Alex Wong

Department of Computer Science
Yale University

alex [dot] wong [at] yale [dot] edu
Google Scholar
LinkedIn
@alexk_wong

I am an Assistant Professor at Yale University and previously a postdoctoral research scholar under the supervision of Stefano Soatto at the UCLA Vision Lab. My long term research goal is to realize embodied cognition in fully autonomous agents, where their understanding of our world evolves as they interact with the physical environment.

My current work largely focuses on learning to infer the 3D structure of the scene without any human supervision, i.e. unsupervised. This is done by designing predictive cross-modal validation criterions that measure the compatibility between the estimated scene and multi-sensory observations. These criterions incorporate priors and inductive biases that enable my systems to achieve state-of-the-art performance, while using a small computational and memory footprint. Additionally, I am the first to demonstrate adversarial perturbations and universal (across networks, datasets, domains) perturbations for deep stereo networks. By studying these perturbations and analyzing the failure modes of perception systems, I have developed training procedures and architectural designs to yield robust models.

I was co-advised by Professor Stefano Soatto (UCLA) and Professor Alan Yuille (JHU) and successfully defended my thesis ''Exploiting Regularities to Recover 3D Scene Geometry'' on November 21, 2019.

Teaching is a passion of mine. I have taught as a teaching assistant throughout my graduate career and recently as an adjunct faculty at Loyola Marymount University (LMU). Here are some excerpts from recent teaching evaluations on courses that I developed at LMU.

" Dr. Wong's lecture is awesome. Machine learning involves lots of math but he made it easy by explaining basic maths and then applies them to algorithm. I was new to ML, but he made me love the subject."
-- Anonymous teaching evaluation, LMU CMSI 535 - Machine learning, Fall 2020

" Theory and applications of machine learning. I really like the way the course was structure: after every new topic there was a coding exercise. Great course. The professor was always willing to help."
-- Anonymous teaching evaluation, LMU CMSI 535 - Machine learning, Fall 2020

" Really appreciated how Dr. Wong goes around after every class to make sure his students understand the material and know that they can reach out to him if they need help."
-- Anonymous teaching evaluation, LMU CMSI 371 - Computer graphics, Spring 2020

" I think he is a great teacher who is very knowledgeable and furthermore genuinely is invested in his students."
-- Anonymous teaching evaluation, LMU CMSI 371 - Computer graphics, Spring 2020

News:

07/08/2022: Two of our papers (Monitored Distillation for Positive Congruent Depth Completion and Not Just Streaks: Towards Ground Truth for Single Image Deraining ) have been accepted by ECCV 2022! This marks another milestone in my career as Monitored Distillation for Positive Congruent Depth Completion is my first senior-author paper without my advisors (Stefano Soatto and Alan Yuille) and the authorship is led by three undergraduate students (Tian Yu Liu, Parth Agrawal, and Allison Chen).

05/02/2022: Great news! I've accepted an offer to join Yale University as a tenure-track Assistant Professor of Computer Science!

03/08/2022: Our paper Stereoscopic Universal Perturbations across Different Architectures and Datasets has been accepted to CVPR 2022! I am particularly proud of this paper because all of the leading authors are undergraduate students whom I have been advising. This is also my first last author paper.

10/09/2021: We will be presenting our talk on Unsupervised Depth Completion with Calibrated Backprojection Layers during Session 10A (2pm-3pm PST, 5pm-6pm EST) and 10B (7am-8am PST, 10am-11am EST) of ICCV 2021. Feel free to drop by for a chat!

09/25/2021: We will be presenting our talk on Small Lesion Segmentation in Brain MRIs with Subpixel Embedding on 9/27 at 8:46 AM EST during the MICCAI Brain Lesion Workshop Session 2. If you are around, be sure to tune in!

09/13/2021: Our work Small Lesion Segmentation in Brain MRIs with Subpixel Embedding has been accepted as an oral paper at MICCAI Brain Lesion Workshop 2021! I want to recognize the two undergrads (Allison Chen and Yangchao Wu) who were amongst the lead authors of the paper. It is tough publishing papers as a graduate student and postdoc, so being able to have an oral paper as undergrads is an impressive feat!

07/30/2021: Our work Unsupervised Depth Completion with Calibrated Backprojection Layers has been accepted as an oral paper at ICCV 2021!

05/15/2021: I will be co-chairing the Adaptive System session (1:00 PM PST, June 1, 2021) at ICRA 2021! I will also be presenting our paper An Adaptive Framework for Learning Unsupervised Depth Completion during the session.

05/3/2021: We will be presenting our paper Learning Topology from Synthetic Data for Unsupervised Depth Completion at the Machine Learning II session (11:00 AM PST, June 1, 2021) at ICRA 2021!

02/28/2021: Two of our papers, Learning Topology from Synthetic Data for Unsupervised Depth Completion and An Adaptive Framework for Learning Unsupervised Depth Completion, have been accepted to ICRA 2021!

02/08/2021: Our paper An Adaptive Framework for Learning Unsupervised Depth Completion has been accepted to RA-L 2021!

02/06/2021: A huge thanks to everyone who came by to visit our poster Stereopagnosia: Fooling Stereo Networks with Adversarial Perturbations at AAAI 2021. The code is opensourced on github and can be accessed here (and below under Publications)!

01/21/2021: Our paper Learning Topology from Synthetic Data for Unsupervised Depth Completion has been accepted to RA-L 2021!

12/08/2020: We are excited to present Targeted Adversarial Perturbations for Monocular Depth Prediction at NeurIPS 2020, Poster session 3 (9 PM PST)!

12/02/2020: Our paper Stereopagnosia: Fooling Stereo Networks with Adversarial Perturbations has been accepted to AAAI 2021!

09/25/2020: Our paper Targeted Adversarial Perturbations for Monocular Depth Prediction has been accepted to NeurIPS 2020!

06/06/2020: We will be presenting Unsupervised Depth Completion from Visual Inertial Odometry at ICRA 2020!

Dataset:

Publication:

Preprint:

Technical Report:

Teaching:

Loyola Marymount University (LMU)
Academic Semester Course Number Course Title
Fall 2020 CMSI 535 Machine Learning
Spring 2020 CMSI 371 Computer Graphics
Fall 2019 CMSI 533 Data Science and Machine Learning
Spring 2019 CMSI 371 Computer Graphics
Fall 2018 CMSI 281 Data Structures
Spring 2018 CMSI 371 Computer Graphics

University of California, Los Angeles (UCLA)
Academic Quarter Course Number Course Title
Spring 2017 ENGR 110 Introduction to Technology Management and Economics
Winter 2017 ENGR 111 Introduction to Finance and Marketing for Engineers
Fall 2016 ENGR 111 Introduction to Finance and Marketing for Engineers
Spring 2016 ENGR 110 Introduction to Technology Management and Economics
Winter 2016 ENGR 111 Introduction to Finance and Marketing for Engineers
Fall 2015 ENGR 111 Introduction to Finance and Marketing for Engineers
Spring 2015 ENGR 110 Introduction to Technology Management and Economics
Winter 2015 ENGR 111 Introduction to Finance and Marketing for Engineers
Fall 2014 ENGR 110 Introduction to Technology Management and Economics
Spring 2014 MGMT 237G Computational Methods in Finance
Spring 2014 PIC 10B Data Structures
Fall 2013 PIC 10A Introduction to Programming
Spring 2013 PIC 10A Introduction to Programming
Winter 2013 PIC 10A Introduction to Programming