HPDS Group Meeting Presentation

Schedule and Selected Titles

Please sort the Date column from latest to former.

Link to a paper is recommended to be a DOI URL.

Default order: 柏叡→承瀚→守維→俊凱→冠宏

Date Speaker Minute Taker Presentation Title Keywords
6/8 俊凱 柏叡 Working Report (None)
6/8 守維 柏叡 Working Report (None)
6/1 承瀚 俊凱 Working Report (None)
6/1 柏叡 俊凱 Working Report (None)
6/1 冠宏 俊凱 Working Report (None)
5/25 俊凱 承瀚 Working Report (None)
5/25 守維 承瀚 Working Report (None)
5/18 承瀚 守維 Working Report (None)
5/18 柏叡 守維 Working Report: Thesis Introduction and Experiments (None)
5/18 冠宏 守維 Working Report (None)
5/11 俊凱 冠宏 Working Report (None)
5/11 守維 冠宏 Working Report (None)
5/4 冠宏 俊凱 Working Report (None)
4/27 俊凱 柏叡 Working Report (None)
4/27 守維 柏叡 Working Report (None)
4/20 承瀚 守維 Working Report (None)
4/20 柏叡 守維 Magpie: Automatically Tuning Static Parameters for Distributed File Systems using Deep Reinforcement Learning Performance optimization,
Parameter tuning,
Reinforcement learning,
Distributed storage systems,
Cluster configuration
4/20 冠宏 守維 Working Report (None)
4/13 俊凱 承瀚 Working Report (None)
4/13 守維 承瀚 Working Report (None)
3/29 承瀚 俊凱 Working Report (None)
3/29 冠宏 俊凱 Working Report (None)
3/23 俊凱 冠宏 Working Report (None)
3/23 守維 冠宏 Working Report (None)
3/16 承瀚 守維 Working Report (None)
3/16 柏叡 守維 Working Report: Thesis Introduction and Related Work (None)
3/16 冠宏 守維 Working Report (None)
3/9 俊凱 柏叡 Working Report (None)
3/9 守維 柏叡 Working Report: RPL Evaluation and Proposed Enhancement Method (None)
3/2 承瀚 俊凱 DroidEvolver: Self-Evolving Android Malware Detection System Feature extraction,
Malware,
Aging,
Training,
Adaptation models,
Manuals,
Labeling
3/2 柏叡 俊凱 Working Report and Storage Benchmarking with Deep Learning Workloads [PDF] Storage systems,
Object storage,
Database,
DL,
Performance evaluation
3/2 冠宏 俊凱 Working Report (None)
2/23 俊凱 承瀚 Working Report (None)
2/23 守維 承瀚 Working Report (None)
2/16 承瀚 守維 Working Report (None)
2/16 柏叡 守維 Working Report: Disk Benchmark Summary (None)
2/16 冠宏 守維 Working Report (None)
2/9 守維 柏叡 Working Report: Experimental Insights: Problems and Solutions (None)
1/26 承瀚 俊凱 Working Report (None)
1/26 柏叡 俊凱 Working Report: Degree Thesis Proposal (None)
1/26 冠宏 俊凱 Working Report (None)
1/26 俊凱 冠宏 Working Report (None)
1/26 守維 冠宏 Working Report (None)
1/12 柏叡18 守維 Working Report: Degree Thesis Proposals (None)
1/12 冠宏16 守維 Working Report (None)
1/5 守維17 承瀚 Working Report: TabNet Implementation & Evaluation (None)
12/29 承瀚17 俊凱 Working Report (None)
12/29 柏叡17 俊凱 Working Report: A Survey of Metrics for Distributed System Benchmarks (None)
12/28 冠宏15 俊凱 Working Report (None)
12/22 俊凱16 柏叡 A Lean and Modular Two-Stage Network Intrusion Detection System for IoT Traffic Training,
Measurement,
Network intrusion detection,
Complexity theory,
Proposals,
Internet of Things,
Security,
Computer crime,
Optimization,
Open source software
12/22 守維16 柏叡 Working Report: BERT NIDS Implementation (None)
12/15 承瀚16 守維 Working Report (None)
12/15 柏叡16 守維 Working Report: Disk & Compression Benchmark (None)
12/15 冠宏14 守維 Working Report (None)
12/8 俊凱15 冠宏 Working Report (None)
12/8 守維15 冠宏 Unknown Network Attack Detection Based on Open-Set Recognition and Active Learning in Drone Network (Unknown)
12/1 承瀚15 俊凱 Working Report (None)
12/1 柏叡15 俊凱 Working Report: Grafana and Gzip (None)
12/1 冠宏13 俊凱 Working Report (None)
11/24 俊凱14 承瀚 Working Report (None)
11/24 守維14 承瀚 Working Report (None)
11/17 承瀚14 俊凱 Working Report (None)
11/17 柏叡14 俊凱 Working Report: Disk Benchmakrk & Environment Setup (None)
11/17 冠宏12 俊凱 Predicting Network Attacks with CNN by Constructing Images from Netflow Data NetFlow data,
Intrusion detection,
