CV

Eduardo Dadalto, PhD in ML

Dedicated to making AI a technology that benefits everyone.

Contact Information

• Full Name: Eduardo Dadalto Câmara Gomes
• Email: edadaltocg[at]gmail.com
• Location: Paris, Île-de-France, France
• Website: edadaltocg.github.io
• GitHub: github.com/edadaltocg
• LinkedIn: linkedin.com/in/edadaltocg
• PDF Version: cv.pdf

Research Interests: deep learning, computer vision, NLP, human-centric AI & robustness.

Education

CentraleSupélec, L2S, Université Paris-Saclay, CNRS & IBM Research

PhD in Machine Learning, Computer Science Department

Paris, France | Nov. 2020 - Mar. 2024

• Thesis: Improving artificial intelligence reliability through out-of-distribution and misclassification detection.
• Keywords: Safety in AI, Out-of-Distribution Detection, Uncertainty Estimation, Computer Vision, NLP.
• Advisors: Pablo Piantanida and Florence Alberge.
• Committee: Nicolas Vayatis, Yann Chevaleyre, Yves Grandvalet, and Nicolas Thome.

Institut de Mathématiques de Toulouse, Université Toulouse III

MSc in Applied Mathematics

Toulouse, France | Sep. 2019 - Oct. 2020

• Dissertation: "Training Binarized Deep Neural Networks" supervised by Franck Mamalet and François Malgouyres.

ISAE-SUPAERO

Diplôme d'Ingénieur, Major in Data & Decision Sciences

Toulouse, France | Sep. 2018 - Apr. 2020

• Selected Coursework: Supervised, Unsupervised, Deep, and Reinforcement Learning with Emmanuel Rachelson.
• Double Degree with a full tuition fee waiver scholarship.

Instituto Tecnológico de Aeronáutica (ITA)

BE in Aerospace Engineering

São José dos Campos, Brazil | Feb. 2015 - Jul. 2018

• Dissertation: "Machine Learning Applied to Communication Channels." Advisor: Meryem Bennamar.

Experience

Future Frame

Co-founder

San Francisco, CA, USA & Paris, France | Apr. 2024 - Nov. 2024

• Adapted transformer models to scale on structured data, eliminating the need for feature engineering.
• Released an API platform for data scientists to fine-tune and run inference with our model (B2B SaaS).
• Raised a pre-seed fund from Transpose VC and Entrepreneur First (EF).
• Acquired a deep understanding of aligning product development with customer needs.

Mila - Quebec AI Institute & International Laboratory on Learning Systems

Graduate Research Intern

Montreal, QC, Canada | Oct. 2022 - Jan. 2023

• Developed misclassification detection algorithms for medical image segmentation to improve diagnostics.
• Advised by Pablo Piantanida and Jose Dolz.

IRT Saint Exupéry, DEEL Team & Airbus

Undergraduate Research Intern

Toulouse, France | Apr. 2020 - Oct. 2020

• Trained neural networks with binary weights for embedded computer vision.
• Supervised by Franck Mamalet and co-supervised by François Malgouyres and Adrien Gauffriau (Airbus).

Publications

For an up-to-date list of publications, please visit my Google Scholar page.

1. Eduardo Dadalto, Marco Romanelli, Georg Pichler, Pablo Piantanida. A Data-Driven Measure of Relative Uncertainty for Misclassification Detection. ICLR2024. ARXIV PDF CODE

2. Eduardo Dadalto, Florence Alberge, Pierre Duhamel, Pablo Piantanida. Combine and Conquer: A Meta-Analysis on Data Shift and Out-of-Distribution Detection. TMLR2024. ARXIV PDF CODE

3. Maxime Darrin, Guillaume Staerman, Eduardo Dadalto, Jackie CK Cheung, Pablo Piantanida, Pierre Colombo. Unsupervised layer-wise score aggregation for textual ood detection. AAAI2023. ARXIV PDF

