2 months ago
Montréal, CanadaSenior / Staff+
Responsibilities
- Lead the design and development of biometric computer vision systems for face attributes, detection, quality assessment, and recognition.
- Architect, train, optimize, and productionize machine learning models using PyTorch, TensorFlow, and/or JAX.
- Own end-to-end ML pipelines from data ingestion and cleaning through training and deployment, including automated Airflow workflows.
- Curate balanced training datasets and generate synthetic data to address quality and diversity gaps.
- Optimize models for low-latency inference using quantization, distillation, TensorRT, and ONNX, and manage deployment on AWS.
- Conduct fairness analysis and benchmarking across datasets and operating conditions, including measuring and mitigating disparate impact.
- Mentor ML engineers, conduct code and design reviews, and establish technical best practices for the Computer Vision team.
Requirements
- At least 5 years of industry experience in machine learning, including at least 3 years dedicated to biometrics or face analysis.
- Deep expertise in computer vision, biometrics, and especially face recognition.
- Practical experience understanding, measuring, and mitigating algorithmic bias and disparate impact in computer vision.
- Expert proficiency in Python and machine learning or computer vision libraries such as Pillow, OpenCV, and PyTorch.
- Experience designing end-to-end ML pipelines and working with workflow orchestrators such as Airflow.
- Hands-on experience scaling training jobs on multi-GPU clusters and deploying services on AWS, including SageMaker, EC2, and EKS.
- Preferred experience includes research publications in CVPR, ICCV, ECCV, or FG related to face recognition, image quality assessment, or fairness.
- Preferred experience with vector databases such as Milvus or Faiss and approximate nearest neighbor search.
- Familiarity with privacy, security, and compliance in biometric systems is preferred.
- Preferred mobile or edge experience includes porting models using CoreML, LiteRT, or TFLite.
- Experience with GANs or diffusion models for generating synthetic training faces and strong communication skills are beneficial.
Categories