Deep Learning,
CNN,
ResNet
11/21 俊凱13 自己 Working Report (None)
11/10 守維13 柏叡 Working Report: Open Set Recognition (OSR) Task (None)
11/3 承瀚13 守維 Confidence May Cheat : Self-Training on Graph Neural Networks under Distribution Shift Graph Neural Networks,
Self-Training,
Information Gain
11/3 柏叡13 守維 Working Report: DFS Choices for Out Lab: Weighing Pros and Cons (None)
11/3 冠宏11 守維 Working Report (None)
10/20 俊凱12 冠宏 Working Report: Transformer-LSTM (None)
10/20 守維12 冠宏 Working Report: Open Set Recognition (OSR) Task (None)
10/13 承瀚12 守維 Working Report (None)
10/13 冠宏10 守維 Working Report (None)
10/13 柏叡12 守維 Handover from Tse-An Lin (see below) (None)
10/6 守維11 承瀚 Working Report (None)
10/6 俊凱11 承瀚 An Intrusion Detection Method Based on Transformer-LSTM Model Intrusion detection,
Transformer,
LSTM,
Deep learning
9/29 柏叡11 俊凱 Handover from Tse-An Lin (see below) (None)
9/29 承瀚11 俊凱 Energy-based Out-of-distribution Detection (None)
9/29 冠宏9 俊凱 Considerably Improving Clustering Algorithms Using UMAP Dimensionality Reduction Technique: A Comparative Study Dimensionality reduction,
UMAP,
Clustering,
Embedding manifold,
Big data analytics,
ML,
Comparative study
9/22 俊凱10 柏叡 Network Intrusion Detection via Flow-to-Image Conversion and Vision Transformer Classification NIDS,
Flow-to-image conversion,
CNN,
Vision transformers,
Image classification
9/22 守維10 柏叡 A Hybrid Approach to Network Intrusion Detection Based on Graph Neural Networks and Transformer Architectures Graph neural network,
GraphSAGE,
Transformer,
NIDS
9/15 柏叡10 守維 Handover from Tse-An Lin (see below) (None)
9/15 承瀚10 守維 Working Report (None)
9/8 柏叡9 冠宏 Handover from Tse-An Lin: A MLOps Framework for Enhancing Self-Training utilizing In-Memory Caching of Pseudo-Labels (None)
9/8 俊凱9 冠宏 Using Consumer Lag for Autoscaling on Kafka-Centric Model Serving (None)
9/8 守維9 冠宏 Working Report: Implementation of FlowTransformer (None)
8/25 承瀚9 俊凱 Working Report (None)
8/25 冠宏8 俊凱 Mean teachers are better role models:Weight-averaged consistency targets improve semi-supervised deep learning results (None)
8/18 俊凱8 柏叡 An Attention-Based Convolutional Neural Network for Intrusion Detection Model NIDS,
CNN,
Computational complexity,
Security,
Network systems,
Image synthesis,
Computational efficiency
8/18 守維8 柏叡 FlowTransformer: A Transformer Framework for Flow-Based Network Intrusion Detection Systems Transformers,
NIDS,
Machine learning,
Generative pre-trained transformer,
Network flow
8/11 承瀚8 守維 Handover: Enhancing Self-Training by Feature-Fusion and Similarity-Based Weighting for Botnet Detection
8/11 柏叡8 守維 A Cost-Efficient Container Orchestration Strategy in Kubernetes-Based Cloud Computing Infrastructures with Heterogeneous Resources Cluster management,
Container orchestration,
Resource heterogeneity,
Cost efficiency
8/11 冠宏7 守維 Handover: Using a Hierarchical Clustering Approach with a Distribution Similarity Strategy for Multi-Class Botnet Labeling in Real-World Traffic
8/4 俊凱7 柏叡 Sentiment Analysis Using Pre-Trained Language Model With No Fine-Tuning and Less Resource Sentiment analysis,
Natural language processing
8/4 守維7 柏叡 RTIDS: A Robust Transformer-Based Approach for Intrusion Detection System NIDS,
Feature representation,
Self-attention mechanism,
Transformer
6/2 承瀚7 守維 Feature extraction for machine learning-based intrusion detection in IoT networks Feature extraction,
Machine learning,
NIDS,
IoT
5/26 柏叡7 承瀚 Heats: Heterogeneity-and Energy-Aware Task-Based Scheduling (None)
5/19 冠宏6 柏叡 Adaptive Intrusion Detection in the Networking of Large Scale LANs with Segmented Federated Learning Cybersecurity,
Deep learning,
NIDS,
Segmented-federated learning,
LAN,
CNN
