cv

Post-doctoral Researcher

Mila Quebec AI Institute & McGill University, Montreal, Quebec, Canada

• Conduct research in deep learning on domain adaptation and out-of-distribution problems.
• Write research papers for publication and present current work at conferences (2).
• Organize weekly meetings. Mentor graduate students (4). Assist with writing grant applications.

Post-doctoral Assistant Professor

University of Michigan, Ann Arbor, Michigan, US

• Research activities
  • Conducted research in numerical analysis, under the guidance of Dr. Selim Esedoglu, focusing on threshold dynamics algorithms which are ideal for large scale simulations of grain growth.
  • Wrote research articles for publication and presented work at 4 conferences and 3 seminars.
• Teaching activities
  • Taught undergraduate mathematics courses. Designed and delivered lectures, facilitated group work, and wrote homework assignments and exams.
  • Courses included multivariable and vector calculus, linear algebra, differential equations and numerical analysis.
  • Received a student Honored Instructor Nomination that recognizes the teaching efforts of instructors who had a positive impact on their experience.

A Reproducible and Realistic Evaluation of Partial Domain Adaptation Methods

Tiago Salvador and Kilian Fatras and Ioannis Mitliagkas and Adam M. Oberman

In NeurIPS Distribution Shifts Workshop 2022

• Code
• A benchmark study on the impact of model selection strategies in partial domain adaptation methods.

ImageNet-Cartoon and ImageNet-Drawing: two domain shift datasets for ImageNet

Tiago Salvador and Adam M. Oberman

In ICML Shift Happens Workshop 2022

• Project Page
• Code
• We propose two new datasets, using data augmentation, to measure model robustness to dataset shift.

FairCal: Fairness Calibration for Face Verification

Tiago Salvador, Stephanie Cairns, Vikram Voleti, Noah Marshall and Adam M. Oberman

In International Conference on Learning Representations 2022

• Project Page
• Code
• We remove bias from face recognition models leveraging unsupervised clustering to bypass the need for sensitive attribute (such as race, ethnicity, etc). We improve different fairness metrics while increasing accuracy and without any additional training.

Frustratingly Easy Uncertainty Estimation for Distribution Shift

Tiago Salvador, Vikram Voleti, Alexander Iannantuono and Adam M. Oberman

arXiv preprint 2021

• We propose two simple post‐hoc calibration methods that leverage data augmentation to improve calibration in the presence of distribution shift (unsupervised domain adaptation and domain generalization) for deep neural networks.

ImageNet-Cartoon and ImageNet-Drawing: two domain shift datasets for ImageNet

In ICML Shift Happens Workshop, July 2022

FairCal: Fairness Calibration for Face Verification

In International Conference on Learning Representations (ICLR), April 2022

Fairness Calibration for Face Verification

Montreal Machine Learning and Optimization (MTL MLOpt) Internal Meeting, May 2021

PhD in Mathematics

McGill University, Montreal, Canada

M.Sc in Mathematics and Applications

Instituto Superior Técnico, Universidade de Lisboa, Portugal

B.Sc in Applied Mathematics and Computation

Instituto Superior Técnico, Universidade de Lisboa, Portugal

Coursera Specializations

• Applied Data Science (Univ. of Michigan)
• Deep Learning (DeepLearning.AI)
• Reinforcement Learning (Univ. of Alberta)