information tech-
Kantarcioglu’s research focuses on the integration of cyber secu- nology has placed a significant impetus on cybersecurity analysts to
rity, machine learning, data science and blockchains for creating enhance the confidentiality, integrity, and availability of their ever-
technologies that can efficiently and securely store, analyze and increasing asset-bases from a rapidly evolving threat landscape. In
share data and machine learning results. His research has been particular, methodologies such as deep learning, machine learning,
supported by grants including from NSF, AFOSR, ARO, ONR, NSA, network science, text analytics, and others can help human analysts
and NIH. He has published over 170 peer reviewed papers in top sift through large quantities of heterogeneous cybersecurity data
tier venues such as ACM KDD, SIGMOD, ICDM, ICDE, PVLDB, with unprecedented efficiency and effectiveness to detect patterns
NDSS, USENIX Security and several IEEE/ACM Transactions as missed by conventional approaches. To date, AI for cybersecurity
well as served as program co-chair for conferences such as IEEE has been leveraged in four major cybersecurity applications areas:
ICDE, ACM SACMAT, IEEE Cloud, ACM CODASPY. He is the re- (1) cyber threat intelligence (CTI) to create, manage, and leverage
cipient of various awards including NSF CAREER award, the AMIA information about emerging threats and key threat actors to enable
(American Medical Informatics Association) 2014 Homer R Warner effective cybersecurity decision making, (2) security operations cen-
Award and the IEEE ISI (Intelligence and Security Informatics) 2017 ters (SOCs) to assist human analysts in tactical tasks such as alert
Technical Achievement Award presented jointly by IEEE SMC and management, vulnerability management, security orchestration,
IEEE ITS societies for his research in data security and privacy. and others, (3) disinformation and computational propaganda to
Statement of Cuneyt Gurcan Akcora identify how fake content can sway public actions and perceptions,
Recent years have seen increased user participation in digital and (4) adversarial machine learning (AML) that relies on tech-
technologies. In the past, social networks had been the primary niques such as generative adversarial networks (GANs) to generate
venue for a user to reach a global audience. Nowadays a user has fake or synthesized content to enhance offensive and defensive
many devices that collect data and share on the web. From self- cybersecurity postures. Despite significant advancements in these
driving cars to mobile and smart home devices, users connect to a areas from both industry and academia alike, key challenges remain
wider audience through popular applications. This increased con- that require significant attention. First, many practitioners and aca-
nection has not always been a boon. Data collected from other demics often work in siloes. Second, there is currently a dearth of
users without proper auditing is used to train ML models and make publicly accessible datasets that accurately depict the nuances and
decisions with them. Companies seem to be most interested in in- complexities of production environments. Third, many AI-based
creasing their revenues by using the data - any data. Suddenly, we models deployed in cybersecurity contexts often lack interpretabil-
have found ourselves governed by decisions that are conditioned on ity. Finally, many students and faculty may lack the resources to
other people’s preferences and actions. Well-documented cases of start and/or execute their AI for cybersecurity research.
bias and racism seeping into machine learning models have caused Sagar Samtani is an Assistant Professor and Grant Thornton
outrage. With these harmful effects, it is imperative to learn how Scholar in the Department of Operations and Decision Technolo-
Machine Learning models use data to make decisions that affect gies at the Kelley School of Business at Indiana University (IU).
users’ lives. Governments have joined efforts as well and passed He is also a Fellow within the Center for Applied Cybersecurity
legislation that forces companies to explain their ML models’ deci- Research at IU. Samtani graduated with his Ph.D. in May 2018 from
sions. AI and ML researchers have responded to this growing issue the Artificial Intelligence Lab in University of Arizona’s Manage-
with novel directions called Interpretable (IAI) and Explainable ment Information Systems (MIS) department at the University of
(XAI) AI. Interpretability attempts to explain the cause and effect Arizona. From 2014 – 2017, Samtani served as a National Science
observed within a system, whereas Explainability deconstructs a Foundation (NSF) Scholarship-for-Service (SFS) Fellow. Samtani’s
machine learning system to explain it in human terms. Both areas research centers around Explainable Artificial Intelligence for Cy-
are developing tools that wrap around existing ML models and ex- bersecurity and cyber threat intelligence. Samtani has published
plain algorithm decisions and predictions. ML research must note over three dozen journal and conference papers on these topics
the growing fields of IAI and XAI, and attempt to explain their ML in leading venues such as MIS Quarterly, Journal of Management
models before model deployment. The benefits can be far-reaching Information Systems, ACM Transactions on Privacy and Security,
as we believe that this practice will speed up ML adoption in society. IEEE Intelligent Systems, Computers and Security, IEEE S&P, IEEE
Cuneyt Gurcan Akcora is an Assistant Professor of Computer ICDM, and others. His research has received nearly $1.8M (in PI
Science and Statistics at the University of Manitoba, Canada. He and Co-PI roles) from the NSF CICI, CRII, and SaTC-EDU programs.
received his Ph.D. from the University of Insubria, Italy. His primary Samtani has won several awards for his research, including the
research interests are Data Science on complex networks and large- ACM SIGMIS Doctoral Dissertation award in 2019.
scale graph analysis, with applications in social, biological, IoT

334