About me
I'm a final-year Electrical Engineering student and a Research Assistant at Mila (Quebec AI Institute), where I focus on NeuroAI. My current research explores the geometry of object representations in recurrent neural networks, with the goal of understanding how these networks retain and process information across multiple cognitive tasks.
Previously, I worked at CNRS (National Centre for Scientific Research), where I helped develop a novel convolution method known as Dilated Convolution with Learnable Spacings (DCLS). This work aimed to improve the flexibility of convolutional neural networks through gradient-based learning techniques.
I have also contributed to research at the RIML Lab at Sharif University of Technology, where I focused on the interpretability of histopathology images, adversarial attacks, and model robustness.
Outside of academia, I have experience working as a Machine Learning Engineer, applying AI to real-world challenges. I also founded Neura, a community that fosters collaboration between neuroscience and AI students, promoting interdisciplinary learning and research.
Education
- Machine Learning 20/20
- Neuroscience 20/20
- Natural Language Processing 19.4/20
- Artificial Intelligence 19.3/20
- Linear Algebra 20/20
- Probability and Statistics 18.9/20
Studied courses such as Visual Neuroscience, Deep Learning, Fundamentals of Neuroscience, Biomedical Engineering, and other related subjects.
After passing the entrance exam for public schools, I studied math, physics, and programming.
I was also selected to join the student council, where we addressed challenges to improve the school’s educational environment.
Experience
Research assistant 2024-Present
Research Assistant 2024
CNRS (National Centre for Scientific Research)
At CNRS, I contributed to the development of Dilated Convolution with Learnable Spacings (DCLS), a novel method that enhances the flexibility of dilated convolutions in deep neural networks. By learning kernel positions through gradient-based optimization, DCLS improves feature extraction across varying scales, making it a valuable tool in tasks like image classification and object detection. I focused on implementing this method in PyTorch, helping to advance CNN architecture for more adaptable deep learning models.
Research Assistant 2022-2023
RIML Lab - Sharif University of Technology (not paid and/or contract-based)
- In my research, I have delved deeply into the intricate domain of adversarial attacks, focusing on the development and implementation of methods to identify and counteract such threats within the realm of computer vision.
- Prior to this, I worked on the interpretability of histopathology images, where I employed various neural network models and vision transformers to enhance our understanding of images. These experiences have not only advanced my technical skills but also deepened my commitment to contributing meaningful advancements in the field
Teaching Assistant 2023-Present
Sharif University of Technology (not paid and/or contract-based)
- Natural Language Processing
- Artificial Intelligence
- Linear Algebra
- Artificial Intelligence and Biological Computing
- Fundamentals of Python Programming
Research and Development 2022-2024
Sharif Entrepreneurship Center
In the Research and Development team at the Entrepreneurship Center of Sharif University of Technology, I drive initiatives to cultivate entrepreneurial and managerial acumen, crafting programs and events tailored for aspiring entrepreneurs. By meticulously assessing conditions and audience needs, I devise innovative strategies, charting pathways for feasible and impactful endeavors. This involves not only idea generation but also strategic financial planning, analyzing national budgets to ensure resource allocation for staffing and implementation. My goal is to create an environment conducive to startup success.
Machine Learning Engineer 2022
ToobaTech
I worked as a junior machine learning engineer in ToobaTech. The job requirements included a deep understanding of machine learning and deep learning with a focus on big data and computer vision. I’ve experienced working with different machine learning frameworks such as PyTorch and Scikit-Learn I’ve also worked with different tools such as Apache Spark and MySQL.
Research
My Research List
| Mechanism-aware Self-supervised Learning Conducted research on Mechanism-aware Self-supervised Learning, leveraging the SimCLR base model. Implemented diverse photo augmentations coupled with embedded metadata, such as rotation angles or color variations, to enhance the model’s awareness of image transformations. This approach aimed to improve the model’s ability to accurately represent and understand different types of photo alterations for enhanced learning outcomes. |
| Vectorized NFGSM Developed and implemented a vectorized approach for the NFGSM defense mechanism to combat adversarial attacks on deep learning models within the PyTorch framework. Achieved superior performance compared to the original NFGSM implementation when subjected to PGD and FGSM attacks. This project involved advanced techniques in optimizing defense strategies against adversarial attacks, showcasing adeptness in enhancing model robustness within the realm of deep learning security. |
Experience as a volunteer
| 2022 | Member of Artificial Intelligence in Medical Sciences Associatio — University of Tehran |
| 2024 | Scientific Staff of Winter Seminar Series (WSS) — Sharif University of Technology |
| 2022 | Scientific staff of Artificial Intelligenc — Summer school of Sharif University of Technology (Rasta) |
| 2023 | Head of Social and Content Team of ASM Masters — Sharif University of Technology |
| 2023 | Social and Content Staff of Winter Seminar Series — Sharif University of Technology |
| 2023 | Social Media Manager — Byte Publication, Sharif University of Technology |
Selected Projects
Movie Retriever
The Movie Retriever is a Python-based machine learning system using NLTK for content-based filtering and TF-IDF vectorization to recommend movies based on user preferences.
Visual Neuroscience Project
This project examines how learning impacts decision-making efficiency and accuracy in a perceptual task by using the Wong neural decision model to simulate reaction times and accuracy changes across learning phases.
GAN-BERT
The project explores Transformers, Semi-Supervised Learning, and GAN-based labeling, using BERT embeddings for enhanced bidirectional semantics in classification, with data limitations to apply a data-scarcity technique.
Selected Courses
Deep Learning
Developed deep learning models, including ID3-based Decision Trees, PCA on MNIST, SVM for heart disease prediction, and Neural Networks with Backpropagation, leveraging PyTorch for CUDA parallelization and applying the Forward Forward Algorithm for image classification.
Natural Language Processing
Developed advanced text analysis solutions using Hazm and Dadmatools for keyword extraction, natural language task management, fake news detection, and news summarization with classification and text generation.
Artificial Intelligence
Developed solutions for AI course assignments at Sharif University, covering search algorithms, constraint satisfaction, and machine learning.