The Architecture of Impact.
I believe that the most impactful research is actionable. My lab focuses on releasing production-grade software, calibrated models, and high-fidelity datasets to empower the global AI community.
PICASO
Probabilistic Conceptual Spaces Sensemaking under Uncertainty
Funded by the Alan Turing Institute’s Defence & Security Programme, PICASO revolutionizes intelligence analysis by making uncertainty a "first-class citizen." Our PROCS Engine replaces brittle, static knowledge graphs with probabilistic clouds, enabling systems to mirror human cognitive reasoning and provide "honest confidence" in high-stakes decision-making.
O(1) time complexity reasoning matching standard embedding speeds.
Selective prediction models that know exactly when not to trust their outputs.
FLiP methodology reducing memory footprint by 90% for secure collaboration.
MEmbER
Max-Margin Based Embeddings for Entity Retrieval. A robust framework for learning entity representations in large-scale systems.
Source CodeCALM Checkpoint
Pre-trained weights for Culturally Self-Aware Language Models, adapted for multi-lingual nuanced environments.
Model WeightsUrban Lifestyles
The Microsoft Research Asia Urban Informatics dataset for modeling temporal mobility and human behavior.
Download Metadata