cv

Bachelor

Sharif University of Technology, Tehran, Iran

Computer Engineering

High School

National Organization for Development of Exceptional Talents, Tehran, Iran

Mathematics and Physics

Internship in Interpretability and Sparsity in Deep Neural Networks

Supervised by Prof. Dan Alistarh

Institute of Science and Technology Austria (IST Austria)

In this study, we enhanced the performance of various interpretability methods by initially sparsifying the network on a selected sample, followed by the application of the interpretability method. This approach proposes a solution to the challenges presented by polysemantic neurons, which are activated by multiple distinct concepts, allowing for an in-depth investigation into the activation triggers of a neuron on a sample. This is particularly useful when the sample may pertain to a lesser known functionality of the neuron.

• Sparsity-Guided Debugging for Deep Neural Networks

Volunteer (Unpaid) Research Assistant in Robust and Interpretable Machine Learning Lab

Supervised by Prof. Mohammad Hossein Rohban

Sharif University of Technology

Initially, we identified vulnerabilities in existing Robust Out-of-Distribution (OOD) detection methods to end-to-end adversarial attacks. Subsequently, we proposed a novel OOD detection method, inspired by Generative Adversarial Network (GAN) architecture and adversarial training, which achieved state-of-the-art results in robust OOD detection.

In an other work, we explored a scenario where neurons are self-interested, aiming to increase the strength of their connection weights to subsequent layer neurons. We demonstrated that when all neurons rationally update their weights, the resulting behavior closely resembles Gradient Descent. This characteristic enables the training of deeper networks under a self-interested neurons scenario compared to previous approaches, thereby extending the applicability of this method to real-world problems.

• Robust Out-of-Distribution Detection Using GAN Architecture
• Aligning Self-Interested Neurons in Deep Neural Networks

Game Theory Course Design, Team Leader

Student-run Rastaiha education group

The team designed a two parts online course for high school students. The first part topic was 'Truthful Allocation Mechanisms Without Payments.' The second part was about collaboration. Core Value and Shapley Value.

Algorithm Instructor

Allameh Helli High School

Affiliated with the National Organization for Development of Exceptional Talents.

Algorithm Instructor

National Team camp for Iran National Olympiad in Informatics

The national team consists of the four chosen students from 10000 students. I am responsible for some graph algorithms, e.g., maximum flow.

Main Graph Theory Instructor

Summer camp for Iran National Olympiad in Informatics

40 students who are chosen from 10000 students by three exams participate in the summer camp.

Algorithm tutor

Hiva Karami & Negar Arj

Hiva has achieved a national gold medal in INOI. She was the first woman to achieve this medal in the past ten years.

World Final participation

International Collegiate Programming Contest (ICPC)

First Place

Sharif University Codejam

Silver Medal

International Olympiad in Informatics (IOI)

First Team

The 7-th Ferdowsi Collegiate Programming Contest

Bronze Medal

Info1Cup

National Gold medal (first place)

National Olympiad in Informatics

National Silver medal

National Olympiad in Informatics