Tiago Salvador

I’m a post-doctoral researcher at McGill University and Mila working at the intersection of applied math, deep learning and machine learning. My main focus is in computer vision, developing algorithms for domain adaptation and out-of-distribution uncertainty problems.

Previously, I was a post-doctoral Assistant Professor at the University of Michigan in applied math, working on threshold dynamics algorithms which are ideal for large scale simulations of grain growth. I did my PhD in applied mathematics at McGill University supervised by Professor Adam Oberman. My research focused on developing numerical methods for partial differential equations.

news

Oct 20, 2022	Short paper accepted at the NeurIPS2022 Distribution Shifts Workshop “A Reproducible and Realistic Evaluation of Partial Domain Adaptation Methods” [Open Review].
Oct 3, 2022	Released “A Reproducible and Realistic Evaluation of Partial Domain Adaptation Methods” on [arXiv].
Jul 18, 2022	Attended ICML 2022 virtually.
Jul 8, 2022	Short paper accepted at @ICML2022 Shift Happens Workshop “ImageNet-Cartoon and ImageNet-Drawing: two domain shift datasets for ImageNet” [Open Review].
Apr 25, 2022	Attended ICLR 2022 virtually.
Mar 30, 2022	Released the GitHub repo Calibration Baselines, a starting point for researchers into post-hoc calibration.

selected publications

NeurIPS Workshop

A Reproducible and Realistic Evaluation of Partial Domain Adaptation Methods

Tiago Salvador, Kilian Fatras, Ioannis Mitliagkas, and Adam Oberman

In NeurIPS Distribution Shifts Workshop, 2022

abstract arXiv html pdf code

Unsupervised Domain Adaptation (UDA) aims at classifying unlabeled target images leveraging source labeled ones. In this work, we consider the Partial Domain Adaptation (PDA) variant, where we have extra source classes not present in the target domain. Most successful algorithms use model selection strategies that rely on target labels to find the best hyper-parameters and/or models along training. However, these strategies violate the main assumption in PDA: only unlabeled target domain samples are available. Moreover, there are also inconsistencies in the experimental settings - architecture, hyper-parameter tuning, number of runs - yielding unfair comparisons. The main goal of this work is to provide a realistic evaluation of PDA methods with the different model selection strategies under a consistent evaluation protocol. We evaluate 7 representative PDA algorithms on 2 different real-world datasets using 7 different model selection strategies. Our two main findings are: (i) without target labels for model selection, the accuracy of the methods decreases up to 30 percentage points; (ii) only one method and model selection pair performs well on both datasets. Experiments were performed with our PyTorch framework, BenchmarkPDA, which we open source.
ICML Workshop

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

Tiago Salvador, and Adam M. Oberman

In ICML Shift Happens Workshop, 2022

abstract html pdf code Slides

Benchmarking the robustness to distribution shifts traditionally relies on dataset collection which is typically laborious and expensive, in particular for datasets with a large number of classes like ImageNet. An exception to this procedure is ImageNet-C (Hendrycks & Dietterich, 2019), a dataset created by applying common real-world corruptions at different levels of intensity to the (clean) ImageNet images. Inspired by this work, we introduce ImageNet-Cartoon and ImageNet-Drawing, two datasets constructed by converting ImageNet images into cartoons and colored pencil drawings, using a GAN framework (Wang & Yu, 2020) and simple image processing (Lu et al., 2012), respectively.
ICLR

FairCal: Fairness Calibration for Face Verification

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

In International Conference on Learning Representations, 2022

abstract arXiv html pdf code Poster Slides

Despite being widely used, face recognition models suffer from bias: the probability of a false positive (incorrect face match) strongly depends on sensitive attributes such as the ethnicity of the face. As a result, these models can disproportionately and negatively impact minority groups, particularly when used by law enforcement. The majority of bias reduction methods have several drawbacks: they use an end-to-end retraining approach, may not be feasible due to privacy issues, and often reduce accuracy. An alternative approach is post-processing methods that build fairer decision classifiers using the features of pre-trained models, thus avoiding the cost of retraining. However, they still have drawbacks: they reduce accuracy (AGENDA, PASS, FTC), or require retuning for different false positive rates (FSN). In this work, we introduce the Fairness Calibration (FairCal) method, a post-training approach that simultaneously: (i) increases model accuracy (improving the state-of-the-art), (ii) produces fairly-calibrated probabilities, (iii) significantly reduces the gap in the false positive rates, (iv) does not require knowledge of the sensitive attribute, and (v) does not require retraining, training an additional model, or retuning. We apply it to the task of Face Verification, and obtain state-of-the-art results with all the above advantages.