Publications

Conference Papers

• Detecting Building Growth in High Flood Risk Regions Across African Cities
  ICML 2019 Workshop on Climate Change and AI

  I wrote this paper with the Planet Labs research team. It uses satellite images and machine learning to track how cities grow in areas with a high risk of floods. The goal is to help plan for climate change in fast-growing areas.

Patents

• Weakly Supervised Visual Anomaly Detection and Segmentation
  US Patent No: US10672115B2

  This patent describes a new way to train deep learning models when labels are few or noisy. We first built it to inspect products in industry by camera. Since then, it has influenced other uses in quality control and monitoring.

Technical Reports

• Multi-sensor Fusion for Change Detection in Satellite Imagery
  Internal Report | Planet Labs

  A detailed study of how to combine data from different satellite sensors (optical cameras, radar, and others) to make change detection models more accurate. It includes real case studies in environmental and infrastructure monitoring.

• Scalable Deep Learning Pipelines for Satellite Imagery
  Internal Report | Planet Labs

  Describes how to design and speed up large ML systems that process satellite data for the whole world. It covers system design, how to coordinate many models (orchestration), and how to tune performance for live use.