Student Research

Intern Projects

Some of examples of recent student projects that I've supervised. All information posted with student's permission. 

Conceptual Spaces for Semantic Communication

Dylan Wheeler, Kansas State University

In this ongoing work, we study an approach to semantic communication using conceptual spaces. Semantic communication is concerned with conveying meaning rather than faithfully recovering symbols. Conceptual spaces are geometric models of meaning which are based on theories of human concept learning. By representing data in the conceptual space, we can reduce the amount of information which must be transmitted and increase our robustness to syntactical errors. 

A representation of the color spindle. The vertical axis captures brightness, the distance from this axis represents saturation, and the angle represents hue.
A representation of three users' transmissions in an asynchronous SCMA system, e.g. user 1's third symbol overlaps with user 2's first.

Delay Estimation for Asynchronous SCMA 

Dylan Wheeler, Kansas State University 

We address a gap in the literature for SCMA by considering the case of asynchronous reception with unknown delays. We model the delay estimation problem as a sparse recovery problem whose solution can then be used with any decoding algorithm. 

Compressed Sensing with Nonconvex Penalties

Dhir Patel, The Ohio State University

Over the summer, we worked on extending recovery guarantees from the classical convex compressed sensing literature to a family of well-behaved nonconvex penalties. 

A drawing of two level sets and their intersection with a line. Both intersect the line at a sparse solution but one is convex while the other is not.
A black and white image of the character after an L_0 invariance attack. The number of pixels removed is counted by the L_0 norm.
A black and white image of a character from the KMNIST dataset represented by hashmarks.

Adversarial ML and Invariance Attacks

Sean Persaud, Northeastern University 

During this co-op, Sean gained familiarity with the fundamentals of machine learning, programming in Python, and using Google Collab. Projects including training and testing neural networks for classification, experimenting with clustering techniques and adversarial attacks, and working with BASNET for identifying and masking objects. Sean also worked on applying invariance attacks to KMNIST.

Adversarial Camouflage with Style Transfer

Darshana Jaint, Northeastern University

Over several months, Darshana gained greater familiarity with adversarial machine learning. In particular, we focused on adversarial camouflage, which uses style transfer to blend into the existing image. Darshana also worked on updating and running the Itti and Koch model of human saliency. 

A picture of a small dog laying on the grass in front of flowers.
The picture of the dog with the flowers in the background modified by the Adversarial Camouflage attack.

SU Directed Reading Program

The primary goal of the DRP is to broaden participation in mathematics, especially among members of underrepresented groups, by increasing individuals’ self-identity as mathematicians, by providing them with mentors close to their age, and by welcoming them into the broader mathematical family. Secondary objectives of the program are to extend the undergraduate curriculum and to provide mentoring experience to graduate students.

SU DRP Page


Past project: