Hi there 👋, I'm Masoud, a graduate student in Computing Science at the University of Alberta, supervised by Prof. Osmar R. Zaïane, and currently a Research Intern at Electronic Arts (EA) 🎮.
My thesis research focuses on visual and spatial reasoning in multimodal language models. Currently, I am exploring how inference-time scaling, adaptive context control for embodied applications, and RL post-training can improve the efficiency and adaptability of spatial reasoning. The idea of scaling inference has some roots in the internal world models of language models. Spatial reasoning has broad applications across domains including robotics, autonomous driving, video games, and VR/AR.
At EA, my research focuses on developing small, efficient language models for real-time decision-making and video game applications.
- Spatial & Temporal Understanding for Planning in Language-based Embodied Agents
- Efficient Spatial & Visual Reasoning with LLMs & VLMs
- 3D Representations, Grounding, & Space Understanding in MLLMs
- Building Vision-Language Datasets for Post-training (SFT/RLFT/OPSD) MLLMs on Spatial Reasoning Tasks
- Visual and Geometry Retrieval Systems (non-textual, multi-modal RAG)
- M.Sc. in CS, University of Alberta (Present)
- Ph.D. in ECE, University of Alberta (Transferred to CS)
- M.Sc. & B.Sc. in ME, Sharif University of Technology & Univ. of Tehran
