RISE-MICCAI Journal Club Session - January 24, 2026
Tuesday 20th January 2026
Join the next RISE-MICCAI Journal Club Session
Paper: Implicit neural representations for signal representation and medical image segmentation
Presenting author: Kushal Vyas, Rice University
Saturday, January 24, 2026 at 12:00 pm EST / 6:00 pm CET
Abstract:
Implicit neural representations (INRs) are new and emerging neural signal or data representations that typically represent a single signal or measurement by essentially overfitting the signal to an underlying multi-layer perceptron's parameters. However, INRs are sensitive to their underlying parameter initialization, which largely determines their convergence and the quality of the learned representation.
In this talk, Kushal Vyas will discuss how we can leverage common properties of natural images, such as underlying structures and low-frequency components, to learn efficient data-driven initialization schemes for implicit networks, allowing them to learn high-quality signal representations quickly. Building on the same premise, he will also present a joint, novel task-based initialization scheme for implicit networks, specifically for signal-translation tasks such as segmentation, known as MetaSeg, which received the 2025 MICCAI Best Paper Award. He will dive deeper into MetaSeg and how it performs on 2D and 3D Brain MRI segmentation.
Author bio:
Kushal Vyas is a PhD student at Rice University working on novel machine learning methods and data representations for medical and scientific imaging, alongside research in machine learning for large signals. His work includes implicit neural representations of signals, and he was recently awarded the MICCAI Best Paper Award for developing neural fields that jointly enable signal representation and segmentation. Prior to his PhD, he worked at Samsung Research America on machine learning and computational photography, and earned his master’s degree from Carnegie Mellon University.