Atefeh Termehchi

Machine learning researcher with 5+ years of experience in deep learning interpretability, generalization theory, out-of-distribution generalization, and reliable machine learning for complex real-world systems. Expertise in analyzing generalization and distribution shifts in deep learning models and developing domain-shift rejection methods for recurrent-neural-network-based industrial systems, resulting in a patent application. Experienced in advancing machine learning research from problem formulation and theoretical development through implementation, experimentation, validation, and scientific publication.

Strong background in applying machine learning to industrial and cyber-physical systems, including industrial control systems, autonomous systems, energy forecasting, wireless networks, and industrial cybersecurity. Postdoctoral researcher at the University of Manitoba and formerly at the University of Ottawa, with 10+ years of combined academic research and industrial engineering experience.

Postdoctoral Fellow at University of Manitoba (2024-04 – Present)

Supervised by Prof. Ekram Hossain

• Led and mentored a junior researcher in developing a data-driven stability framework for safe and communication-efficient consensus in multi-agent deep reinforcement learning.
• Developed theoretical generalization bounds and domain-generalization methods for recurrent neural networks, with experiments evaluating robustness under distribution shifts in real-world industrial applications.
• Developed data-driven methods for analyzing out-of-distribution generalization and deriving theoretical generalization bounds for deep reinforcement learning agents.
• Led the development of a science-informed deep learning framework that integrates domain knowledge into model architecture, training, and validation for wireless-system applications.
• Developed and validated domain-shift rejection methods for RNNs, resulting in a patent application and supporting reliable application in industrial systems such as smart-grid energy forecasting.
• Developed and implemented a predictive, physics-based beamforming control method for RIS-aided wireless communications and contributed to a related patent application.
• Developed a consensus-aware co-design framework for jointly optimizing graph-based multi-agent control and wireless resource allocation in next-generation networks.

Postdoctoral Fellow at University of Ottawa (2023-01 – 2024-03)

Supervised by Prof. Melike Erol-Kantarci

• Conducted research in collaboration with Nokia Bell Labs, focusing on autonomous, safe, and communication-efficient deep learning methods for next-generation UAV-aided wireless networks.
• Developed and evaluated a distributed safe multi-agent deep reinforcement learning framework.
• Developed a goal-oriented deep reinforcement learning method combining supervised learning and reinforcement learning to reduce unnecessary environment interactions while maintaining network performance.
• Contributed to the development of a hierarchical deep reinforcement learning framework incorporating age-of-information metrics for UAV trajectory planning in smart-agriculture applications.

Industrial Cybersecurity Researcher and Risk & Compliance Specialist at Industrial Cybersecurity Research Center (2013-05 – 2019-12)

• Developed cyber-risk assessment software for industrial control systems (ICS), enabling automated vulnerability analysis, risk quantification, and structured security evaluation.
• Designed, implemented, and commissioned a hardware-in-the-loop industrial cybersecurity testbed for testing and validating cyberattack scenarios in ICS environments.
• Designed and implemented a Modbus intrusion detection system for detecting anomalous network activity and cyber threats in industrial communication data.
• Conducted cyber-risk assessments for oil, gas, and power facilities, identifying security vulnerabilities and supporting regulatory compliance.
• Conducted research on cybersecurity frameworks for next-generation industrial Internet of Things systems, addressing emerging threats in connected industrial environments.

Electrical and Instrumentation Engineer at Arkan Control Group (2009-05 – 2010-12)

On behalf of Swedish company Rosemount Tank Gauging

• Design, installation, and commissioning of Rosemount Tank Gauging systems for oil and gas plants.

Ph.D. in Information Technology Engineering – Amirkabir University of Technology (Tehran Polytechnic) (2015-09 – 2022-02)

M.Sc. in Electrical Engineering, Control Engineering – Amirkabir University of Technology (Tehran Polytechnic) (2010-12 – 2013-02)

B.Sc. in Electrical Engineering, Control Engineering – Shiraz University of Technology (2004-12 – 2008-08)