Portfolio item number 1
Short description of portfolio item number 1
Posts by Collection
portfolio
Portfolio item number 2
Short description of portfolio item number 2
publications
Scanning Trojaned Models Using Out-of-Distribution Samples
Published in NeurIPS, 2024
In this work, we’ve introduced TRODO, a new method for detecting backdoor attacks in deep neural networks. TRODO identifies trojans by adversarially shifting out-of-distribution (OOD) samples toward in-distribution (ID) and detecting when classifiers mistakenly classify them as ID. This approach is effective even without training data and works against adversarially trained trojaned classifiers, making it adaptable across different scenarios and datasets.
A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts
Submitted to ICLR, 2025
In this work, we designed a novelty detection method which is robust to style shifts in the data distribution. By distinguishing between core features and style features and using a teacher-student scheme, we were able to achieve state-of-the-art results on various dataset pairs.
talks
Research Assistant at RIML Lab
Published:
During my time as a Research Assistant at the Robust and Interpretable Machine Learning Lab under the supervision of Dr. Mohammad Hossein Rohban, I focused on advancing machine learning robustness and interpretability. My initial work involved conducting literature reviews and experiments on out-of-distribution (OOD) detection, particularly addressing spurious correlations. I tested various OOD benchmarks on spurious samples, analyzed the limitations of existing methods, and proposed improvements. This phase of research significantly contributed to understanding the complexities of OOD detection in real-world applications.
Remote Research Assistant at University of South Carolina
Published:
As a Remote Research Assistant at the University of South Carolina under the supervision of Prof. Pooyan Jamshidi, I conducted experiments on Centered Kernel Alignment (CKA) and its variations, such as dCKA, along with other similarity metrics. My research aimed to evaluate and compare these methods for measuring similarity between neural networks, identifying their strengths and proposing improvements to enhance their performance and interpretability.
Research Assistant at Chinese University of Hong Kong
Published:
During my time at the Chinese University of Hong Kong, I worked as a Research Assistant under the supervision of Professor Farzan Farnia. Our research focused on investigating the diversity of text-to-image generative models, specifically how they are conditioned on text descriptions. We explored differential clustering methods as a potential approach to better capture and quantify this diversity. Through this experience, I became familiar with the world of generative models and their evaluation metrics, broadening my understanding while refining my research skills in machine learning and artificial intelligence.
teaching
Advanced Programming
Undergraduate course, Sharif University of Technology, Department of CE, 2021
Data Structure and Algorithms
Undergraduate course, Sharif University of Technology, Department of CE, 2022
Advanced Programming
Undergraduate course, Sharif University of Technology, Department of CE, 2022
Head TA of Fundamental Programming
Undergraduate course, Sharif University of Technology, Department of CE, 2022
Linear Algebra
Undergraduate course, Sharif University of Technology, Department of CE, 2022
Engineering Probability and Statistics
Undergraduate course, Sharif University of Technology, Department of CE, 2022
Neumerical Methods
Undergraduate course, Sharif University of Technology, Department of CE, 2023
Linear Algebra
Undergraduate course, Sharif University of Technology, Department of CE, 2023
Machine Learning
Graduate course, Sharif University of Technology, Department of CE, 2023
Artificial Intelligence
Undergraduate course, Sharif University of Technology, Department of CE, 2023