Perouz Taslakian
Research Lead
Adaptive Agent
Perouz Taslakian is a Research Scientist and Research Lead at ServiceNow AI Research, where she leads research on adaptive AI agent systems. She is also an Adjunct Professor in the School of Computer Science at McGill University and an Associate Industry Member of Mila – Quebec AI Institute. Her research focuses on agentic AI systems, multimodal foundation models, and AI security, privacy.
Prior to ServiceNow, Perouz held research and leadership positions at Element AI, Samsung AI Lab, and Morgan Stanley. She has contributed to a broad range of areas spanning multimodal learning, knowledge representation, graph learning, causality, and AI agents, with work ranging from theoretical foundations to large-scale industrial AI systems.
Before transitioning to industrial research, Perouz served as Professor and Chair of the BSc in Computational Sciences program at the American University of Armenia.
Perouz received her PhD in Computer Science from McGill University under the supervision of Luc Devroye and Godfried Toussaint. Her doctoral research focused on discrete and computational geometry.
Interests
- Machine Learning
- Theory of Machine Learning
- Causality
- Multimodal Models
Publications
Hierarchical Retrieval at Scale: Bridging Transparency and Efficiency.
International Conference on Machine Learning (ICML),
2026.
Hierarchical Retrieval at Scale: Bridging Transparency and Efficiency.
Workshop at the International Conference of Machine Learning (ICML),
2026.
Grounding Computer Use Agents on Human Demonstrations.
International Conference on Learning Representations,
2026.
StarFlow: Generating Structured Workflow Outputs From Sketch Images.
European Chapter of the Association for Computational Linguistics (EACL),
2026.
AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding.
Neural Information Processing Systems (NeurIPS),
2025.
Rendering-Aware Reinforcement Learning for Vector Graphics Generation.
Neural Information Processing Systems (NeurIPS),
2025.
WebMMU: A Benchmark for Multimodal Multilingual Website Understanding and Code Generation.
Conference on Empirical Methods in Natural Language Processing (EMNLP),
2025.
BigCharts-R1: Enhanced Chart Reasoning with Visual Reinforcement Finetuning.
Conference on Language Modeling (COLM),
2025.
BiXSE: Improving Dense Retrieval via Probabilistic Graded Relevance Distillation.
Conference on Language Modeling (COLM),
2025.
UI-Vision: A Desktop-centric GUI Benchmark for Visual Perception and Interaction.
International Conference on Machine Learning (ICML),
2025.
WebMMU: A Benchmark for Multimodal Multilingual Website Understanding and Code Generation.
Workshop at the Computer Vision and Pattern Recognition Conference (CVPR),
2025.
AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Understanding.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
Learning to Defer for Causal Discovery with Imperfect Experts.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
WebMMU: A Benchmark for Multimodal Multilingual Website Understanding and Code Generation.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks.
International Conference of Learning Representations (ICLR),
2025.
InsightBench: Evaluating Business Analytics Agents Through Multi-Step Insight Generation.
International Conference of Learning Representations (ICLR),
2025.
VCR: Visual Caption Restoration.
International Conference of Learning Representations (ICLR),
2025.
BigDocs: A Permissively-Licensed Dataset for Training Vision-Language Models on Document and Code Tasks.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content.
NeurIPS Datasets and Benchmarks Track (NeurIPS Datasets),
2024.
VCR: Visual Caption Restoration.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Conference on Empirical Methods in Natural Language Processing (EMNLP),
2024.
A Sparsity Principle for Partially Observable Causal Representation Learning.
International Conference on Machine Learning (ICML),
2024.
Multi-View Causal Representation Learning with Partial Observability.
International Conference of Learning Representations (ICLR),
2024.
A Sparsity Principle for Partially Observable Causal.
Workshop at the Neural Information Processing Systems (NeurIPS),
2023.
Capture the Flag: Uncovering Data Insights with Large Language Models.
Workshop at the Neural Information Processing Systems (NeurIPS),
2023.
Multi-View Causal Representation Learning with Partial Observability.
Workshop at the Neural Information Processing Systems (NeurIPS),
2023.
Explaining Graph Neural Networks Using Interpretable Local Surrogates.
Workshop at the International Conference on Machine Learning (ICML),
2023.
OC-NMN: Object-centric Compositional Neural Module Network for Generative Visual Analogical Reasoning.
Workshop at the International Conference on Machine Learning (ICML),
2023.
Knowledge Hypergraph Embedding Meets Relational Algebra.
Journal of Machine Learning Research (JMLR),
2023.
Object-centric Compositional Imagination for Visual Abstract Reasoning.
Workshop at the International Conference on Learning Representations (ICLR),
2022.
Typing assumptions improve identification in causal discovery - theory and algorithms.
Causal Learning and Reasoning (CLeaR),
2022.
Typing assumptions improve identification in causal discovery - Report and comments on future directions.
Workshop at the Neural Information Processing Systems (NeurIPS),
2021.
Typing assumptions improve identification in causal discovery.
Workshop at the International Conference on Machine Learning (ICML),
2021.
Knowledge Hypergraphs: Prediction Beyond Binary Relations.
International Join Conference on Artificial Intelligence (IJCAI),
2020.
Knowledge Hypergraphs: Prediction Beyond Binary Relations.
Workshop at the Association for the Advancement of Artificial Intelligence (AAAI),
2020.
Knowledge Hypergraphs: Prediction Beyond Binary Relations.
Women in Machine Learning (WiML),
2019.
Context-Aware Visual Compatibility Prediction.
Computer Vision and Pattern Recognition (CVPR),
2019.
Efficient Multi-Robot Coverage of a Known Environment.
International Conference on Intelligent Robots and Systems (IROS),
2017.
Continuous Yao Graphs.
Journal on Computational Geometry (CG),
2017.