SDS 533: Fusion Energy, Cancer Proteomics, and Massive-Scale Machine Vision — with Dr. Brett Tully

AI Systems

Dr. Brett Tully, Director of AI Output Systems at Nearmap, explains the intricate structure of AI systems within the company. His team focuses on transforming deep learning models into actionable insights from vast amounts of aerial imagery. This involves a complex process of labeling and semantic understanding, resulting in over 250 attributes representing the urban physical world 1. Tully describes the process as "painting pixels with meaning," where each pixel in an image is assigned a confidence level for specific attributes 2.

   

Imagery Tech

Nearmap, a leader in location intelligence, revolutionizes aerial imagery by capturing and processing data at unprecedented scales. The company employs advanced camera systems and machine vision algorithms to create detailed 3D representations of urban environments 3. Tully highlights the use of Kubeflow and Kafka to manage dynamic pipelines, ensuring efficient processing of large-scale data while minimizing costs 4. This innovative approach allows Nearmap to offer subscription-based access to its comprehensive imagery database.

   

MLOps

Tully redefines MLOps at Nearmap, emphasizing the importance of application architecture over traditional model deployment. His team focuses on creating a "pipeline of pipelines" to handle the complexities of geospatial intelligence, where running inference is just a small part of the process 5. He notes that "everything breaks at scale," highlighting the challenges of processing vast amounts of data and the necessity for robust systems 6. This approach ensures that Nearmap can efficiently manage its extensive data operations.