5/12 俊凱6 冠宏 Local LLM + RAG (None)
5/5 守維6 俊凱 Mamba: Linear-Time Sequence Modeling with Selective State Spaces (None)
4/28 承瀚6 守維 Self-organizing Maps (SOM) (None)
4/21 柏叡6 承瀚 KubeShare: A Framework to Manage GPUs as First-Class and Shared Resources in Container Cloud Cloud computing,
GPU,
Container,
Scheduling,
Cloud computing,
GPU,
Container,
Scheduling
4/14 冠宏5 柏叡 Federated Deep Learning for Zero-Day Botnet Attack Detection in IoT-Edge Devices Botnet detection,
Cybersecurity,
Deep learning,
Deep neural network,
Federated learning,
IoT
3/31 守維5 冠宏 Efficiently Modeling Long Sequences with Structured State Spaces (None)
3/24 承瀚5 守維 BotChase: Graph-Based Bot Detection Using Machine Learning Security management,
Botnet detection,
Machine learning
3/17 柏叡5 承瀚 NetMARKS: Network Metrics-AwaRe Kubernetes Scheduler Powered by Service Mesh K8s,
Network statistics,
Scheduling,
Service mesh,
Latency,
Interoperability,
5G,
Containerized network
3/10 俊凱5 柏叡 Azure OpenAI (None)
3/3 冠宏4 俊凱 A Deep Learning Model for Network Intrusion Detection with Imbalanced Data NIDS,
Bi-LSTM,
Attention mechanism,
NSL-KDD
2/4 俊凱4 冠宏 Azure OpenAI (None)
1/28 承瀚4 俊凱 BotGM: Unsupervised Graph Mining to Detect Botnets in Traffic Flows IP networks,
Ports (Computers),
Security,
Windows,
Silicon,
Focusing
1/21 守維4 承瀚 Hunt for Unseen Intrusion: Multi-Head Self-Attention Neural Detector Deep neural network,
NIDS,
Multi-head attention,
Realistic prediction performance evaluation,
Self-attention
2024/
1/14		 柏叡4 守維 A Novel Flow-vector Generation Approach for Malicious Traffic Detection Deep learning,
Malicious traffic,
Embedding,
Attention mechanism
12/24 冠宏3 柏叡 E-GraphSAGE: A Graph Neural Network based Intrusion Detection System for IoT Graph Neural Networks,
NIDS,
IoT
12/17 俊凱3 冠宏 A Fuzzy Logic based feature engineering approach for Botnet detection using ANN Artificial Neural Networ,
BotnetFuzzy Logic,
Fuzzy rules,
CTU-13,
Cyber Security
12/10 守維3 俊凱 A Neural Attention Model for Real-Time Network Intrusion Detection NIDS,
Deep learning,
Attention model,
Network security
12/3 承瀚3 守維 Evading Machine Learning Botnet Detection Models via Deep Reinforcement Learning Botnet,
Adversarial,
Reinforcement learning
11/26 冠宏2 承瀚 IoT Malware Network Traffic Classification using Visual Representation and Deep Learning Network traffic,
Machine learning,
Security,
NIDS
11/19 柏叡3 冠宏 Unsupervised Detection of Botnet Activities using Frequent Pattern Tree Mining Botnet detection,
Internet security,
Frequent pattern tree,
Data mining
11/12 俊凱2 柏叡 An Advanced Computing Approach for IoT-Botnet Detection in Industrial Internet of Things Botnet,
Malware,
Feature extraction,
IIoT,
Static analysis,
Informatics,
Heuristic algorithms
11/5 守維2 俊凱 A Unified Approach to Interpreting Model Predictions [PDF] (none)
10/29 承瀚2 守維 Evading Machine Learning Botnet Detection Models via Deep Reinforcement Learning Botnet,
Adversarial,
Reinforcement learning
10/15 冠宏1 柏叡 Train Without Label: A Self-supervised One-Class Classification Approach for IoT Anomaly Detection IoT,
Self-supervised learning,
One-class classification,
Cyber-attacks,
NIDS
10/8 俊凱1 冠宏 A Hybrid Model for Botnet Detection using Machine Learning Botnet detection,
kmeans,
Rule-based system,
decision tree,
CTU-13 dataset
9/24 守維1 俊凱 BotStop : Packet-based efficient and explainable IoT botnet detection using machine learning IoT,
Botnet,
NIDS,
Explainable machine learning
9/17 承瀚1 守維 Botnet Detection in the Internet of Things using Deep Learning Approaches Botnet,
Computer crime,
Machine learning,
Servers,
IoT
2023/
9/10		 柏叡1 承瀚 A KNN-Based Intrusion Detection Model for Smart Cities Security Smart cities,
IoT,
Security,
AI,
Machine learning,
Classification

