Recording Plan: 

Curated Track: 

, 

Call For Participation
This session aims to advance deep learning applications for multi-source data analysis. We invite you to join us in pushing the boundaries of deep learning and developing impactful solutions to real-world challenges. The session will feature presentations on a range of topics, including but not limited to:

* Multi-Modal Data Fusion and Integration: Methods for combining different data modalities (e.g., imagery, text, sensor data) to create cohesive geospatial and environmental insights, addressing challenges in alignment, normalization, and scaling.
* Time Series Analysis for Multi-Temporal Data: Approaches to analyze time-sequenced data from various sources, capturing trends, changes, and anomalies over time, and the challenges of data synchronization and temporal alignment.
* Multimodal Deep Learning for Geospatial Applications: Applications of deep learning to integrate image, text, and sensor data for geospatial modeling, with a focus on handling and learning from heterogeneous data sources in real-world environments.
* Handling Data Sparsity and Imbalance in Multi-Source Data: Approaches for addressing issues of missing or imbalanced data in multi-source datasets, and the role of deep learning in learning from sparse or incomplete information.
* Explainability and Interpretability in Multi-Modal Deep Learning: Addressing the challenge of making deep learning models more interpretable in the context of multi-source data, focusing on developing tools and methods for better understanding model predictions and decisions.

Description:

In today’s data-rich era, researchers and practitioners encounter both unprecedented opportunities and challenges in deciphering complex environments through multi-source data analysis. The influx of data from remote sensing images, point clouds, social media imagery, and geospatial textual information offers a wealth of knowledge crucial for urban studies, environmental monitoring, and human-centered applications. The heterogeneity of these data sources also brings about unique complexities in terms of integration and processing.

Recent advancements in computer vision (CV), large language models (LLMs), and multimodal deep learning have enabled novel methods to bridge diverse data modalities and extract valuable insights from images, text, and sensor outputs. Techniques such as multimodal fusion, semantic segmentation, object detection, time series analysis, and natural language processing are breaking new ground in transforming multi-source data into actionable intelligence.

This session aims to advance the application of deep learning techniques for multi-source data processing. We seek to explore methods that address the distinctive characteristics of various data types and offer practical solutions for their analysis. We invite researchers to share their work, showcasing innovative approaches to multi-source data integration and analysis. Additionally, we aim to engage potential new users interested in these methodologies, fostering deeper multidisciplinary interactions on this topic.

Non-Presenting Participants