PhD on Multimodal Graphs for Biodiversity Intactness Understading

I looking for a high motivated PhD student to work on Multimodal Graphs for diversity. See more below.

Funding: DLA AI-Intervene, self-funding
Project Title: Multimodal Graphs for Biodiversity Intactness Index
Main Supervisor: Alan Guedes
Co-supervisor: Researcher from NHM Future Lab

The Biodiversity Intactness Index (BII) quantifies the impact of human activities on biodiversity by assessing changes in species abundance across various ecosystems. The Natural History Museum dataset PREDICTS details encompasses species abundance data, land-use types, and environmental changes during time. Futhermore, there is a growing of multimodal biodiversity data, including scientific texts, field images, videos, and environmental reports. However, these valuable data sources often remain siloed, making it challenging to derive interconnected insights and develop a comprehensive understanding of their relationships. To address this challenge, MM-BioGraph focuses on integrating multimodal data into graph-based structures to represent BII relationships. Central to this project is the use novel methods in graph neural networks (GNNs) and large language models (LLMs). GNNs enable learning a graph-structure of BII, allowing the project to make also predictions about unobserved links or even hypothesize new biodiversity phenomenas. The leverage LLMs to create GNNs models that go beyond traditional biodiversity analyses. These graphs can then be analysed to fill gaps in current BII knowledge and forecast potential changes driven by environmental factors. The proposed thesis will include the goals:

The project aims to integrate diverse data into graph-based structures to represent biodiversity relationships and predict unobserved links. The proposed thesis will include the following goals:

BII Multimodal Dataset Generation: Collaborate with NHM specialists to develop a comprehensive multimodal dataset (textual, visual, ecological) to support robust graph representation.
GNN Baseline Development: Establish a Graph Neural Network (GNN) baseline model for initial analysis of biodiversity relationships using the generated dataset.
LLM-Assisted GNN: Build upon the GNN baseline by applying Large Language Models (LLMs) alongside visual data. This approach will iteratively refine predictions and achieve more accurate and comprehensive insights into complex biodiversity relationships. It can also leveraging Chains of Thought (CoT) reasoning.

References

Datasets:

biodiversity graphs