CV

Machine Learning Scientist

Advanced Sensor Research Inc. (ASR)

Building end-to-end time-series machine learning pipelines for non-invasive glucose estimation.

• Built end-to-end time-series machine learning pipelines for non-invasive glucose estimation. [PyTorch, TensorFlow, Hugging Face]
• Developed and benchmarked Transformer-based sequence models alongside baselines (LSTM, TCN), including representation learning with autoencoder-style pretraining. [PyTorch]
• Ran structured architecture and hyperparameter experiments; compared training protocols for stability across sensor noise and session variability. [PyTorch, Scikit-learn]
• Designed evaluation splits to test generalization across subjects/sessions/sensor units; investigated domain generalization and domain adaptation strategies. [PyTorch, Pandas]
• Implemented explainability and error analysis workflows (slice-based analysis, failure mode analysis, calibration/uncertainty checks). [PyTorch]
• Performed exploratory data analysis and signal-quality investigations to identify data issues and guide data-cleaning and collection improvements. [SciPy]

Intern Research Scientist

Samsung Research Montreal

Designed and evaluated deep reinforcement learning methods for network load balancing and energy efficiency.

• Designed and evaluated deep reinforcement learning methods for network load balancing in simulated environments. [PyTorch, OpenAI Gym, Stable Baselines3]
• Implemented curriculum learning to improve training robustness and generalization to harder scenarios. [PyTorch, Stable Baselines3]
• Implemented a Vector-Quantized Variational Autoencoder (VQ-VAE) for representation learning within RL pipelines. [PyTorch]
• Adapted and integrated an Online Decision Transformer into an existing RL framework to enhance network load balancing and energy efficiency. [PyTorch, Hugging Face]

Software Engineer

Pars Cognition

Created engaging video game experiences aimed at cognitive assessment and development.

• Developed mini-serious video games targeting cognitive science applications and research.

Doctor of Philosophy

University of Alberta

Computing Science

Thesis: Automated Personalized Exposure Therapy Based on Physiological Measures Using Experience-driven Procedural Content Generation via Reinforcement Learning

• Intro to Virtual/Augmented Reality and Telepresence
• Machine Learning and the Brain
• Image Processing and Analysis in Diagnostic Imaging

Master of Science,

Shahid Beheshti University,

Computer Engineering/Artificial Intelligence

Thesis: Pattern Extraction and Analysis of the Behavior of Children with Autism Spectrum Disorders Using a Designed Video Game

• Machine Learning
• Neural Network
• Pattern Recognition
• Data Mining
• Image Processing
• Natural Language Processing

Bachelor of Science

University of Tehran,

Computer Engineering/Software Engineering

Thesis: Designed and Developed a Video Game for Children with Autism to Improve their Receptive Language Skills.

• Advanced Programming
• Database Design
• Data Structures
• Artificial Intelligence
• Operating Systems
• Human-Computer Interaction
• Intro to Multimedia
• Intro to eLearning

Spiders Based on Anxiety: How Reinforcement Learning Can Deliver Desired User Experience in Virtual Reality Personalized Arachnophobia Treatment

ACM Transactions on Interactive Intelligent Systems

This paper introduces a framework for personalized virtual reality exposure therapy for arachnophobia that leverages procedural content generation and reinforcement learning to automatically adapt virtual spiders to elicit specific anxiety responses, demonstrating superior performance over traditional rules-based methods.

Stress Detection from Photoplethysmography in a Virtual Reality Environment

ArXiv

This article presents a virtual reality exposure therapy platform that non-intrusively assesses patients' mental states using photoplethysmography (PPG) signals, achieving a 70.6% accuracy in classifying relaxing and stressful states—outperforming more complex methods.

Label-Free Subjective Player Experience Modelling via Let's Play Videos

Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment

This paper introduces a label-free approach to Player Experience Modelling (PEM) using gameplay videos, validated through strong correlations with affective measures in Angry Birds gameplay.

Arachnophobia exposure therapy using experience-driven procedural content generation via reinforcement learning (EDPCGRL)

Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment

This paper proposes a reinforcement learning-based approach that uses physiological measures to automatically adapt therapeutic content in arachnophobia exposure therapy, generating personalized virtual spiders that adapt more quickly and accurately than existing search-based methods.

Automated personalized exposure therapy based on physiological measures using experience-driven procedural content generation

Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment

This research proposes a personalized virtual reality exposure therapy framework that utilizes physiological sensors and machine learning algorithms to automatically adapt exposure parameters through procedural content generation, addressing the limitations of subjective, hand-authored methods, with planned human studies on arachnophobia and fear of public speaking.

A framework for easier designs: Augmented intelligence in serious games for cognitive development

IEEE Consumer Electronics Magazine

This article introduces a portable, rapid development framework for intelligent serious games that leverage human intervention to minimize complex AI requirements, enabling caregivers to foster social interactions with children with special needs across various settings.

The differences between children with autism and typically developed children in using a hand-eye-coordination video game

Ubiquitous Computing and Ambient Intelligence: 11th International Conference, UCAmI

This paper introduces a touch-based hand-eye coordination video game that effectively differentiates between children with autism and typically developing children using only two in-game features, demonstrating its potential as a cost-free tool for initial screening of motor coordination issues in autism.

