World Class Research in Artificial Intelligence
Strategic AI research investments are crucial to EXIT83's mission. Staying at the technological forefront enables us to offer tailored solutions that help our clients thrive in a dynamic digital landscape—while making meaningful contributions to the creation of a smarter, more connected future for us all.
Dr. Oscar Correa, AI/ML Leader at EXIT83
PhD in Computer Science • Multiple published research papers on AI
Led by Oscar Correa, Ph. D., the EXIT83 research team works both independently and in collaboration with world-renowned academics. We present our findings at prestigious conferences like ICML and NeurIPS to share what we learn and continually strive to chart new territory for practical, real-world innovations.
Self-Supervised Learning for Identifying Maintenance Defects in Sewer Footage
We propose a novel application of self-supervised learning (SSL) for automating sewer defect detection using the DINO methodology. We evaluated our method on the Sewer-ML dataset, achieving strong performance with minimal labeled data (50.05 and 87.45 with only 10% of the data).
This research highlights the potential of SSL in specialized fields and sets a foundation for scalable, cost-effective solutions in urban infrastructure maintenance.
Presented at ICML in 2024 in Vienna, Austria.
ResNet-18 validation performance heatmaps across imbalance levels. The x-axis represents the imbalance levels in the validation set, while the y-axis indicates the method and the imbalance level used during training.Expanding the Frontiers of Anomaly Detection with Self-Supervised Learning
While SSL has shown immense promise in computer vision, its application to anomaly detection remains underexplored. Our work aims to bridge this gap by providing a general framework for using SSL in real-world anomaly detection.
Our collaboration with Professor Randall Balestriero at Brown University, a recognized leader in SSL for computer vision, further strengthens our approach. Together, we explore various SSL techniques through a comprehensive ablation study.
To be presented at the upcoming SSL Theory and Practice workshop at NeurIPS 2024 In Vancouver, Canada.
Ready to talk?
Schedule a discovery call today, or read about engagement process first to see how we work.