|
Introduction
|
|
|
DDoS Classification using Machine Learning
|
We are utilizing a specific model, logistic regression to classify network traffic data into benign and attack classes
The data being used is not raw network traffic data, but pre-processed data that is conducive to machine learning analysis
When a lot of training data is provided, machine learning is able to learn fairly accurate models
|
|
SVM Poisoning Attack
|
We are using a SVM model trained to recognize handwritten digits from the popular MNIST database
As part of the poisoning attack, malicious data is being added to the training set
The effectiveness of the attack is measured by evaluating on a test set with and without the poisoned samples
|
|
Adversarial Attacks on Neural Networks
|
We are using a trained neural network model trained to recognize handwritten digits from the popular MNIST database
As part of the attack, adversarial examples are being added which trick the network into returning a different label
The effectiveness of the binary thresholding method is demonstrated on grayscale images
The effectiveness of training the neural network with the adversarial examples added to the training set is also demonstrated
|