After receiving gold medal in Electrical Engineering from University of Western Ontario, I decided to pursue
graduate studies where I completed my MASc in Biomedical Engineering at the University of Toronto under
supervision of Dr. Babak Taati, Vector Affiliated Scientist. I am Currently Working as a Machine Learning
Research Associate at University Health Network (UHN) in Toronto, Canada. My research interests include various
deep learning sub-fields including Natural Language Processing (NLP), Computer Vision, and Speech Processing. I
am also interested in applying machine learning techniques to solve real-world problems in healthcare and
medicine.
Projects
a selection - other projects can be found on myGitHub Profile
Utilized web scraping solutions to gather data from diverse sources using Python and
web scraping frameworks such as beautifulSoup, Scrapy, and Selenium (for JavaScript-heavy websites).
Developed a comprehensive dataset for Natural Language Processing (NLP) tasks, which
was successfully added to the Hugging Face Model Hub.
Collaborated with a team of NLP experts affiliated with Vector Institute to define
annotation guidelines for the dataset, ensuring high-quality annotations and facilitating unsupervised
learning tasks.
Extensively worked with a wide range of NLP models, from traditional approaches like
Word2Vec and GloVe to state-of-the-art models such as BERT, and LLMs such as GPT-4 and LLAMA.
Performed prompt engineering: conducted in-depth research and analysis of prompts for
LLMs such as GPT-4 and LLaMa, understanding the impact of prompting on model behaviour and output
quality.
Applied unsupervised learning techniques to the dataset, including constrained
clustering, to group the wall clues into meaningful clusters based on contextual embeddings.
Paper: Saeid Alavi Naeini, Raeid Saqur, Mozhgan Saeidi, John Giorgi, Babak Taati.
"Large Language Models are Fixated by Red Herrings: Exploring Creative Problem Solving and
Einstellung Effect using the Only Connect Wall Dataset."Advances in Neural Information Processing
Systems (2023).
Curated and collected a large-scale dataset of TTS speech samples covering a wide
range of accents, dialects, and speaking rates, ensuring representation of various linguistic
characteristics.
Developed fine-tuning pipelines and strategies to adapt pre-trained ASR models to the
specific attributes of TTS speech data, addressing challenges such as accent variations, speaking rate
differences, and dysarthric disorders.
conducted extensive data processing, including utilization of Dynamic Time Warping for
temporal segmentation of speech samples.
Conducted thorough evaluations and benchmarking of the fine-tuned ASR models,
comparing their performance with the baseline models on standard evaluation metrics such as Word Error
Rate (WER) and accuracy.
Collaborated with domain experts and linguists to validate the improvements achieved
by proposed models, ensuring the enhanced models accurately captured dysarthric speech.
Implemented MLOps practices by developing a web application that analyzes dysarthric
speech and performs automatic temporal segmentation. The web app was developed using Docker and
Kubernetes to ensure efficient deployment and scalability of machine learning models.
Paper: SA Naeini, L Simmatis, D Jafari, Y Yunusova, B Taati. "Improving
Dysarthric Speech Segmentation with Emulated and Synthetic Augmentation." IEEE Journal of
Biomedical and Health Informatics (JBHI). IEEE, 2023 (submitted).
Paper: SA Naeini, et al. "Concurrent Validity of Automatic Speech and Pause
Measures During Passage Reading in ALS" 2022 IEEE-EMBS International Conference on Biomedical and
Health Informatics (BHI). IEEE, 2022.
Applied transformer-based computer vision techniques to perform repetition counting in
videos, achieving accurate and efficient analysis of repetitive actions or behaviors.
Fine-tuned pre-trained transformer models on the repetition counting task, leveraging
transfer learning to capture visual patterns and repetitive dynamics.
Leveraged automatic repetition counting techniques to perform temporal segmentation
and parsing of neurological assessment video data, enabling accurate analysis and understanding of
repetitive actions in the context of neurological assessments.
Paper: SA Naeini, et al. "Automated Temporal Segmentation of Orofacial Assessment
Videos." 2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI). IEEE,
2022.
Developed a web application for remote data collection during the COVID-19 pandemic,
enabling efficient and secure data gathering from remote locations.
Designed and implemented the user interface (UI) and user experience (UX) of the web
app, focusing on simplicity, intuitiveness, and accessibility to accommodate users with varying
technical expertise.
Developed secure user authentication and authorization mechanisms to protect sensitive
data and ensure proper access control for different user roles.
Provided ongoing maintenance, bug fixes, and feature enhancements to optimize the web
app's performance and address evolving user requirements.
Utilized AWS cloud services such as EC2 to store and process web application’s
content.
Integrated Dropbox API to store data on HIPPA compliant platform.
Paper: Simmatis, L., Alavi Naeini, S., et al. (2023). "Analytical Validation of a
Webcam-Based Assessment of Speech Kinematics: Digital Biomarker Evaluation following the V3
Framework." Digital Biomarkers, 7(1), 7-17.