Transformer-Based Glucose Level Estimation

Built end-to-end time-series pipelines using Transformer, LSTM, and autoencoder models for non-invasive glucose estimation from sensor data. [PyTorch, TensorFlow]

• Benchmarked Transformer architectures against LSTM and TCN baselines
• Implemented representation learning with autoencoder-style pretraining

RLHF-Inspired Personalized Reward Learning

Proposed an approach inspired by RLHF that learns reward models from physiological feedback and user embeddings for faster agent personalization. [PyTorch]

• Designed reward models incorporating physiological signals and user-specific embeddings
• Implemented pretraining and fine-tuning pipelines for rapid adaptation to individual users

Domain Adaptation for Sensor Data

Developed domain adaptation and generalization techniques to handle sensor variability across users, body locations, and sensor units. [PyTorch, Pandas]

• Addressed domain shift where sensors behave differently per person and placement
• Implemented transfer learning strategies across sensor configurations

Online Decision Transformer

Adapted and integrated an Online Decision Transformer into RL frameworks for real-time sequential decision-making. [PyTorch, Hugging Face]

• Extended offline Decision Transformer for online learning scenarios
• Applied to network load balancing and energy efficiency optimization

VQ-VAE for Time-Series Representation Learning

Implemented Vector-Quantized Variational Autoencoder for clustering and representation learning on high-dimensional time-series data. [PyTorch]

• Designed discrete latent representations for improved clustering
• Applied to network traffic patterns for efficient data compression

Explainability and Error Analysis Framework

Built workflows for model interpretability including slice-based analysis, failure mode detection, and uncertainty quantification. [PyTorch, SHAP]

• Implemented systematic error analysis to identify prediction failure patterns
• Designed calibration checks to assess model confidence reliability

Experience-Driven PCG via Reinforcement Learning

Developed RL-based procedural content generation that adapts virtual environments based on real-time physiological feedback. [PyTorch, OpenAI Gym, Stable Baselines3]

• Trained DQN and PPO agents to generate personalized therapeutic content
• Integrated physiological sensors for closed-loop adaptation

Curriculum Learning for RL

Implemented curriculum learning strategies to improve RL training robustness and generalization to harder scenarios. [PyTorch, Stable Baselines3]

• Designed progressive task difficulty scheduling for stable learning
• Improved agent generalization through structured training curricula

Stress Detection from Physiological Signals

Built classification models to detect stress states from PPG and physiological responses using ML and deep learning. [PyTorch, Scikit-learn]

• Compared traditional ML methods with LSTM-based approaches
• Achieved 70.6% accuracy in VR-based stress classification

Heart Rate Estimation from Video

Developed CNN-based remote photoplethysmography (rPPG) for non-contact heart rate estimation from facial video. [PyTorch]

• Implemented end-to-end deep learning pipeline for video-based vital signs
• Enabled non-intrusive physiological monitoring without wearables

Player Experience Modeling

Built data-driven models to predict player affect from gameplay videos for adaptive game content generation. [PyTorch]

• Developed label-free approach using Let's Play videos
• Validated through human studies correlating predictions with affective measures

Adaptive VR Environment

Created VR environments with real-time parameter adaptation based on user interactions and physiological responses. [Unity, C#]

• Integrated physiological sensors for closed-loop environment control
• Applied to exposure therapy for anxiety disorders

Behavior Pattern Analysis

Applied ML techniques to identify and classify user behavior patterns in interactive game environments. [MATLAB]

• Extracted features to differentiate behavioral patterns
• Applied to autism research for early screening applications

Hidden Semi-Markov Model for Mobility Tracking

Implemented probabilistic models to track mobility patterns with missing data and multiple observation sequences. [Java]

• Handled incomplete and noisy sequential data
• Improved tracking accuracy through advanced statistical modeling

Serious Games for Cognitive Development

Developed educational and therapeutic video games targeting cognitive assessment and social skill development. [Unity, GameMaker]

• Designed game mechanics for children with autism spectrum disorders
• Created engaging experiences for cognitive science research

Reviewer

PeerJ Journal

Reviewed academic articles submitted to the PeerJ journal, ensuring the quality and validity of research in AI and related fields.

Program Committee Member

AIIDE Conference 2025

Provided peer reviews for research submissions focused on Artificial Intelligence and Interactive Digital Entertainment.

Reviewer

Transactions on Affective Computing journal

Served as a reviewer for ransactions on Affective Computing journal, providing peer reviews for submissions in human-computer interaction and related fields.

Program Committee Member

Experimental AI in Games (EXAG)

Active committee member served as a reviewer for EXAG workshops focused on experimental AI in games.

Reviewer

CHI Conference 2024

Served as a reviewer for CHI Conference 2024, providing peer reviews for submissions in human-computer interaction and related fields.

Treasurer

Computer Science Graduate Student Association (CSGSA)

Managed the financial operations of the CSGSA, a voluntary group offering support and activities for Computing Science graduate students at the University of Alberta.

Olympiad Competition Award

Math and Literature Olympiads for Pre-University Students

Awarded in the Math and Literature Olympiads for pre-university students.

Ranked 459th in the Undergraduate Nationwide Universities Entrance Exam

Nationwide Universities Entrance Exam, Iran

Achieved a ranking of 459th in the Undergraduate Nationwide Universities Entrance Exam in Iran, competing against over 300,000 participants.

Ranked 65th in the Graduate Nationwide Universities Entrance Exam

Graduate Nationwide Universities Entrance Exam, Iran

Achieved a ranking of 65th in the Graduate Nationwide Universities Entrance Exam in Iran, out of more than 30,000 participants.