Merge pull request #18 from RBhupi/main

more changes

src/pages/science/collaboration.md

@@ -2,15 +2,15 @@
 
 Clouds have long fascinated humans because of their complex and diverse nature. To gain a deeper understanding of these atmospheric phenomena, a multidisciplinary team of computer scientists, meteorologists, and machine learning experts from Northwestern-Argonne Institute of Science and Engineering (NAISE) collaborated on the National Science Foundation (NSF)-supported SAGE project. The project's goal was to develop new edge computing technologies that would allow scientists to collect and analyze large amounts of data from advanced sensors in real time.
 
-Inspired by compelling results after applying a self-supervised machine learning model (called DINO) to bird sound analysis, our team explored the possibility of utilizing similar techniques to study cloud images (referring to it as ClouDINO). Encouraging early results prompted us to seek further validation by leveraging data from the Atmospheric Radiation Measurement (ARM) User Facility at the Department of Energy's Southern Great Plains (SGP) atmospheric observatory.
+Inspired by compelling results after applying a self-supervised machine learning model (called DINO) to bird sound analysis, our team explored the possibility of utilizing similar techniques to study cloud images (referring to it as ClouDINO). Encouraging early results prompted us to seek further application of the model by leveraging data from the Atmospheric Radiation Measurement (ARM) User Facility at the Department of Energy's Southern Great Plains (SGP) atmospheric observatory and Argonne Testbed for Multiscale Observational Science (ATMOS) at the U.S. Department of Energy’s Argonne National Laboratory.
 
 
 ![Zones](imgs/Zones.png)
 
 **Fig:1** _The feature vectors trained with DINO in PCS space._
 
 
-We used a self-supervised learning model capable of autonomously extracting prominent features from ground-based sky camera images. Using a joint embedding architecture with Vision Transformers, our model learned to autonomously segment cloudy images and classify them based on their properties such as coverage, diurnal variation, and cloud base height. Notably, the self-supervised model even showed potential for semantic segmentation without the need for labeled data which we would like to explore further. The results were comparable to the other studies of our team lead by Bobby Jackson, an atmospheric scientist and instrument specialist at Argonne, who primarily utilized Doppler Lidar data. This made our results more robust due to intercomparison from the different perspectives. Seongha Park's previous work in solar energy prediction not only complemented our results but also greatly boosted our validation efforts.
+We used a self-supervised learning model capable of autonomously extracting prominent features from ground-based sky camera images. Using a joint embedding architecture with Vision Transformers, our model learned to autonomously segment cloudy images and classify them based on their properties such as coverage, diurnal variation, and cloud base height. Notably, the self-supervised model even showed potential for semantic segmentation without the need for labeled data which we would like to explore further. The results were comparable to the other studies of our team lead by Bobby Jackson, an atmospheric scientist and instrument specialist at Argonne, who primarily utilized Doppler Lidar data. Furthermore, Seongha Park's prior work in solar energy prediction, leveraging her expertise in computer science and information technology, significantly complemented our findings and greatly boosted our validation efforts with new a dataset. This made our results more robust due to intercomparison from the different perspectives.
 
 ![Clusters_Labelled](imgs/Clusters_Labelled.png)