Perouz Taslakian

Perouz Taslakian

Research Scientist and Program Lead in Machine Learning

Service Now Research

Biography

I am a Research Scientist in machine learning at Service Now Research in Montreal, and lead the Multimodal Learning Program.

My research at Service Now focuses on developing deep learning techniques to enhance large language models, and study their reasoning capacity through causal representation learning.

Previously, I was a research scientist at Samsung Research, where I developed AI techniques to address challenges that come up in wireless communication networks.

For a long time, I was also a Research Scientist and Research Lead at Element AI (currently Service Now), where I conducted fundamental research in machine learning, specializing in the area of graph learning and causality. I was concurrently the research lead of the Human Decision Support Program, whose goal was to develop decision-making models in a setting where data is volatile, relations are ambiguous, and past is not a good predictor of the future. The program effectively combined time-series analysis with graph and causal representation learning.

I obtained my PhD in Computer Science from McGill University. My academic research focused on theoretical and algorithmic aspect of discrete structures, with problems on coloring and guarding line arrangements, aligning necklaces, deflating polygons, flipping linkages, and proving things about of geometric graphs. Among my own papers, one of my favourites is perhaps Transversals in Trees, mainly because I had a lot fun working with Vašek, Luc, and Victor. But also because it shrank my Erdős number to 2, and gave me a flaming grade on Vašek’s course The mathematics of Paul Erdős.

I used to be a professor and chair of the BS in Computational Sciences Program at the American University of Armenia.

For about four years, I ran a local meetup group called All-Girl Hack Night, whose purpose was to bring together, support and empower women in IT.

Download my resumé.

Interests
  • Learning from graph-like data
  • Relational Reasoning
  • Causal Discovery
Education
  • PhD in Computer Science, 2009

    McGill University

  • MSc in Computer Science, 2004

    Concordia University

  • BSc in Computer Science, 1998

    Haigazian University

Publications

(2024). Multi-View Causal Representation Learning with Partial Observability. Submitted to International Conference on Learning Representations (ICLR).

Cite

(2023). A Sparsity Principle for Partially Observable Causal Representation Learning. Proceedings of NeurIPS Workshop on Causal Representation Learning.

Cite

(2023). Capture the Flag: Uncovering Data Insights with Large Language Models. NeurIPS 2023 Foundation Models for Decision Making Workshop.

Cite

(2023). Explaining Graph Neural Networks Using Interpretable Local Surrogates. Topological, Algebraic and Geometric Learning Workshops 2023.

Cite

(2023). OC-NMN : Object-centric Compositional Neural Module Network for Generative Visual Analogical Reasoning. ICML Workshop on Knowledge and Logical Reasoning in the Era of Data-driven Learning.

Cite

Conference Organization

BCL 2024

BCL 2024

Bellairs Workshop on Causality - Inference and Representation Learning

BCL 2023

BCL 2023

Bellairs Workshop on Causality - Inference and Representation Learning

GCP 2022

GCP 2022

Armenian Workshop On Graphs, Combinatorics, Probability

GroundedML 2022

GroundedML 2022

Workshop on Anchoring Machine Learning in Classical Algorithmic Theory (@ICLR)

GCP 2019

GCP 2019

Armenian Workshop On Graphs, Combinatorics, Probability

Contact