Lilly Thomas joins Development Seed as our latest machine learning engineer hire, and we couldn’t be more excited! She will challenge our team to re-think and further develop how we apply deep learning to do feature extraction on high-resolution imagery, and while she expands our technical chops, she also tips the gender balance to 75% female on our machine learning team!
Initially, Lilly will focus on applying deep reinforcement learning to assist humans in mapping roads from overhead imagery. This is a novel take on a popular problem (deriving road networks from imagery), so it will be interesting to see if, and how, this work might offer an alternative solution to common pixel-wise segmentation techniques.
Monitoring volatile land using data to track changes over time is an area Lilly is quite adept at professionally and personally. She worked with heterogeneous data formats at both Eagleview Technologies and OmniEarth, where we overlapped working together, to deploy and scale cloud-based machine learning pipelines and help provide insight into resource conservation measures. Together we worked on a project leveraging land cover classification from satellite imagery, in combination with regional weather and evapotranspiration data, to assist California’s public water agencies in targeting inefficient users within their areas of jurisdiction.
Lilly’s day job has also morphed into her hobby outside of work. As an avid surfer based in Los Angeles, she has spent time looking at coastal erosion and its impacts on daily surfing conditions to detect rip currents using multispectral satellite imagery with sediment-specific band ratios.