Adversial attack on machine learning

Project Title

Adversial attack on machine learning

Project Area of Specialization Artificial IntelligenceProject Summary

Machine learning algorithms have shown tremendous potential in dealing with classification and regression problems. They have shown that they compute complex data sets and achieve accurate results. This has led to boom in ML based products. It as become an integral part of many apps because it ML algorithms have the ability to learn from data and use this knowledge to enhance their capability. But with so much capability the ML algorithms are highly vulnerable to attacks. Adversarial attacks can fool a ML algorithm in making the wrong choice. It has the ability to manipulate the ML algorithm. In white box adversarial attack, the attack has all the knowledge of the model and may manipulate the data set to reduce efficiency of the model. black-box, which resembles a real-life scenario with the adversary having almost no knowledge of the model to be attacked. Considering how almost every app has some type of ML algorithms these types of attacks may prove costly.

Project Objectives

We aim to exploit the vulnerability of Machine learning algorithms through adversarial attacks. These attacks are vital for testing the robustness of a machine learning algorithm. Attacking machine learning algorithms can have devastating results on the model’s efficiency.

Project Implementation Method

Using Python, we will first create a classification algorithm using different Machine learning algorithms such as SVM, Linear regression etc. For the dataset we will use benchmark datasets to test out algorithm’s classification potential. After the above-mentioned task is completed will create an adversarial attack to target our dataset an attempt to misguide the classifier

Benefits of the Project

Machine learning is employed in multiple sectors such as Social media, Virtual personal assistants, Product Recommendations, Google Translation, Fraud Detection, Health Sector, Transportation and Commuting etc.

Since ML is used in numerous sectors it is essential to test is robustness. Our project aims to showcase the power of its adversaries. The vulnerability of ML could lead to serious losses

Technical Details of Final Deliverable

Python libraries such as NumPy, pandas, seaborn, Scikit-learn,Tkinter,Deep fool and The Adversarial Robustness Toolbox (ART),Datasets

Final Deliverable of the Project Software SystemCore Industry ITOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	0
Python libraries	Miscellaneous	10	0	0
Data sets	Miscellaneous	4	0	0