|
BGK 596E Seminars
CRN 21028
|
- This page is updated weekly.
- You may stay tuned.
|
| 5 |
| Title |
| Deep Learning-Based Network Intrusion Detection System |
| Date |
| April 24th, 2026 |
| Time |
| 10:30 (GMT+3) |
| Microsoft Teams |
Meeting ID:
***
Passcode:
***
(Please, contact the secretary's office.)
|
| Abstract |
|
The article titled "An Intrusion Detection Model With Hierarchical Attention Mechanism" introduces a Network Intrusion Detection System (NIDS) that uses deep learning techniques to enhance network security. Researchers have developed a model that treats network traffic as time-series data and is based on a bidirectional gated recursive unit (BiGRU) architecture. The model's key feature is its prioritization of critical data using a hierarchical attention mechanism that focuses on both traffic characteristics and time periods. Tests on the UNSW-NB15 dataset demonstrate that the system achieves a high accuracy rate exceeding 98.76% and a very low false alarm rate (lower than 1.2%). Furthermore, the visualization maps used make the model's decision-making process transparent by identifying which features are more critical in intrusion detection. This approach provides a more adaptable and sensitive defense mechanism against complex and constantly evolving cyber threats, unlike traditional methods.
|
|
| 4 |
| Title |
| A Neural Probabilistic Language Model |
| Date |
| April 17th, 2026 |
| Time |
| 10:30 (GMT+3) |
| Microsoft Teams |
Meeting ID:
***
Passcode:
***
(Please, contact the secretary's office.)
|
| Abstract |
|
A goal of statistical language modeling is to learn the joint probability function of sequences of words in a language. This is intrinsically difficult because of the curse of dimensionality: a word sequence on which the model will be tested is likely to be different from all the word sequences seen during training. Traditional but very successful approaches based on n-grams obtain generalization by concatenating very short overlapping sequences seen in the training set. We propose to fight the curse of dimensionality by learning a distributed representation for words which allows each training sentence to inform the model about an exponential number of semantically neighboring sentences. The model learns simultaneously (1) a distributed representation for each word along with (2) the probability function for word sequences, expressed in terms of these representations. Generalization is obtained because a sequence of words that has never been seen before gets high probability if it is made of words that are similar (in the sense of having a nearby representation) to words forming an already seen sentence. Training such large models (with millions of parameters) within a reasonable time is itself a significant challenge. We report on experiments using neural networks for the probability function, showing on two text corpora that the proposed approach significantly improves on state-of-the-art n-gram models, and that the proposed approach allows to take advantage of longer contexts.
|
|
| 3 |
| Title |
| Research on Efficient Packet Filter Technology Based on eBPF |
| Date |
| April 10th, 2026 |
| Time |
| 10:30 (GMT+3) |
| Microsoft Teams |
Meeting ID:
***
Passcode:
***
(Please, contact the secretary's office.)
|
| Abstract |
|
The research on efficient packet filter technology based on eBPF aims to address the performance bottleneck of traditional kernel protocol stacks in high concurrency scenarios. This paper proposes an efficient network packet filtering scheme based on eBPF and XDP technologies. By mounting the eBPF program at the kernel driver layer, zero copy processing is achieved by bypassing the TCP/IP protocol stack, and combining hash table rule matching to reduce time complexity to O(1). The experiment shows that this scheme supports dynamic loading of filtering rules and can accurately achieve packet filtering in both single rule and multi rule scenarios, significantly improving processing efficiency compared to traditional tools such as Iptables. The technical core includes: using XDP hooks to directly arbitrate data packets at the network card driver layer, ensuring program security through eBPF validators, and using mapping systems to achieve efficient interaction between user mode and kernel mode. This technology can be applied to DDoS defense, sensitive port access control, and other scenarios.
|
|
| 2 |
| Title |
| Finding Preimages in Full MD5 Faster Than Exhaustive Search |
| Date |
| April 3rd, 2026 |
| Time |
| 10:30 (GMT+3) |
| Microsoft Teams |
Meeting ID:
***
Passcode:
***
(Please, contact the secretary's office.)
|
| Abstract |
|
In this paper, we present the first cryptographic preimage attack on the full MD5 hash function. This attack, with a complexity of 2^116.9, generates a pseudo-preimage of MD5 and, with a complexity of 2^123.4, generates a preimage of MD5. The memory complexity of the attack is 2^45 × 11 words. Our attack is based on splice-and-cut and local-collision techniques that have been applied to step-reduced MD5 and other hash functions. We first generalize and improve these techniques so that they can be more efficiently applied to many hash functions whose message expansions are a permutation of message-word order in each round. We then apply these techniques to MD5 and optimize the attack by considering the details of MD5 structure.
|
|
| 1 |
| Title |
| Human Aspects of Cyber Security for Computing Higher Education: Current Status and Future Directions |
| Date |
| March 6th, 2026 |
| Time |
| 10:30 (GMT+3) |
| Microsoft Teams |
Meeting ID:
***
Passcode:
***
(Please, contact the secretary's office.)
|
| Abstract |
|
The human aspects of cybersecurity are fundamental as these are interlinked with processes and technology in building resilience against an evolving and complex threat landscape. It is vital to teach future cybersecurity specialists sufficient knowledge about human aspects in order to strengthen defenses and defend against malicious actors. The challenges in the cybersecurity course and curriculum design in higher education arise from the ambiguity of what is meant by 'human aspects' in cybersecurity and the differing demands of a multitude of cybersecurity job profiles. As a result, choosing what human aspects of knowledge and skills to teach and balancing this with technical content is a multiplex problem. We review the existing cybersecurity curricular guidance on the human aspects and perform a systematic literature review of academic research. The review covers what human aspects content is included in existing cybersecurity courses and curricula, what the design philosophies and pedagogical approaches to teaching human aspects are, and how the effectiveness of the content coverage and pedagogical approaches is evaluated in higher education with cybersecurity specialisation. The main findings are that although high-level guidance is available, there is no academic agreement and common vocabulary on the human aspects of cybersecurity. The research covering teaching human aspects is often very high-level or focuses on a narrow topic (e.g., social engineering). The published research and sharing knowledge on the pedagogical approaches and evaluation varies in quality and is often published based on the USA's universities' experience. The interdisciplinary approach in cyber security, where human behaviour is a crucial component (i.e., human-centric cyber security), has been discussed for years. However, only when giving enough human aspects focus in the curricula and choosing appropriate pedagogical approaches will academia shape the future of cybersecurity education to achieve this goal.
|
|