Open-World Detection and Inference at Scale

This project develops machine learning methods for detecting and predicting diverse phenomena in large-scale real-world data under minimal supervision. We focus on open-world settings where training data is limited, noisy, and highly imbalanced—conditions typical of sociotechnical systems research.

Key methodological contributions include:

• Weakly supervised learning frameworks for content detection with limited labeled data
• Multi-task learning architectures that leverage shared representations across related detection tasks
• Imbalanced learning techniques for identifying rare but important content (e.g., hate speech, misinformation)
• Open-world classification methods that can detect novel content categories not seen during training
• Multimodal fusion approaches combining text, images, video, and network structure

These computational methods underpin our empirical studies of online platforms, enabling large-scale measurement and analysis of phenomena that would be impossible to manually annotate. Our frameworks are designed for robustness to the inherent noise and evolving nature of real-world social media data.