Fall 2020 CS Cyber Security Reading Group
Members: Yashwanth Bandala, Yu Cai, Bo Chen (co-advisor), Niusen Chen, Shashank Reddy Danda, Siva Krishna Kakula, Vishnu Kamaraju, Alex Larkin, Jean Mayo (co-advisor), Joe Muhle, Jagi Rinchin, Jonah Schulte, Sai Venkateswaran, Sankalp Shastry, Shuo Sun, Deepthi Tankasala, Yuchen Wang, Wen Xie, Xiaoyong Yuan
Detailed schedule:
Time: 3:00 – 4:00pm Friday, September 18, 2020
Location: Zoom
Presenter: Dr. Xiaoyong Yuan (Assistant Professor, Applied Computing and
Computer Science)
Title: Beyond Class-Level
Privacy Leakage: Breaking Record-Level Privacy in Federated Learning
Abstract:
Federated learning (FL)
enables multiple clients to collaboratively build a global learning model
without disclosing their own data, such as the virtual keyboard (Gboard), for privacy protection. However, recent research
found privacy leakage of FL, especially on image classification tasks, such as
the client-level reconstruction of class representative. Nevertheless, such
analysis on image classification tasks is not applicable to uncover the privacy
threats against natural language processing (NLP) tasks, such as the language
modeling used in Gboard, whose records composed of
sequential texts cannot be grouped as class representatives. Moreover, due to
the finer granularity, it is challenging to distinguishably extract individual
records from a specific client, which is a serious threat to leak more precise
private information. This paper gives the first attempt to explore the
record-level privacy leakage against NLP tasks in FL. We propose a framework to
investigate the exposure of the record of interest in federated aggregations
based on the perplexity of language modeling. By monitoring the exposure
patterns, two correlation attacks are proposed to identify the corresponding
clients when extracting their specific records. Extensive experiments
demonstrate the effectiveness of the proposed attacks. We then investigate
several countermeasures to mitigate the attacks, which, however, are
ineffective.
Time: 2:00 – 3:00pm Friday, October 2, 2020
Location: Zoom
Presenter: Shashank Reddy
Title: Towards Mitigating Spreading of Coronavirus via
Mobile Devices
Abstract:
Recently,
the impact of coronavirus has been witnessed by almost every country around the
world. To mitigate spreading of coronavirus, a fundamental strategy would be
reducing the chance of healthy people from being exposed to it. Having observed
the fact that most viruses come from coughing/sneezing/runny nose of infected
people, in this work we propose to detect such symptom events via mobile
devices possessed by most people in the modern world and, to instantly
broadcast locations where the symptoms have been observed to other people. This
would be able to significantly reduce risk that healthy people get exposed to
the viruses. The mobile devices today are usually equipped with various sensors
including microphone, accelerometer, and GPS, as well as network connection
(4G, LTE, Wi-Fi), which makes our proposal feasible. Further experimental
evaluation shows that coronavirus-like symptoms (coughing/sneezing/runny nose)
can be detected with an accuracy around 90%; in addition, the dry cough (more
likely happening to COVID-19 patients) and wet cough can also be differentiated
with a high accuracy. Implementation using mobile devices is shown along with
some demo videos.
Time: 2:00 – 3:00pm
Friday, October 23, 2020
Location: Zoom
Presenter: Niusen Chen
Title and Abstract will be added.
Time: 3:00 – 4:00pm
Friday, November 6, 2020
Location: Zoom
Presenter: Jonah Schulte
Title: Mouse Trap: Forget USB Sticks, Attacks Could
Come from Essential Peripheral Devices
Abstract:
In
our continually digitizing world, the threat of malware has continued to rear
its ugly head, and expand, despite humanity's best efforts. The practice of
infecting a USB storage device, and leaving on the ground near a target has
become all too common the attempt to compromise a system, and more often than
not, it succeeds. This study, and subsequent paper, endeavor to highlight the
very real, and potentially very dangerous, act of using peripheral devices,
instead of common USB sticks, to carry and execute malware on a device. In this
study, the researchers strived to show the weakness of peripheral devices and
their liability for malware. They worked on the mouse pictured here, the
Logitech G600.
Time: 3:00 – 4:00pm
Friday, November 20, 2020
Location: Zoom
Presenter: Deepthi Tankasala
Title: Blockchain Technology on different
platforms
Abstract:
When
we say Blockchain, the prominent application we remember is cryptocurrency.
There are many more technologies where Blockchain can be applied starting from
the Internet of Things, Artificial Intelligence, and many more. This
presentation talks about the different applications along with their
architectures that are being developed where the inclusion of blockchain can be
a game-changer in terms of security and availability. In this talk, I plan to
present some of the most prominent applications of Blockchain technology other
than the famously known cryptocurrencies.