Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Alliance Grants

The University of British Columbia

$25,000.00

Assessing the Suitability of Datasets in Foundation Models Development: Foundational Models for Remote Sensing A Case Study

FY 2024-25

Summary
Foundation models, which are models trained on broad datasets with powerful generalization tasks such as large language models from ChatGPT, have revolutionized natural language processing and are widely deployed. There has been recent rise in computer vision foundation models in terms of multi-modal large language models (e.g., Flamingo and GPT-4v) and pixel-level image understanding models (e.g., SAM). In addition to the recent rise in building foundational models for remote sensing (e.g., SSL4EOL). Most of these foundation models rely on training with large-scale datasets, usually in a self supervised manner. However, training foundation models has significant environmental costs as reported in other works. To address this, we propose to answer a meta question when building such foundation models. We investigate whether we can assess the dataset suitability for self supervised learning and its impact on models generalization ability, without repeated and extensive training of these foundation models with emphasis on foundation models for remote sensing. Our proposal will reduce the environmental cost of building foundation models generally and has wide impact on the Computer Vision field, without being tied to a specific instance of a foundation model or task. We propose a suitability measure (s-measure) that is computed using two methods, one is an analytical approach that is designed to handle large-scale datasets, while the other is large language models based.
Program
Alliance Grants
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
595973-2024
Name nom
Siam, Mennatullah M
Department département
Computer Science
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
25000
Program id
ALLRP
Program name en
Alliance Grants
Program name fr
Subventions Alliance
Group en
RESEARCH PARTNERSHIPS
Group fr
PARTENARIATS DE RECHERCHE
Committee code
96
Committee name en
RPP Internal Decision Cttee
Committee name fr
Cté de décision interne - PPR
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2603
Research subject group en
Robotics
Research subject group fr
Robotique
Research subject en
Computer vision
Research subject fr
Vision artificielle
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Assessing the Suitability of Datasets in Foundation Models Development: Foundational Models for Remote Sensing A Case Study
Keywords
computer vision; datasets; deep learning; foundational models; remote sensing
Application summary
Foundation models, which are models trained on broad datasets with powerful generalization tasks such as large language models from ChatGPT, have revolutionized natural language processing and are widely deployed. There has been recent rise in computer vision foundation models in terms of multi-modal large language models (e.g., Flamingo and GPT-4v) and pixel-level image understanding models (e.g., SAM). In addition to the recent rise in building foundational models for remote sensing (e.g., SSL4EOL). Most of these foundation models rely on training with large-scale datasets, usually in a self supervised manner. However, training foundation models has significant environmental costs as reported in other works. To address this, we propose to answer a meta question when building such foundation models. We investigate whether we can assess the dataset suitability for self supervised learning and its impact on models generalization ability, without repeated and extensive training of these foundation models with emphasis on foundation models for remote sensing. Our proposal will reduce the environmental cost of building foundation models generally and has wide impact on the Computer Vision field, without being tied to a specific instance of a foundation model or task. We propose a suitability measure (s-measure) that is computed using two methods, one is an analytical approach that is designed to handle large-scale datasets, while the other is large language models based.