September 21-22, 2024, Copenhagen, Denmark
Kamronbek Yusupov1, Md Rezanur Islam1, Insu Oh2, Mahdi Sahlabadi2, and Kangbin Yim2, 1Software Convergence, Soonchunhyang University, Asan-si, South Korea, 2Department of Information Security Engineering, Soonchunhyang University, Asan-si, South Korea
This research focuses on evaluating the security of an intrusion detection system in a CAN bus-based vehicle control network. A series of studies were conducted to evaluate the performance of models proposed by previous researchers, testing their effectiveness in real-world scenarios as opposed to those on which they were trained. The article demonstrates that models trained and tested on the same dataset can only sometimes be considered adequate. An approach that included models trained only on CAN ID, Payload, or full data was chosen. The research results show that such methods are ineffective enough in real-world attack scenarios because they cannot distinguish between new scenarios not presented during training. The results of testing the models in various attack scenarios are presented, and their limitations are identified. In addition, a new method is proposed explicitly for attack scenarios that may occur in the real-world use of an in-vehicle CAN communication system.
Intrusion Detection System, Controller Area Network, In-Vehicle Network, LSTM.
Omar Saad Almousa, Jordan University of Science and Technology, Jordan
Passwords are ubiquitous and this will continue for long. Strong passwords are a necessity to protect sensitive information. However, users not only tend to pick weak passwords, but also reuse them over several authentication systems. The existence of weak passwords in a system not only jeopardize that system, but also other systems with overlapping users because of password reuse phenomena. Investigating users’ behaviour in password creation leads to finding ways to avoid weak passwords. One aspect of that is to study the very passwords. In this study we analyse 662 passwords created by fresh students in our faculty. The students picked their passwords to authenticate themselves to a platform for programming practice and assignment solving. Our analysis relied on basic structural parameters such as password length, constructing characters, and entropy. To that end, we coined two definitions for weak and strong passwords. One is alphabet-based, and the other is entropy based. Accordingly, we found that majority of students do not tend to create strong passwords. We believe that this is due to the lack of enforcement of a strong password policy.
Passwords, Analysis, Weak password, Strong password.
Batuhan Erdogan, Bogazici University, Istanbul, Turkey
This study examines the limitations of OpenAIs ChatGPT models (GPT-3.5 and GPT-4) in interpreting and utilizing indexicals. While GPT-4 shows some performance improvements over GPT-3.5, both models frequently misinterpret indexicals in prompts and occasionally err in producing them in specific contexts. The models abilities vary with the type of contextual environment simulated by the user, demonstrating better competence in discrete environments and conversational implicatures. ChatGPT generally avoids context-dependent language in its responses. Through word frequency analysis of four demonstrative indexicals across essays written by humans and the two GPT models, we found GPT-4 significantly more likely to produce such indexicals than GPT-3.5. Inspired by Heideggers concept of Being-in-the-World, we propose a new training method using narratives with multiple first-person perspectives within a fictional world to enhance the models handling of pronominal indexicals.
Artificial Intelligence, Pragmatics, Semantics, Linguistics, Indexicals, LLMs, Artificial Neural Networks, Philosophy of Artificial Intelligence.
Sükran Sungur and Gülbin ÖzkanDepartment of Mathematics and Science Education, Uludag University, Bursa, Turkiye
The purpose of this study is to determine the opinions of pre-service science teachers about web-based teaching and distance education. A case study was carried out with undergraduate science teacher students (n=15) studying at a state university in Istanbul. The study was carried out through Material Design in Science Teaching lesson. Students took this course for 12 weeks and at the end of this course student opinions about web-based teaching and their opinions about distance education by moving from the experiences of students during the pandemic were received. Examining all the data reveals that while pre-service science teachers have many favorable opinions of web-based learning, they have few favorable opinions about distance education. Since students work with web 2.0 tools, they stated the advantages and disadvantages of using these tools in science education. Students suggestions regarding web-based and distance education will contribute to future studies about web-based and distance education.
Web-based teaching, distance education, science teaching, pre-service science teaching .
Simon M¨unker, Kai Kugler, and Achim Rettinger
Filtering and annotating textual data are routine tasks in many areas, like social media or news analytics. Automating these tasks allows to scale the analyses wrt. speed and breadth of content covered and decreases the manual effort required. Due to technical advancements in Natural Language Processing, specifically the success of large foundation models, a new tool for automating such annotation processes by using a text-to-text interface given written guidelines without providing training samples has become available. In this work, we assess these advancements in-the-wild by empirically testing them in an annotation task on German Twitter data about social and political European crises. We compare the prompt-based results with our human annotation and preceding classification approaches, including Naive Bayes and a BERT-based fine-tuning/domain adaptation pipeline. Our results show that the prompt-based approach – despite being limited by local computation resources during the model selection – is comparable with the fine-tuned BERT but without any annotated training data. Our findings emphasize the ongoing paradigm shift in the NLP landscape, i.e., the unification of downstream tasks and elimination of the need for pre-labeled training data.
foundation models, automating text annotation, zero-shot classification, social and political EU crises.