Part-Time Research Intern - Kesidis Research - Engineering

• CURRENT PENN STATE EMPLOYEE (faculty, staff, technical service, or student), please login to Workday to complete the internal application process. Please do not apply here, apply internally through Workday.

• CURRENT PENN STATE STUDENT (not employed previously at the university) and seeking employment with Penn State, please login to Workday to complete the student application process. Please do not apply here, apply internally through Workday.

• If you are NOT a current employee or student, please click "Apply" and complete the application process for external applicants.

JOB DESCRIPTION AND POSITION REQUIREMENTS:

Computer Science and Engineering seeks to hire multiple part-time research interns.

Interns will Implement a successful gradient-free test-time evasion attack against a classifier trained for the MNIST domain.

Approaches to gradient-free test-time evasion attacks can be found in a paper that has been added to the course's Extra Materials. You are asked to implement, in Matlab, an algorithm to perform gradient-free TTE attacks (targeted attacks). You should attack each test sample and misclassify it to a (randomly chosen) target class. You should report the attack success rate and the average number of classifier queries needed to achieve a successful attack. You should display representative attacked image examples to show whether the attack is perceptible or imperceptible. You should consider at least two different gradient-free attack algorithms, or two variants of the same algorithm. Also suggest a simple algorithm for detecting attacks and reports its performance (true detection rate at a fixed false positive rate, or ROC AUC curve, by varying the true detection rate versus false positive rate, by changing the detection threshold).

CAMPUS SECURITY CRIME STATISTICS: