About Me

Alireza Keshavarzian

Ph.D. Candidate in Electrical and Computer Engineering at University of Toronto
Data Scientist | Machine Learning Researcher | AI Engineer

Education

Ph.D. in Electrical and Computer Engineering

University of Toronto

2020 – Present

Research focus on Random Representation Learning for Time Series, developing novel deep learning methods for time-series data analysis.

M.Sc. in Electrical Engineering

Amirkabir University of Technology (Tehran Polytechnic)

2016 – 2019

Thesis: Human Action Recognition on Distributed Big Data Infrastructure

B.Sc. (Double Major) in Computer Engineering and Information Technology

Amirkabir University of Technology (Tehran Polytechnic)

2014 – 2019

B.Sc. in Electrical Engineering

Amirkabir University of Technology (Tehran Polytechnic)

2012 – 2016

Work Experience

Data Scientist Intern

DraftAid

June – September 2025

Developed a conditional autoregressive GNN that predicts annotation edges sequentially, conditioning each step on previously accepted graph states.

Data Scientist

University Health Network (UHN)

December 2020 – Present

Design deep learning models for respiratory disease analysis, including the OscilloFusion Attention architecture for multi-channel oscillometry fusion and phenotype classification. Developed an interface enabling clinicians to explore oscillometry signals, attention maps, and predicted mechanical parameters.

Data Scientist

Snappfood

May 2020 – October 2020

Developed a recommendation system for food and restaurants, leveraging a model to generate dynamic meta-tags for foods and restaurant listings, increasing click-through rates. Conducted in-depth analysis of conversion rates across application pages, identifying key areas for optimization and improving user engagement.

Machine Learning Engineer

RISE

January 2020 – August 2020

Researching uncertainty measurement and conformal prediction.

Chief Technical Officer (CTO)

Atrovan

November 2016 – April 2020

Team leader and software architect of Smart Home Gateway based on Zigbee protocol. Developer of Atrovan edge layer. Worked on Modbus and MQTT gateways deployed on the fog layer of smart solutions. Led engineering teams for the Atrovan IoT platform.

Teaching Assistant

University of Toronto

January 2021 – March 2023

Courses: Probabilistic Machine Learning (CSC412), Computer Networks I (ECE361), Programming Fundamentals (ECE244), Design and Analysis of Data Structures (CSCB63).

Research Interests

Machine Learning in Healthcare

Developing deep learning models for medical data analysis, particularly in respiratory disease diagnosis and treatment.

Time Series Analysis

Random representation learning and segmentation-based methods for time-series classification and feature extraction.

High-Dimensional Data Integration

Feature selection and dimensionality reduction techniques for complex, high-dimensional datasets.

Contrastive and Few Shot Learning

Developing contrastive learning frameworks for scalable feature selection and representation learning.

Publications

CoMIND: A Contrastive Multi-metric Approach to INformative and Discriminative Feature Selection

A. Keshavarzian, S. Valaee

Submitted to: International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2026)

CURVE: Contrastive Unsupervised Representation-based Variable Elimination

A. Keshavarzian, S. Valaee

Submitted to: IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)

Random Representation Learning based on Segmented Pooling Layer and Information Spread Detector

A. Keshavarzian, S. Valaee

Submitted to: IEEE Transactions on Artificial Intelligence (TAI)

HIERVAR: A Hierarchical Feature Selection Method for Time Series Analysis

A. Keshavarzian, S. Valaee

Published in: International Workshop on Machine Learning for Signal Processing (MLSP 2024)

RASTER: Representation Learning for Time Series Classification using Scatter Score and Randomized Threshold Exceedance Rate

A. Keshavarzian, S. Valaee

Published in: International Workshop on Machine Learning for Signal Processing (MLSP 2023)

Representation Learning of Clinical Multivariate Time Series with Random Filter Banks

A. Keshavarzian, H. Salehinejad, S. Valaee

Published in: International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2023)

Time Series Classification Using Convolutional Kernel and Adaptive Dynamic Thresholding

A. Keshavarzian, J. Wue, C. Chow, S. Valaee

Published in: IEEE International Conference on Communication (ICC 2024)

Modified Deep Residual Network Architecture Deployed on Serverless Framework of IoT Platform Based on Human Activity Recognition Application

A. Keshavarzian, S. Sharifian, S. Seyedin

Published in: Future Generation Computer Systems Journal

Skills & Expertise

Programming Languages

C, C++, Python, Golang, Java

Frameworks & Libraries

Spark, Hadoop, PyTorch, TensorFlow, Keras, OpenCV, Numba, OpenAI SDK, CrewAI

Data Analysis & Visualization

Pandas, Seaborn, Plotly, Power BI, PySpark

Large Language Models

CrewAI, OpenAI SDK, LangChain, LLM fine tuning

Hardware & Embedded Systems

Xilinx, Verilog, VHDL, Microprocessors, Raspberry Pi

Web Technologies

HTML, CSS, JavaScript, Bootstrap, React.js, MongoDB, CassandraDB, RDBMS

Academic Projects

Ph.D. Project: Random Representation Learning for Time Series

Supervised by: Dr. S. Valaee

Develops random representation learning and segmentation-based deep learning methods for time-series data, including RASTER, ISD, HIERVAR, and T-ROCKET, to identify informative regions and enhance feature discriminability. Extends these models to unsupervised and contrastive learning frameworks for scalable feature selection in high-dimensional settings. Focuses on theoretical analysis of segment-wise informativeness and contrastive objectives to build robust, ML-ready representations.

Technologies: Python, PyTorch, TensorFlow, Keras

M.Sc. Thesis: Human Action Recognition on Distributed Big Data Infrastructure

Supervised by: Dr. S. Sharifian and Dr. S. Seyedin

Built and optimized deep neural models for smartphone sensor–driven action recognition, integrating distributed training and inference on Apache Spark.

Technologies: Python, Keras, TensorFlow, Spark

B.Sc. Capstone: Real-time Static Logo Detection and Inpainting in Sports Videos

Supervised by: Prof. S.A. Motamedi

Designed and implemented GPU-accelerated algorithms for logo detection and inpainting using CUDA.

Technologies: C/C++, CUDA, Matlab (2017)

IoT Platform: Data Aggregation and Real-time Analytics

Architected a distributed microservices IoT platform for high-throughput data ingestion, dashboarding, and low-latency analytics. Implemented a FaaS-based rule engine using OpenWhisk for scalable event processing.

Technologies: Golang, OpenWhisk, Cassandra, Postgres (2019)