Teresa Yeo

I am a PhD student at the VILAB, EPFL supervised by Amir Zamir and Pierre Dillenbourg. My research is on the intersection of computer vision and machine learning and in particular, making models robust and adaptive! I am also interested in the theory and methods for algorithmic teaching and inverse reinforcement learning.

My undergrad degree was at the National University of Singapore in the Mathematics department. I went on to do my Master in Financial Engineering at Columbia University and became a quant developing on systematic investment strategies for equity portfolios. I then returned to school to do a Master in Machine Learning at Cambridge University.

Towards robust and adaptive systems. Click to see the relevant projects. Full publication list below.

Controlled Training Data Generation with Diffusion Models

Yeo, T.*, A. Atanov*, A. Alekseev, H. Benoit, R. Ray, A. Zamir

A method to control text-to-image generative models to produce training data useful for supervised learning.

Website

4M: Massively Multimodal Masked Modelling

D. Mizrahi*, R. Bachmann*, O. F. Kar, Yeo, T., M. Gao, A. Dehghan, A. Zamir

NeurIPS 2023 (Spotlight)

A scalable multimodal training scheme.

Paper Website

Fast Adaptation of Neural Networks Using Test-Time Feedback

Yeo, T., Kar, O., Sodagar, Z., Zamir, A.

ICCV 2023

Learning to use test-time feedback to adapt neural networks.

Paper Website

Task Discovery: Finding the Tasks that Neural Networks Generalize on

Atanov, A., Filatov, A., Yeo, T., Sohmshetty, A., Zamir, A.

NeurIPS 2022

Instead of fixing the task and searching in the model space, we fix the model and search in the task space.

Paper Website Code

3D Common Corruptions and Data Augmentation

Kar, O., Yeo, T., Atanov, A., Zamir, A.

CVPR 2022 (Oral)

Using 3D information to generate real-world corruptions.

Paper Website Code

Robustness via Cross-Domain Ensembles

Yeo, T.*, Kar, O.*, Zamir, A.

ICCV 2021 (Oral)

Ensembling via a diverse set of middle domains.

Paper Website Code

Robust Learning Through Cross-Task Consistency

Zamir, A.*, Sax, A.*, Yeo, T., Kar, O., Cheerla, N., Suri, R., Cao, Z., Malik, J., Guibas, L.

Arxiv 2020, ECCV 2020 (demo), CVPR 2020 (Best paper award nominee)

Given a dictionary of tasks, augment the learning objective with constraints for cross-task consistency.

Paper Website Code

Iterative Classroom Teaching

Yeo, T., Kamalaruban, P., Singla, A., Merchant, A., Asselborn, T., Faucon, L., Dillenbourg, P., Cevher, V

AAAI 2019 (Oral), AMLD 2019 (Invited talk)

Finding the optimal sequence of examples to train a class of linear algorithms with applications to teaching to write and to classify.

Paper

Assisted Inverse Reinforcement Learning

Kamalaruban P., Devidze R., Yeo, T., Mittal T., Cevher V., Singla A.

NIPS 2018 Learning by Instruction Workshop

Finding the optimal sequence of examples to train an Inverse Reinforcement Learning agent.

Paper

2017-Present
École Polytechnique Fédérale de Lausanne
Computer and Communication Sciences

PhD in Machine Learning and Computer Vision
2015-2016
Cambridge University,
Information Engineering Division

MPhil in Machine Learning and Machine Intelligence
Thesis: Bayesian Optimization for Natural Language Processing
2011-2012
Columbia Univerity,
Industrial Engineering and Operations Research Department

MSc in Financial Engineering
2007-2011
National Univeristy of Singapore,
Mathematics Department

BSc (Hons) in Quantitative Finance
Thesis: High Order Numerical Methods for Solving PDE with Optimal Stopping Times

École Polytechnique Fédérale de Lausanne

Teaching Assistant
Fall 2021: CS503 Visual Intelligence: Machines and Minds
Spring 2018, 2019, 2020: EE559 Deep Learning
Fall 2019: CS433 Machine Learning
2016-2017, Shift Technology (Singapore)

Data Scientist
Designed and impelmented models for automated fraud detection.
2012-2015, UBS Global Asset Management (New York, London)

Quantitative Research, Structured Alpha/Beta
Researched on systematic strategies for equity portfolios.

first.lastname@epfl.ch

Somewhere in INJ, EPFL