PhD Student
I focus on building robot policies that move us closer to artificial general intelligence. My work emphasizes visual generalization in online visual reinforcement learning. I am supervised by Glen Berseth.
Research Interests
My research combines reinforcement learning, robotics and computer vision. I aim to create learning systems that can operate in varied environments with minimal task-specific tuning. Key areas include:
- Artificial general intelligence and generalist robot policies
- Visual generalization in online visual reinforcement learning
- Using large pretrained vision models in RL pipelines
- Efficient methods for real-world deployment
Publications
SegDAC: Segmentation-Driven Actor-Critic for Visual Reinforcement Learning
SegDAC integrates large vision models with online visual reinforcement learning to produce object-centric scene representations. It improves visual generalization by up to twice compared to previous state-of-the-art online visual RL methods, without relying on human labels, data augmentation, or frame stacking.
Selected Projects
Canada Wildfire Prediction using Deep Learning
End-to-end system for large-scale wildfire forecasting across Canada at 250 m resolution. Handles the full pipeline from multi-source geospatial data collection and cleaning to model training and evaluation. Achieves strong test set performance, making it a practical tool for predicting future wildfire risk.
Aerial Semantic Segmentation
End-to-end deep learning pipeline for high-resolution segmentation of aerial imagery. Processes large-scale geospatial datasets, optimizes models for small-object detection, and delivers robust performance across diverse environments. Useful for mapping, infrastructure monitoring, and environmental analysis.
Cloud Cover Segmentation
Developed deep learning models for accurate detection of clouds and cloud shadows in satellite imagery. Includes full data preprocessing and model training pipeline to handle diverse atmospheric conditions. Enhances the quality of remote sensing datasets by removing noise, improving analysis for environmental and mapping applications.
Experience
PhD Student, Artificial Intelligence • 2025 to present
MSc in Artificial Intelligence • 2023 to 2025
2021 to 2023
2018 to 2021
Contact
Email: alexandre.brown@mila.quebec