๐๐๐ฉ๐๐ซ ๐๐๐๐๐ฉ๐ญ๐๐ ๐๐ญ International Conference on Learning Representations (๐๐๐๐) ๐๐๐๐!
๐๐๐จ๐ฎ๐ญ ICLR
The International Conference on Learning Representations (ICLR) is a top-tier venue in Artificial Intelligence and learning systems. According to Google Scholar Metrics, ICLR ranks 8th overall across all disciplines and publication types (2nd in the AI category), placing it among the most prestigious and highest-impact publications, including Nature, The Lancet, The New England Journal of Medicine, Science, and CVPR. ICLR records a h5-index of 362 and h5-median of 652, reflecting very strong citation impact and wide research influence.
๐๐๐จ๐ฎ๐ญ the paper
This work introduces ๐๐๐๐๐-๐๐๐, a novel complex modulation-based framework to mitigate spectral attenuation, a common phenomenon that degrades INR performance. The work is grounded on a strong theoretical foundation based on Chebyshev polynomial approximations and harmonic distortion analysis.
This includes a thorough experimental analysis showing that COSMO-INR significantly outperforms the state of the art across numerous tasks, including image reconstruction, denoising, inpainting, neural radiance fields (NeRFs), and 3D object reconstruction.
Full paper can be accessed at: https://lnkd.in/gezcqYcw
Why it is so important? ๐๐ง๐ญ๐ข๐ซ๐๐ฅ๐ฒ ๐๐จ๐ฆ๐-๐๐ซ๐จ๐ฐ๐ง ๐๐๐ฌ๐๐๐ซ๐๐ก ๐๐ญ ๐๐จ๐
The achievement is especially meaningful because the research was led entirely by recent graduates and faculty from the Department of Electrical & Electronic Engineering – UOP, through the MARC (Multidisciplinary AI Research Centre – University of Peradeniya) INR team. From idea to implementation and writing, the work was developed locally, showcasing the strength of home-grown talent and mentorship at UoP in producing globally competitive AI research.
Team Members
Pandula Thennakoon, Avishka Ranasinghe, Mario De Silva, Buwaneka Epakanda
Supervisors
Roshan Godaliyadda, Mervyn Parakrama Ekanayake, Vijitha R. Herath