4. Pierre Colombo, Eduardo Dadalto, Guillaume Staerman, Nathan Noiry, Pablo Piantanida. Beyond Mahalanobis Distance for Textual OOD Detection. NeurIPS 2022. ARXIV PDF

5. Eduardo Dadalto, Florence Alberge, Pierre Duhamel, Pablo Piantanida. Igeood: An Information Geometry Approach to Out-of-Distribution Detection. ICLR 2022. ARXIV PDF CODE

6. Meryem Bennamar, Eduardo Dadalto, Pablo Piantanida. CSI-aided Robust Neural-based Decoders. IEEE ISTC, 2021. PAGE PDF

Preprints

1. Eduardo Dadalto, Marco Romanelli. Optimal Zero-shot Regret Minimization for Selective Classification With Out-of-Distribution Detection. 2024.

2. Eduardo Dadalto, Marco Romanelli, Federica Granese, Siddharth Garg, Pablo Piantanida. Trusting the Untrustworthy: A Cautionary Tale on the Pitfalls of Training-based Rejection Option. 2023.

3. Eduardo Dadalto, Pierre Colombo, Guillaume Staerman, Nathan Noiry, Pablo Piantanida. A Functional Perspective on Multi-Layer Out-of-Distribution Detection. 2023.

Workshops

1. Eduardo Dadalto, Marco Romanelli, Georg Pichler, Pablo Piantanida. A Data-Driven Measure of Relative Uncertainty for Misclassification Detection. NeurIPS Workshop on Mathematics of Modern Machine Learning. 2023.

2. Eduardo Dadalto, Florence Alberge, Pierre Duhamel, Pablo Piantanida. Igeood: An Information Geometry Approach to Out-of-Distribution Detection. NeurIPS Workshop on Distribution shifts: connecting methods and applications (DistShift). 2021.

Teaching

CentraleSupélec, Université Paris-Saclay

Teaching Assistant (TA) for the course Introduction to Deep Learning (3SQ2050)

Gif-sur-Yvette, France | 2021-2023

Awards and Scholarships

Nordic Probabilistic AI Summer School (ProbAI)

Norwegian University of Science and Technology

Trondheim, Norway | June 2023

• Sponsored by DeepMind. Coursework on probabilistic modeling, variational inference, and generative models.

Globalink MITACS Scholarship

Awarded for a winning project in a four-month full mobility scholarship program

Montreal, QC, Canada | Oct. 2022 - Feb. 2023

• Partnership between Mila, ÉTS, McGill, Inria, CNRS, and Université Paris-Saclay.

BCG Gamma Data Science Hackathon

Organized by BCG Gamma and Instituto Tecnológico de Aeronáutica

São José dos Campos, Brazil | February 2020

• Fine-tuned a YOLOv3 model for object detection on infrared drone footage.
• Received the Best Presentation Award.

Services & Volunteering

Reviewer: Neural Information Processing Systems (NeurIPS 2022, 2023), Computer Vision and Pattern Recognition (CVPR 2023), International Conference on Learning Representations (ICLR 2024), Conference on Artificial Intelligence (AAAI 2023).

Teaching: Volunteered as a high school math teacher at a Brazilian public school (2011).

Research Grants

Compute research grant (AD011012803R) at Jean-Zay a French HPC/AI cluster. | 2021 - 2024

PhD funding from the PSPC AIDA project (2019-PSPC-09) granted by BPI-France. | 2020 - 2024

Open Source

Detectors

May 2023 Open source package to accelerate research on out-of-distribution (OOD) detection for computer vision applications.

Todd / ToddBenchmark

February 2023 Open source package to accelerate research on out-of-distribution (OOD) detection for textual applications.

Skills & Tools

Languages: English (fluent), French (fluent), Portuguese (native).

Programming: Python (PyTorch, Numpy, Tensorflow/Keras, Scikit-Learn, Transformers, etc.), distributed computing (Slurm), Go, C, CUDA (basics), JavaScript, HTML, SQL, LaTeX, git, Linux, Docker, cloud (Terraform IaC, AWS, GCP), etc.