Fatemeh N. Nokabadi

Machine Learning Scientist | Computer Vision, VLMs & Video Understanding

Email LinkedIn GitHub Scholar Publications
Fatemeh N. Nokabadi

About

I’m a Machine Learning Scientist with a Ph.D. in Electrical Engineering from Université Laval and Mila. I build and evaluate machine learning systems, bridging research and real-world applications.

My work spans vision-language models, document understanding, video object tracking, and model robustness. At Employment and Social Development Canada, I develop and adapt multimodal models for document intelligence, with a focus on improving accuracy, diagnosing failure modes, and making models reliable in production environments.

My doctoral research explored how learning-based vision systems behave over time and under adversarial conditions. I’ve published and presented research at venues including TMLR, NeurIPS, and CRV.

I’m interested in practical AI: understanding why models fail, developing better learning methods, and translating research into systems that work beyond the lab.

Projects

Multimodal Models for Document Understanding
  • Led model development for a domain-specific visual document-understanding system based on vision-language models (VLMs), covering model architecture, fine-tuning, training, and evaluation.
  • Fine-tuned open-weight models from the Qwen-VL family and TrOCR using Hugging Face Transformers, integrating them into an existing multi-model recognition pipeline and developing the associated training, validation, and testing workflows.
  • Developed data-type-specific prompting strategies to improve the reliability of VLM outputs on complex documents.
  • Adapted model development to strict privacy requirements and limited computational resources in an offline environment, supporting production deployment.
Robustness Evaluation of Vision Models
  • Developed frameworks to evaluate failure modes of modern vision models.
  • Exposed critical weaknesses under adversarial conditions.
  • Improved understanding of model reliability in safety-critical scenarios.
Reproducibility Study Webpage Paper Code Presentation

TrackPGD Webpage Paper Code Presentation

ABBG Paper Code Presentation

Transactions on Machine Learning Research 2024 · Proceedings of the 22nd Conference on Robots and Vision 2025 · Journal-to-conference track & AdvML Frontiers at NeurIPS2024

Representation Learning for Video Understanding
  • Framed single-object tracking as a reinforcement-learning episodic task with a Soft-Actor-Critic agent learning a motion policy from video frames.
  • Proposed the β-Multivariational Autoencoder (βMVAE) and a U-Net variant (βMVUNet) to learn an entangled Multivariate Gaussian prior (mean vector and covariance matrix) over the tracker's correlated motion parameters directly from frame patches.
  • Jointly trained the decoder to predict the target's binary segmentation mask, so a single network learns the motion distribution and the per-frame object mask used by the tracker.
Paper Code Presentation

WiML at ICML 2024

Visual Authentication System
  • Built models for fine-grained visual verification tasks.
  • Leveraged subtle visual cues for authenticity detection.
  • Achieved high accuracy in real-world scenarios.
Graph-Based Salient Object Detection
  • Developed graph-based saliency detection algorithms using local, global, and mixed feature representations.
  • Proposed multiple methods including random graph and global contrast graph approaches for salient object detection in images.
Paper 1 Paper 2 Paper 3

Multimedia Tools and Applications 2018 · Signal, Image and Video Processing 2018 · SPIS 2015

Real-Time Multi-Object Tracking for Moving Cameras
  • Researched and implemented real-time multi-object tracking algorithms for cluttered video frames, incorporating salient object detection to prioritize targets.
  • Designed and deployed a MATLAB and C++ tracking system using OpenCV for moving-camera video streams, improving tracking speed by 30%.

Academic Service