Kevin Kasa

I am a Visiting Researcher at ServiceNow Research, where I work on fundamental research towards improving the safety, security, and performance of LLM agentic systems.

I previously graduated with an MASc in AI from the University of Guelph and Vector Institute, where I was advised by Graham Taylor.

From 2023-2025, I was a member of the Canada's Chief Science Advisor's Youth Council.

I graduated with a B.Eng in Aerospace Engineering from the Toronto Metropolitan University (previously Ryerson University). Previously, I've worked on applying machine learning to healthcare with the Orthopaedic Biomechanics Lab at the Sunnybrook Research Institute. I also spent one year working on machine learning for Geointelligence applications at MDA. In the past I've worked on research projects including robotics, orbital dynamics, synthetic aperture radar (SAR) and remote sensing, as well as at companies like IBM and Deloitte.

In addition to machine learning research, I am also very interested in deep tech entrepreneurship, having previously worked on bringing autonomous high-payload drones to the skies.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

profile photo

Research

I am interested in topics on agentic systems, adaptaion, and AI safety / security.

project image

Adapting Prediction Sets to Distribution Shifts Without Labels


Kevin Kasa, Zhiyu Zhang, Heng Yang, Graham W. Taylor
Uncertainty in Artificial Intelligence (UAI) 2025, 2025
paper / code

Improving prediction set coverage on arbitrary and continous distribution shifts.
project image

Empirically Validating Conformal Prediction on Modern Vision Architectures Under Distribution Shift and Long-tailed Data


Kevin Kasa, Graham W. Taylor
ICML 2023 workshop on Structured Probabilistic Inference & Generative Modeling, 2023
paper / code

Large-scale evaluations demonstrating the challenges of utilizing conformal prediction in distribution shift or long-tailed data regimes.
project image

Multi-Modal Deep Learning for Assessing Surgeon Technical Skill


Kevin Kasa, David Burns, Mitchell G. Goldenberg, Omar Selim, Cari Whyne, Michael Hardisty
MDPI Sensors, 2022
paper

Multi-modal (kinematic + image) deep learning model for surgical skill assessment.
project image

Deep Learning for Vessel Detection and Identification from Spaceborn Optical Imagery


Giona Matasci, Jonathan Plante, Kevin Kasa, Payam Mousavi, Andrew Stewart, Andrew Macdonald, Anne Webster, Jennifer Busler
ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXIV ISPRS Congress, 2021
paper

Siamese networks for re-identifying vessels via spaceborn optical imagery. Developed an automated pipeline for collecting a large geospatial dataset (WorldView-2/3), processing, & detecting 200k+ vessels.
project image

A modular system for behaviour analytics from airborne full-motion video


Andrew Stewart, Fouad Faraj, Helen You, Jonathan Plante, Michael Lim, Kevin Kasa, Austin Beauchamp, Anne Webster, Leigh Martin-Boyd, Ken Wong, Alexander Braun, Andrew J. Macdonald
SPIE Defense + Commercial Sensing, 2021
paper

Object detection + tracking from UAS video.




Design and source code from Jon Barron's website