Skip to main content

Ayushi Mehrotra

Ayushi Mehrotra

2025 Davidson Fellow
$25,000 Scholarship

Age: 18
Hometown: Irvine, CA

Technology: “H-Sets: Uncovering Feature Interactions in Image Classifiers using Hessian”

About Ayushi

Hi! My name is Ayushi Mehrotra, and I am an incoming freshman at the California Institute of Technology (Caltech). I am interested in anything and everything related to AI security to enable trustworthy machine learning models and have dipped my toes into social choice, mechanism design, and causality. Above all, I am drawn to areas that intersect the human value of trust and mathematics.

Outside of research, I have spent the past twelve years training in Bollywood dance. It has been a constant source of joy — from competing at national competitions such as Starpower to dancing at Garba during Navratri with friends and family. I never miss a chance to dance. At the same time, my academic interests continue to grow at the intersection of AI and security. In the future, I hope to launch a research-based startup focused on AI security, ensuring that theoretical work in this field is directly applied to real-world models and systems. My goal is to help bridge the gap between research and practice in safety-critical AI applications.

Skip testimonial carousel

"I am incredibly honored to be named a Davidson Fellow. To me, being a Davidson Fellow means I get to join an incredible community of bright students also chasing weird, hard questions with deep-seated curiosity."

Project Description

Understanding how computer vision models “see” and make decisions is crucial for building trust, especially in high-risk areas such as self-driving cars or medicine. Most current tools explain decisions by showing which individual pixels were important, but they often miss how combinations of pixels work together to influence a prediction. My research introduces H-Sets, a new method that uncovers these interactions between features. Compared with existing methods across four models and two datasets, H-Sets consistently identifies sparse, more meaningful image regions that better explain what the model is actually paying attention to.

Deeper Dive

My project, H-Sets: Uncovering Feature Interactions in Image Classifiers Using Hessian, focuses on improving how we understand the decisions made by AI models. Most current explanation methods highlight which individual pixels in an image are important for a model’s prediction, but they miss how combinations of pixels might work together to influence that decision. This missing piece—called feature interaction—is essential for understanding complex model behavior. With guidance from Professor Nidhi Rastogi and mentorship from Ph.D. student Dipkamal Bhusal, I developed a new method that can detect these interactions using a mathematical tool called the Hessian. Over several months of research, I designed and implemented H-Sets, a method that explains not only which parts of an image matter but also how those parts interact to shape the model’s output. This contributes to making AI systems more interpretable and trustworthy.

AI systems are increasingly being used in areas where decisions carry real-world consequences, such as diagnosing diseases or detecting objects in self-driving cars. In these cases, it is critical to understand not just what the model is focusing on, but why. H-Sets helps bridge that gap by revealing how different parts of the input interact to influence the model’s prediction. This added layer of understanding can improve confidence in AI systems, especially in settings where safety, fairness, and accountability matter. I hope my work contributes to building AI that people can understand and trust.

Q&A

What are the top three foreign countries you’d like to visit?

Argentina, Spain, and Japan.

What is your favorite hobby?

I love to dance, especially Bollywood. I've been dancing for the past twelve years, performing at national competitions and Diwali parties.

What is your favorite tradition or holiday?

Diwali! In my family, we wear new clothes, light firecrackers, and dance the night away.

Skip carousel

Click image to download high resolution files

In The News

IRVINE, Calif. — Ayushi Mehrotra, 18, of Irvine, has been awarded a $25,000 Davidson Fellows Scholarship for her technology project, H-Sets: Uncovering Feature Interactions in Image Classifiers Using Hessian. The Davidson Fellows Scholarship is one of the nation’s most prestigious honors for students 18 and younger. Mehrotra’s award is part of the program’s 25th anniversary year, which is granting a record $825,000 to 21 students nationwide.

Download the full press release here