Florida International University
South Florida Coastal Environmental Data and Modeling Center
Large Language Models
Large Language Models (LLMs) have rapidly transformed various sectors by enabling more intelligent and adaptable solutions through advanced natural language processing. This presentation delves into the capabilities of LLMs, focusing on their tuning and learning mechanisms, and highlights practical applications in real-world projects. We will explore how tuning LLMs enhances their adaptability to specific domains, and how continuous learning improves their problem-solving capabilities over time. The demonstration section will feature CLLMate, a platform that leverages LLMs to provide personalized learning experiences, showing how these models have been integrated to facilitate user interaction, adapt responses, and streamline knowledge-based tasks. This session aims to shed light on the potential of LLMs in enhancing both project outcomes and user engagement, providing insights into the evolving role of AI in various applications.
Remote Sensing and Data Analytics: Understanding Climate-Driven Natural Disasters
This presentation will discuss harnessing geospatial data science and remote sensing data for climate-related natural disasters. The need for fast processing and classification of massive remote sensing datasets for near-real-time monitoring of natural disasters, such as floods and landslides, highlights the importance of integrating deep learning methods into sensing approaches to create fast and accurate maps of affected areas. These maps can provide critical information to support emergency response planning and facilitate damage assessment in both spatial and temporal contexts.
Open Science Pool
What is the OSPool?
The OSPool is a source of computing capacity that is accessible to any researcher affiliated with a US academic institution. Capacity is allocated following a Fair-Share policy.
To harness the full capacity of the OSPool you will need to obtain an account via the OSG Portal.
How Can I Harness the OSPool Capacity?
Researchers can submit computational work to the OSPool via Access Points operated by the OSG, which serves researchers affiliated with projects at US-based academic, non-profit, and government institutions.
Namely, you can benefit from the OSPool Capacity if you are a
Researcher affiliated with a project at a US-based academic, government, or non-profit institution (via an OSG-Operated Access Point).
Researcher affiliated with such an institution or project that operates a local own access point.
Institutions or collaborations that would like to harness the capacity of the OSPool should contact support@osg-htc.org
What types of work run well on the OSPool?
For problems that can be run as numerous, self-contained jobs, the OSPool provides computing capacity that can transform the types of questions researchers are able to tackle (see the table below). A wide range of research problems and computational methods can be broken up or otherwise executed in this high-throughput computing (HTC ) approach, including:
image analysis (including MRI, GIS, etc.)
text-based analysis, including DNA read mapping and other bioinformatics
parameter sweeps
model optimization approaches, including Monte Carlo methods
machine learning and AI executed with multiple independent training tasks, different parameters, and/or data subsets
The OSPool is made up of mostly opportunistic capacity – contributing clusters may interrupt jobs at any time. Thus, the OSPool supports workloads of numerous jobs that individually complete or checkpoint within 20 hours.
Downscaling
Climate forecasting downscaling transforms global climate models with low-resolution forcasting data into high-resolution projections to better predict local weather patterns.
Project 1
Project2
Project3
Research