Philxxw's repositories
ELAD
We proposes a model to segment the original data and generate time subsequences. And we use the random concatenation within a certain time range to learn more complex data distributions. Then we designed a mask method of random continuous segments or random discrete points in partial position range of data and a corresponding objective function for this method. These methods are applied to the dual-stream self-attention model to learn a good distribution of training data. Anomaly detection depend on construction error between the reconstructed data output by the model and the original test data. The evaluation experiments are conducted on two public datasets from aerospace and a public server machine dataset from an Internet company. Our model has a total F1 score of 0.9526 in the three real datasets, which is significantly better than the best performing baseline.
Best_README_template
🌩最好的中文README模板⚡️Best README template
Network-Intrusion-Detection
Network Intrusion Detection KDDCup '99', NSL-KDD and UNSW-NB15