About Me

I am currently a Research Resident at FAR AI, exploring critical challenges in AI safety and mechanistic interpretability, with a particular interest in uncovering how large language models (LLMs) persuade and influence people. Concurrently, I am pursuing my PhD as an NSERC CGS-D scholarship recipient at York University, where I am part of the CVIL Lab under the supervision of Dr. Kosta Derpanis. My doctoral research emphasizes mechanistic interpretability in multi-modal video understanding systems.

My journey in AI research includes impactful industry experiences, such as internships at Ubisoft La Forge, where I worked on generative modeling for character animations, and at Toyota Research Institute, contributing to interpretability research for video transformers within the machine learning team. Additionally, I hold a position as a faculty affiliate researcher at the Vector Institute and previously served as Lead Scientist in Residence at NextAI (2020–2022).

My academic path began with a Bachelor of Applied Science (B.A.Sc) in Applied Mathematics and Engineering, specialized in Mechanical Engineering from Queen’s University in Kingston, Ontario. Following graduation, I gained practical engineering experience at Morrison Hershfield, collaborating in multidisciplinary teams to design buildings, laboratories, and residential projects.

I then earned my Master’s degree at the Ryerson Vision Lab in August 2020, co-supervised by Dr. Neil Bruce and Dr. Kosta Derpanis. My thesis focused on evaluating the effectiveness of various modalities in recognizing human actions.

Outside of research, I enjoy maintaining an active lifestyle focused on health and fitness, engaging competitively in Super Smash Bros. Melee, birdwatching, practicing close-up magic, and immersing myself in progressive house music.

Project Highlights

Project 1 Image
PrePrint
Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment
We train SAEs to discover universal and unique concepts across different models.
Paper
Project 1 Image
PrePrint
Archetypal SAE: Adaptive and Stable Dictionary Learning for Concept Extraction in Large Vision Models
We design SAEs that solve the issue of stability across different training runs
Paper
Project 1 Image
Spotlight @ CVPR 2024
Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models
Unsupervised discovery of concepts and their interlayer connections.
Paper, project page, demo.
Project 2 Image
Spotlight @ CVPR 2024
Understanding Video Transformers via Universal Concept Discovery
We discover universal spatiotemporal concepts in video transformers.
Paper, project page.
Project 3 Image
CVPR 2022 and Spotlight @ XAI4CV Workshop
A Deeper Dive Into What Deep Spatiotemporal Networks Encode
We develop a new metric for quantifying static and dynamic information in deep spatiotemporal models.
Paper, project page.
Project 4 Image
ICCV 2021
Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNs
We show how spatial position information is encoded along the channel dimensions after pooling layers.
Paper, code.
Project 5 Image
ICLR 2021
Shape or Texture: Understanding Discriminative Features in CNNs
We develop a new metric for shape and texture information encoded in CNNs.
Paper, project page.
Project 6 Image
Oral @ BMVC 2020
Feature Binding with Category-Dependent MixUp for Semantic Segmentation and Adversarial Robustness
Source separation augmentation improves semantic segmentation and robustness.
Paper.

News

  • Gave an invited talk at David Bau’s Lab at Northeastern University!
  • Gave an invited talk at Thomas Serre’s Lab at Brown University!
  • Paper accepted to TPAMI! Quantifying and Learning Static vs. Dynamic Information in Deep Spatiotemporal Networks. Paper
  • TWO papers accepted as Highlights at CVPR 2024!
  • CVPR 2024 paper accepted as a Highlight, a result of my Internship at Toyota Research Institute - Understanding Video Transformers via Universal Concept Discovery. Paper and project page! We will also be presenting this work as a poster at the Causal and Object-Centric Representations for Robotics Workshop
  • CVPR 2024 paper accepted as a Highlight - Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models. Paper and project page. We will also be presenting this work as a poster at the CVPR Explainable AI for Computer Vision Workshop
  • CAIC 2024 long paper accepted - Multi-modal News Understanding with Professionally Labelled Videos (ReutersViLNews) . Paper.
  • Paper accepted to the International Journal of Computer Vision (IJCV) - Position, Padding and Predictions: A Deeper Look at Position Information in CNNs . Paper.
  • I have been awarded the NSERC CGS-D Scholarship with a total value of $105,000! (Accepted)
  • I have accepted an offer to do a research internship at Toyota Research Institute for the Summer of 2023 at the Palo Alto HQ office!
  • I gave a talk at Vector’s Endless Summer School program on Current Trends in Computer Vision and a CVPR 2022 Recap
  • Paper accepted to the International Journal of Computer Vision (IJCV) - SegMix: Co-occurrence Driven Mixup for Semantic Segmentation and Adversarial Robustness. Paper.
  • I presented a spolight presentation at the Explainable AI for Computer Vision Workshop at CVPR 2022. You can watch the recorded talk here.
  • Paper Accpted to CVPR 2022 - A Deeper Dive Into What Deep Spatiotemporal Networks Encode: Quantifying Static vs. Dynamic Information. Paper and Project Page.
  • Paper Accepted to ICCV 2021 - Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNs. Paper.
  • Paper Accepted to BMVC 2021 - Simpler Does It: Generating Semantic Labels with Objectness Guidance. Paper.
  • Paper Accepted to ICLR 2021 - Shape or Texture: Understanding Discriminative Features in CNNs. Paper.
  • Paper Accepted as an Oral to BMVC 2020 - Feature Binding with Category-Dependant MixUp for Semantic Segmentation and Adversarial Robustness. Paper.