Perouz Taslakian

Timeline

1. 1998–2009

   Degrees from Haigazian to McGill

   She earns a BSc in computer science at Haigazian University in 1998, an MSc in computer science at Concordia University in 2004 and a PhD in computer science at McGill University in 2009. Her academic research focuses on theoretical and algorithmic aspects of discrete structures, with problems on coloring and guarding line arrangements, aligning necklaces, deflating polygons and flipping linkages.

2. 2007

   The distance geometry of music

   She is a coauthor of The Distance Geometry of Music, posted to arXiv in 2007 with Erik Demaine, Godfried Toussaint and others. The paper shows how the structure of the Euclidean algorithm defines a family of rhythms covering over forty timelines from traditional world music, and proves that these Euclidean rhythms spread their onsets as evenly as possible.

3. 2014–2015

   Program chair in Yerevan

   In October 2014, she is Computational Sciences Program Chair at the American University of Armenia and joins a workshop in Leuven, Belgium for program chairs and deans. In March 2015, as Assistant Professor and Program Chair of the BS in Computational Sciences, she takes part in the Third Annual Workshop on Geometry and Graphs at the Bellairs Research Institute in Barbados. Hayk Nersisyan is appointed the new program chair in July 2015, and the College of Science and Engineering thanks her as former chair.

4. By 2018

   Applied research at Element AI

   She is an applied research scientist on Element AI's AI Core team in 2018, when MIT CSAIL lists her to present the team's progress on representation learning and vision at its tech talk in October 2018. Her Element AI research centers on graph learning and causality, and she leads the Human Decision Support Program, which combines time-series analysis with graph and causal representation learning.

5. 2019–2023

   Hypergraphs and causal representations

   In 2019, she coauthors Knowledge Hypergraphs: Prediction Beyond Binary Relations with Bahare Fatemi, David Vazquez and David Poole, which introduces the embedding methods HSimplE and HypE for link prediction on relations that involve any number of entities. In 2023, she coauthors Multi-View Causal Representation Learning with Partial Observability, a framework for the identifiability of representations learned from simultaneously observed views such as different data modalities.

6. 2024–2025

   Open datasets and agent benchmarks

   In December 2024, she is among the authors of BigDocs, an open dataset of 7.5 million multimodal documents across 30 tasks. In 2025, she coauthors UI-Vision, a benchmark for evaluating computer use agents in desktop environments across 83 software applications, and Rendering-Aware Reinforcement Learning for Vector Graphics Generation, which introduces RLRF for SVG generation in vision-language models. The UI-Vision paper appears at ICML 2025 and the RLRF paper at NeurIPS 2025.

7. As of October 2026

   Research lead at ServiceNow

   She is a Staff Research Scientist and Research Lead at ServiceNow, an adjunct professor in McGill University's School of Computer Science and an associate industry member at Mila. Her earlier industry work includes a post as research scientist at Samsung Research, where she applies AI techniques to problems in wireless communication networks. In September 2026, she coauthors Continual Enterprise World Model Discovery in Dynamic Systems, which introduces EnterpriseWorldShift, a benchmark built on a live ServiceNow environment, and a Continual Discovery Agent that learns an enterprise system's hidden business rules.