Skip to Main Content
 

Global Search Box

 
 
 
 

Files

ETD Abstract Container

Abstract Header

Distribution-Based Adversarial Multiple-Instance Learning

Abstract Details

2023, Master of Sciences, Case Western Reserve University, EECS - Computer and Information Sciences.
This thesis studies the effect of adversarial attacks on multiple-instance learning (MIL) methods and techniques to defend against such attacks. MIL uses data in the form of labeled sets (bags) of objects (instances). In this work, we show that the multiple-instance representation admits novel attacks where an adversary can alter the learned concept without manipulating any instance features, simply by changing the distribution of instances in bags. We introduce the False-positive Resampling Offense With Noise (FROWN), a bag-level attack that uses a resampling strategy to create adversarial bags that skew a dataset’s bag distribution. We also introduce the SMILe Defense (SMILeD), a bag-level defense that uses the previously studied Shuffled Multiple-Instance Learning (SMILe) resampling approach to recover from an attack by restoring the original bag distribution. Finally, we empirically evaluate the FROWN attack on a multiple-instance learner, as well as SMILeD’s mitigation of FROWN and an instance-level MIL attack from previous work.
Soumya Ray (Advisor)
Michael Lewicki (Committee Member)
Erman Ayday (Committee Member)
67 p.

Recommended Citations

Citations

  • Chen, S. (2023). Distribution-Based Adversarial Multiple-Instance Learning [Master's thesis, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1660307172211722

    APA Style (7th edition)

  • Chen, Sherry. Distribution-Based Adversarial Multiple-Instance Learning. 2023. Case Western Reserve University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1660307172211722.

    MLA Style (8th edition)

  • Chen, Sherry. "Distribution-Based Adversarial Multiple-Instance Learning." Master's thesis, Case Western Reserve University, 2023. http://rave.ohiolink.edu/etdc/view?acc_num=case1660307172211722

    Chicago Manual of Style (17th edition)