Benjamin Thérien Email & Phone Number

Canada

Montreal, Quebec, Canada

Montreal, Quebec, Canada

Co-Advised by Irina Rish and Eugene Belilovsky• Collaborated as part of the U.S. DOE’s INCITE grant for "Scalable Foundation Models for TransferableGeneralist AI”.• Wrote a script to convert pre-trained large language models (LLMs) between transformers and GPT-NeoX. • Debugged and ran large-scale (9B param. models) experiments investigating the Continual Pre-training of LLMs on the SUMMIT supercomputer.• Optimized 3D parallelism model configuration to maximize GPU utilization (TFLOPs) for 410M, 3B, and 9B LLMs

New York, New York, United States

Researching pre-training of MoE LLMs.

Waterloo, Ontario, Canada

During my masters at the University of Waterloo, under the supervision of Krzysztof Czarnecki, my research on object reidentification using point clouds not only resulted in a significant publication at WACV 2024 but also in the crafting of a first-of-its-kind paired object re-identification dataset of LiDAR scans and images. I led the project through its entire lifecycle, from ideation to implementation, analysis, and writing. This process was pivotal in my growth as a researcher and machine learning engineer, enhancing my ability to lead a research project from start to completion and providing meaningful experience in creating efficient and scalable data processing tools. Beyond the work done for my thesis, I also conducted research with Dr. Hongyang Zhang and my colleague Luke Rowe on the difficulty of composite adversarial examples, contributed to a research publication "Out-of-distribution detection for lidar-based 3d object detection" with members of the WISE Lab, and was the teaching assistant for two courses: Neural Networks and Machine Learning.

Waterloo, Ontario, Canada

Instructor: Jeff Orchard

Waterloo, Ontario, Canada

Instructor: Yaoliang Yu

Montreal, Quebec, Canada

During my NSERC-funded internship at Mila, under the supervision of Dr. Eugene Belilovsky and Dr. Guy Wolf I jointly worked towards researching and benchmarking Parametric Scattering Networks and a multi-disciplinary project to classify the severity of chronic liver disease (CLD) on B-mode ultrasound (US) images. Both projects culminated in different research papers. Subsequently, I had the pleasure of presenting Parametric Scattering Networks at CVPR2022 in person as we received an oral spotlight. My hands-on experience in building software for research experimentation was instrumental in pushing both projects forward, as I was able to drive greater experimental organization and efficiency through these skills. Beyond my skills as an engineer, I proposed a number of new research experiments which ended up in our final paper —a testament to my evolving expertise as a researcher.Notable Milestones:- DeepMath 2021 Abstract- CVPR 2022 (Oral) Co-first author: Parametric Scattering Networks- Radiology 2023 co-author: Comparison of Radiologists and Deep Learningfor US Grading of Hepatic Steatosis

Montreal, Quebec, Canada

Under the mentorship of Dr. Sabine Bergler, I began my first research-focused internship at the CLaC Lab, making a pivot in my career trajectory. During my time at the lab, I spearheaded the development of an NLP pipeline, advancing it through iterative experimentation, which culminated in an 8th-place finish at MeasEval (NLP measurement evaluation competition) and a first-authored system description paper. This experience was instrumental in refining my technical skills, particularly by improving my familiarity with deep learning frameworks, and in cultivating my ability to propose and integrate new research ideas.

Montreal, Quebec, Canada

Instructor: Sabine Bergler. Responsibilities: giving tutorials (105 minutes/week) and grading.

Montreal, Canada Area

While at Morgan Stanley, I led the automation of Operational Acceptance Tests (OATs), creating a program that allows non-coders to manage these tests. By designing a robust data model and a user-friendly command-line interface, I deepened my skills in maximizing modularity, minimizing assumptions, and ensuring intuitive user interaction in software development.

Montreal, Canada Area

During my internship at Accedian, I focused on automating and enhancing the company's device simulation software for large-scale testing. By innovatively using Python, I coded frameworks for swift and efficient network topology generation and orchestration, achieving notable speed improvements (up to 15x). My expertise also extended to developing APIs, managing virtual machines, and delving deep into Linux-based environments and tools.

Montreal, Canada Area

Working at Marianopolis, one of Montreal's most eminent CEGEPs, I tutored students in differential calculus. This involved creating customized problem sets tailored to the student's needs and simplifying mathematical concepts to improve their understanding.

Doctor Of Philosophy - Phd, Computer Science

Mmath, Computer Science

Master of Mathematics (Computer Science) thesis option candidate supervised by Krzysztof Czarnecki

Bachelor Of Science - Bs, Computer Science

Bachelor'S Degree, Mathematics And Statistics

College Degree, Business/Commerce, General