News
Next Seminar on 08.05.2024
Written on 06.05.2024 10:24 by Mang Zhao
Dear All,
The next seminar(s) take place on 08.05.2024 at 14:00 (Session A). Please note that there will be only ONE session.
Session A: (14:00-15:00)
Mario Beluri, Somrita Ghosh
https://cispa-de.zoom.us/j/96786205841?pwd=M3FOQ3dSczRabDNLb3F1czVXVUpvdz09
Meeting-ID: 967 8620 5841
Kenncode: BT!u5=
Session A:
14:00 - 14:30
Speaker: Mario Beluri
Type of talk: Master Intro
Advisor: Prof. Dr. Thorsten Holz, Dr. Bhupendra Acharya
Title: Exploration of the Dynamics of Buying and Selling of Social Media Accounts
Research Area: RA3: Threat Detection and Defenses / RA5: Empirical and Behavioural
Abstract: In recent years, there has been a rise in social media users with projections indicating approximately 4.95 billion active users as of 2024. These platforms have become an integral part of the daily lives of internet users, serving as one of the primary ways for socializing, communication, and information sharing. Unfortunately, social media has become a double-edged sword that while allowing its utilizers to take advantage of the connectivity and communication paths created, on the other hand, provides an environment that can be easily exploited.
Lately, social media platforms have been targeted for a variety of malicious activities including propaganda, hijacking, phishing, and scams. This has led to the growth of social media marketplaces that offer social media accounts for sale, where fraudulent accounts, among others, are openly being sold and used as harmful playgrounds rather than a means of communication.
In this work, we plan to conduct in-depth research on the buying and selling dynamics of social media profiles across a variety of frameworks. We aim to acquire complete metadata and user engagement facts from multiple social media networks. Our central objective is to find any patterns in these data and carry out a thorough analysis to identify any attack targets associated with these entries. Additionally, we intend to categorize such accounts based on the web category of the targeted brand.
We plan to unveil the scammer’s modus operandi of buying and selling social media profiles by segmenting target attacks into distinct categories. Thus, through our study, we aim to provide an end-to-end scam life cycle of scammers buying these social media profiles and launching attacks against the targeted market segments.
14:30 - 15:00
Speaker: Somrita Ghosh
Type of talk: Master Intro
Advisor: Dr. Xiao Zhang
Title: Enhancing Robust Training through Selective Unlabeled Data
Research Area: RA1
Abstract: Training for robustness is recognized to require a significantly larger dataset compared to standard training, as evidenced by extensive research in the field. The existing literature emphasises a substantial gap in the number of samples needed for achieving robust learning as opposed to standard learning. Previously, addressing this gap involved implementing semi-supervised learning techniques with a considerable amount of unlabeled data. However, even with semi-supervised learning, achieving high robust accuracy demands an extensive pool of unlabeled data and imposes substantial computational overhead. In our work, we propose alternative approaches to mitigate the mentioned complexity gap. Instead of relying on a large pool of unlabeled data, we aim to employ selection algorithms to choose a subset of this data, aiming to ensure that robust accuracy is not compromised. Our strategy involves selecting data closer to decision boundaries, with the intention of focusing more on data that genuinely contributes to improving robust accuracy by smoothing the decision boundary. By opting for a subset of data, we aim to reduce the memory and time complexity associated with general robust self-training algorithms.