Parvin Malekzadeh

Hi!

I'm a Ph.D. candidate at the Edward S. Rogers Sr. Department of Electrical and Computer Engineering (ECE) at the University of Toronto (UofT), mentored by Kostas Plataniotis. Currently, I'm collaborating with FinHub. My academic journey includes completing my Master of Research (MRes) at Concordia University under the guidance of Arash Mohammadi, where interned at Dormakaba Inc. for two years.

Research

My research focuses on advancing autonomous artificial intelligence, targeting decision-making and interaction within real-world environments. This goes beyond the limitations of task-specific agents and involves substantial work in Reinforcement Learning and Active Inference, spanning three critical areas:

Transfer Learning: increasing the adaptability of autonomous agents through efficient knowledge transfer between varied tasks, leading to enhanced learning and adaptability across different environments. Key papers: AKF-SR , Makf-Sr , UaMB-SF
Uncertainty-Aware Decision Making: Enabling agents to effectively handle uncertainty in environments with partial observability, tackling uncertainties in environment, state, or algorithms. This aims to prepare AI systems for making informed decisions in situations with incomplete or unpredictable information. Key papers: Unified Inference , MM-KTD , AKF-SR , Makf-Sr , UaMB-SF .
Robust Loss Function Design: Acknowledging uncertainty's impact on decision-making, focused on creating robust loss functions that can effectively manage uncertainty, thereby improving the reliability and performance of AI agents in dynamic and uncertain scenarios. Key papers: A Robust Quantile Huber Loss, UUaE .

Highlighted Publications

See Google Scholar for more publications.

	UaMB-SF: Uncertainty-aware transfer across tasks using hybrid Model-Based Successor Feature reinforcement learning P Malekzadeh, M Hou, KN Plataniotis Neurocomputing, 2023 paper / webpage
	Active Inference and Reinforcement Learning: A unified inference on continuous state and action spaces under partially observability P Malekzadeh, KN Plataniotis Preprint paper / webpage
	A robust quantile huber loss with interpretable parameter adjustment in distributional reinforcement learning P Malekzadeh, KN Plataniotis, Z Poulos, Z Wang ICASSP , 2024 paper / code / webpage
	AKF-SR: Adaptive Kalman filtering-based Successor Representation P Malekzadeh, M Salimibeni, M Hou, A Mohammadi, KN Plataniotis Neurocomputing, 2022 paper / code / webpage
	UUaE: A Unified Uncertainty-aware Exploration combining epistemic and aleatory uncertainty P Malekzadeh, M Hou, KN Plataniotis ICASSP , 2023 paper / webpage
	MM-KTD: Multiple Model Kalman Temporal Differences for reinforcement learning P Malekzadeh, M Salimibeni, A Mohammadi, A Assa, KN Plataniotis IEEE Access, 2020 paper / code / webpage