some related work about traffic and large language model

1. Transformer 是 RNN：具有线性注意力的快速自回归 Transformer

内容很好，略。

2. A survey of transformers

较全的一个survey

3. On The Computational Complexity of Self-Attention

结论：在强指数时间假设下，在最坏情况下，这里可能存在一种基本的“没有免费午餐”现象：似乎不太可能获得自注意力的可证明的次二次算法，同时对于所有输入也是可证明的近似准确的算法。

4. Training Vision Transformers with Only 2040 Images

1. ET-bert

2. BFCN: A Novel Classification Method of Encrypted Traffic Based on BERT and CNN

数据包level bert + 序列的CNN特征

3. Yet Another Traffic Classifier: A Masked Autoencoder Based Traffic Transformer with Multi-Level Flow Representation

实际上是flow 转5个数据包的图 （40*40），然后利用图像transformer进行。

4. M3F: A novel multi-session and multi-protocol based malware traffic fingerprinting

一阶马尔可夫

5. TSFN: A Novel Malicious Traffic Classification Method Using BERT and LSTM

数据包level bert + 序列的LSTM特征 和 BFCN同一个作者，设计细节没写。

6. I^2 RNN: An Incremental and Interpretable Recurrent Neural Network for Encrypted Traffic Classification

略

7. Low-Quality Training Data Only? A Robust Framework for Detecting Encrypted Malicious Network Traffic

略

8. FusionTC: Encrypted App Traffic Classification Using Decision-Level Multimodal Fusion Learning of Flow Sequence

略

9. FA-Net: More Accurate Encrypted Network Traffic Classification Based on Burst with Self-Attention

一眼看上去有点怪的模型

10. NetGPT: Generative Pretrained Transformer for Network Traffic

3个数据包的bert (转NLP)

11. Spatial-Temporal Feature with Dual-Attention Mechanism for Encrypted Malicious Traffic Detection

cnn+gru

12. Rosetta: Enabling Robust TLS Encrypted Traffic Classification in Diverse Network Environments with TCP-Aware Traffic Augmentation

2页短文？没看懂

13. Transformer-Based Device-Type Identification in Heterogeneous IoT Traffic

transformer + 一些trick做分类

14. FastTraffic: A lightweight method for encrypted traffic fast classification

略

1. Network Traffic Generation: A Survey and Methodology

分析了92个流量生成器工具

2. Generating practical adversarial network traffic flows using NIDSGAN

利用GAN生成对抗性flow，但是生成的是特征，不是flow。缺了一半

3. 用于物联网流量生成的知识增强 GAN

GAN+LSTM+知识图 生成 flow的时间序列

4. Flow-based network traffic generation using generative adversarial networks

GAN+flow特征的生成

5. (SIGCOMM'22) Practical GAN-based synthetic IP header trace generation using NetShare

合成数据包和流标头跟踪, gan +sequence +header

6. PAC-GAN: Packet Generation of Network Traffic using Generative Adversarial Networks

GAN + packet

7. DBWE-Corbat：使用动态词嵌入和网络范围对比学习生成后台网络流量

数据包-->ip+port

8. Design and Implementation of Traffic Generation Model and Spectrum Requirement Calculator for Private 5G Network

GAN学习流量的时间--字节数矩阵

9. DPNeT：具有生成对抗网络的差分专用网络流量合成

GAN+差分隐私+flow特征

10. FlowGAN - Synthetic Network Flow Generation using Generative Adversarial Networks

WGAN + 流特征

11. Generative Adversarial Networks (GANs): A Survey on Network Traffic Generation

一个比较全的survey

12. SyNIG: Synthetic Network Traffic Generation through Time Series Imaging

使用格拉姆角场(GAF)将时间序列数据转换为图像

13. Stan: Synthetic network traffic generation with generative neural models

卷积神经层 (CNN) 与混合密度神经层 (MDN) 和 softmax 层集成

1. Segmented Recurrent Transformer: An Efficient Sequence-to-Sequence Model

2. Block-State Transformers

3. 扩散器：具有长序列多跳注意力扩散的高效变压器

4. FlashAttention

1. 包头分布 + JensenShannon Divergence（netgpt）

2. 【NLP技术分享】文本生成评价指标的进化与推翻

3. 基于模型的相似度评价。

1. bert cls token.

1. Datasets are not enough: Challenges in labeling network traffic

1. A review on the attention mechanism of deep learning

1. my experiments