Keyword abbreviations:

  1. NIDS: Network intrusion detection, network intrusion detection system, intrusion detection system, intrusion detection
  2. ML: Machine Learning
  3. IoT: Internet of Things
  4. IIoT: Industrial Internet of Things
  5. CNN: Convolutional neural network, convolutional neural networks
  6. K8s: Kubernetes
  7. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
  8. LSTM: Long short-term memory

Notes and Guidelines

1. In Preparation

Recommended outlines of a paper (including its slides):

  1. Abstract (摘要): Write your own English version. Read out the English version word by word, and then briefly introduce it in Chinese.
  2. Introduction (介紹含背景)
  3. Related Works (相關研究成果)
  4. Proposed Approach (或Methodology，提出的解決方案)
  5. Experimentation (或Evaluation，實驗及結果)
  6. Conclusion and Future Work (結論和可改善的點)

Related to the choice of paper to deliver a speech:

  1. Avoid papers whose contribution is applying some approaches only. This is not researches, but experiment reports. (2023-09-10 謝老師給柏叡)
  2. Proposed approach can train with or handle newer (10 years ago to present) botnet datasets. (2023-09-10 俊又給柏叡)
  3. Figure out what aspects are challenging and forward-looking related to our researches. Specifically, for session-based approaches (instead of packet-based.) (2023-09-10 謝老師給柏叡) For example, RNN and attention are not practical in our researches, for such techniques require time sequence in packets which NetFlow doesn't provide. (2023-09-17 謝老師給承瀚)
  4. Avoid outdated approaches. Specifically, rule-based machine learning (RBML). (2023-11-19 謝老師給柏叡)
  5. Understand implementation details to the extent that you can reproduce it. Details include where and how each procedure runs, how to set initial parameters and update them, how to generate/preprocess each dataset, how to retrieve or calculate each metrics. You cannot understand partially or superficially. (2023-09-17 張老師給承瀚、2023-09-24 謝老師和張老師給守維、2023-10-08 謝老師和張老師給冠宏、2023-10-08 謝老師給以薰、2023-10-15 謝老師和張老師給冠宏)
  6. Prevent presenting papers whose statements and experiment results are suspicious (possibly fake). (2023-10-08 謝老師給冠宏)
  7. If some related work is comparable to the propose approach, such a proposition is less valuable. (2023-09-24 謝老師給守維) (2023-10-15 張老師給冠宏)

Related to slides, highly recommended to follow:

  1. Tell audience only the bottom line (重點、重要的事實). (2023-09-10 俊又給柏叡) You could highlight them in paragraphs. (2023-09-17 張老師給承瀚)
  2. Replace ambiguous adjectives with precise numbers or practical solutions, or supplement them after adjectives. It's because we are researchers. (2023-09-10 俊又給柏叡)
  3. Prevent typos and grammar errors. (2023-09-17 張老師給承瀚) Only humans can "propose" an approach; a thesis cannot. (2023-10-08 謝老師給冠宏) Machines are infected before attacking. Self-supervised learning vs self learning: Self-supervised learning focuses on learning representations from data, while self-learning focuses on learning behaviors or policies through interaction with an environment. (2023-10-15 謝老師給冠宏)

Related to slides, better if followed:

  1. Distinguish different approaches by preprocessing time, system-designing effort, training time, testing time, testing memory usage, testing F\(_1\) score (or testing accuracy, precision and recall.) (2023-09-10 俊又給柏叡)
  2. Think how can propose approaches apply in real world. (2023-10-15 謝老師給冠宏)

Related to working reports:

  1. It is a good direction to improve seniors' work (like standing on the shoulders of giants) by identifying potential improvements, making it complete, applying it in real world after researching related work. (2023-09-10 謝老師給碩星、2023-09-17 謝老師給哲賢、2023-10-08 謝老師和張老師給碩星、2023-10-08 謝老師給以薰)
  2. The following directions are hard for us to research: federated learning. (2023-09-17 謝老師給子豪) (2023-09-17 張老師給子豪)
  3. Figure out the bottlenecks of the proposed approach before finding their solutions. (2023-09-24 張老師給澤安)
  4. Try to gradually transfer (redirect) workloads from old to new system for migration. (2023-09-10 俊又給碩星)
  5. Think of data dependency and parallelism when designing architecture and system. (2023-09-10 俊又給紘維)
  6. Testing with more datasets, maybe with the latest or real-world ones, proves viability of proposed approaches. Old datasets and researches cannot reflect the network traffic today. (2023-09-17 謝老師給俊昇) (2023-09-17 張老師給哲賢)
  7. Try to derive applications from experiments as contributions. (2023-09-17 謝老師給俊昇)
  8. In implementation, you could try to request source code from the authors of related work to compare. (2023-09-17 張老師給俊昇)
  9. In experiments, you have to label real-world traffic before evaluation. (2023-09-17 張老師給哲賢)
  10. Generalize your ideas/policies rather than experiments with method of exhaustion (all cases). (2023-09-24 謝老師給紘維、2023-10-15 張老師給哲賢)

todo: from 2023-10-15哲賢Da-Jun

2. Two Days before Meeting

When you're about (in 2 days) to have paper report, please send the title of paper to the thread 報論文的時間和主題. If possible, update the schedule part in this notebook to prevent the others from choosing the same paper for presentation.

3. During Meeting

Speaker (報告人)

(todo)

Minutes Taker (作會議記錄的人)

The next scheduled speaker* has to take the meeting minutes, including the following content for each speaker:

  1. Name of the speaker.
  2. Title of the presentation.
  3. Questions and suggestion from teachers Shieh and Chang, and from the senior Jun-You. (Answers from the speaker can also be included.)

The following text block is a sample meeting minutes. (Thanks 守維 and 柏叡.)

Chinese version:

2023/09/17 會議紀錄

林俊昇：QLD-IDS working report

謝老師：
* 多找幾個dataset試看看
* 做完實驗後要思考如何應用，發揮最大效果

張老師：
* 可以嘗試向原作者詢問原始碼，看和自己重現的部分有哪些差異

------------------------------

(Minutes in the same format as the above for the other speakers.)

English version:

Meeting Minutes 2023-09-17

SPEAKER:

Title: TODO

Teacher Shieh:
1. TODO

Teacher Chang:
1. TODO

Senior Jun-Yo:
1. TODO

------------------------------

(Minutes in the same format as the above for the other speakers.)

* The next scheduled speaker: For example, when 柏叡 is reporting, the next speaker is 承翰, referred from the default order in the schedule.

4. After Meeting

After the meeting ends, send the meeting minutes to the channel 會議記錄 in Discord.

Pronunciation Correction

Select a repo