Hi! My name is Arash. I am currently working as a PhD Research Assistant at InfoLab, Computer Science Department at the University of Southern California under the supervision of Prof. Cyrus Shahabi.
My research interests include Graph Neural Networks and Spatio-Temporal Data Management and Forecasting.
Please check out my CV if you want to find more information about me and my research!
PhD in Computer Science, 2022 - Current
University of Southern California
MSc in Computer Science, 2022 - 2024
University of Southern California
BSc in Computer Engineering, 2017 - 2021
Amirkabir University of Technology, CGPA 19.28 / 20
Led the development of the W4H Integrated Toolkit, an open-source toolkit centralizing both real-time and offline wearable data from various sources (e.g., Garmin, Apple Watch, Fitbit).
Designed a scalable system architecture separating data engineering, analysis, and visualization.
The toolkit comprises the following open-sourced tools:
StreamSim: Real-time data streaming simulator using Python and Flask.
W4H ImportHub: Integrates stored datasets with Python, SQLAlchemy, and Streamlit.
pyGarminAPI: Python library for interacting with the Garmin API.
Integrated Analytics Dashboard: Core component for data extraction and analysis using Streamlit, pandas, Flask, Spark, and Kafka.
Released the toolkit in two modes: a Docker image for local setup and a centralized version on USC clusters.
I developed a GNN-based deep learning model to analyze brain correlations across spatial, semantic, and temporal dimensions using EEG signals.
Addressing the complex task of predicting visits to Points of Interest (POIs), I formulated the problem as a multivariate time-series forecasting challenge.
Trained and guided two undergraduate students on an academic project during the Summer of 2022, enhancing their research capabilities and ensuring project success.
Responsibilities include:
The study presents the Busyness Graph Neural Network (BysGNN), designed to forecast visits to urban Points-of-Interest (POIs). BysGNN uncovers multi-context correlations among POIs across temporal, spatial, and semantic dimensions, resulting in a comprehensive dynamic graph. This approach significantly improves forecasting accuracy over traditional methods, offering a deeper understanding of urban dynamics.