Friday, Oct. 24, 2025
10:20 – 11:10 a.m. in ETB 1020
Jarin Ritu
Ph.D. student, Department of Electrical & Computer Engineering
Texas A&M University
Title: “Texture-Informed Knowledge Distillation from Foundation Models for Efficient Deployment“
Abstract:
Large machine learning models achieve state-of-the-art performance but are often too computationally expensive for deployment in resource-constrained environments. To address this, researchers have turned to knowledge distillation (KD), a widely used method that transfers knowledge from large models into smaller, lightweight student models. However, conventional methods often fail to preserve fine-grained, domain-specific structures that are critical for robust generalization.
In this talk, we will introduce Statistical and Structural Audio Texture Knowledge Distillation (SSATKD), a framework that extends KD beyond prediction and feature-matching by embedding texture awareness into the transfer process. The framework incorporates modules to capture both structural patterns (e.g., periodicity, edges) and statistical properties (e.g., histograms, co-occurrence). By doing so, we enable compact student models to retain domain-relevant signal properties while remaining computationally efficient. We will present results showing that SSATKD improves performance over baseline KD methods in environmental sound tasks and demonstrate the framework’s adaptability to vision through segmentation of scanning electron microscopy images.
Biography:
Jarin Ritu is a third-year Ph.D. student in the Department of Electrical and Computer Engineering at Texas A&M University, working in the Advanced Vision and Learning Lab with Dr. Joshua Peeples. Her research lies at the intersection of machine learning and computer vision with applications in audio and image analysis, focusing on developing texture-informed knowledge distillation frameworks that transfer information from large foundation models to lightweight models for real-world deployment.
Her collaborative work with the Massachusetts Institute of Technology Lincoln Laboratory and Baylor University explores texture-informed AI methods to improve environmental sound analysis and quantify cellular morphology changes in microscopy images related to toxic exposures. She has presented her research at venues such as the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics.
Beyond research, Jarin serves as the Vice President of Media for the Electrical and Computer Engineering Graduate Student Association at Texas A&M University and was selected for the National Science Foundation-funded Software-Tailored Architectures for Quantum Co-Design Quantum Ideas Summer School at Duke University. She also mentors undergraduate students in research projects related to her field.
Jarin Ritu’s homepage: avll.engr.tamu.edu/people/jarin-ritu/