Mohamadali Tofigh

Postdoctoral Researcher · University of Alberta

Mohamadali Tofigh

I work on reinforcement learning and multimodal foundation models for agents that act in the physical world: LLM-guided reward design, spatial reasoning in vision-language models for motion planning, foundation models as teachers for sample-efficient RL learning, and agentic workflows for efficient embodied intelligence.

A track record in deep learning for sequential data, from interpretable sequence models to temporal graph networks, and hands-on experience putting ML into production on industrial fleets.

  • Reinforcement Learning
  • Multimodal LLMs
  • Embodied Agents
  • Deep Sequence Models
Email LinkedIn Google Scholar GitHub CV ↓
Mohamadali Tofigh

About

I'm a postdoctoral researcher in the Department of Computing Science at the University of Alberta, working with Dr. Osmar Zaïane and affiliated with the Alberta Machine Intelligence Institute (Amii).

My research asks how foundation models can make reinforcement learning agents practical for physical systems. That splits four ways: getting language models to write dense reward functions that actually shape behaviour, understanding where multimodal models succeed and fail at spatial reasoning, building teacher–student setups where a foundation model's action advice buys sample efficiency in online learning, and creating agentic frameworks to automate high-level planning and low-level control.

I got here through deep learning for time series. My PhD built sequence architectures for complex physical systems, interpretable temporal convolution models, Transformer-based forecasting for early-stage degradation, Kolmogorov–Arnold networks for time-varying dynamics, and temporal graph neural networks for anomaly detection. Twelve journal papers came out of asking what it takes to make a learned model of a real, noisy, sequential process reliable enough to act on. That question didn't change when I moved to embodied AI — it just got harder. Foundation models or RL agents are sequence models that also have to plan, choose, and act.

As a Machine Learning Resident at Amii, I built failure-risk prediction and anomaly detection for a commercial vehicle fleet, end-to-end on Azure. I currently mentor several graduate students and research assistants at the CS Department at UAlberta.

Position
Postdoctoral Researcher, Computing Science
Institution
University of Alberta | Amii
Advisor
Dr. Osmar Zaïane
Location
Edmonton, Alberta, Canada
PhD
University of Alberta, 2026 — deep learning for modelling and anomaly detection in dynamical systems
Open to
Research collaborations, faculty and industry research roles