Sepehr Rezaee

Sepehr Rezaee

Computer Science Student | AI & Machine Learning Enthusiast

Hi, I'm

Sepehr Rezaee

A passionate and driven Computer Science student at The National University of Iran, specializing in Deep Learning, Machine Learning, and AI Safety. With a solid foundation in mathematics and computer science, I have developed expertise in building robust and interpretable machine learning models, particularly for applications in scientific computing and AI safety. My academic and professional journey is marked by significant contributions to research, hands-on experience in various roles, and a strong commitment to advancing knowledge in AI and its applications.

CV Google Scholar GitHub Email LinkedIn

Publications

Research Experiences

Robust and Interpretable Machine Learning Lab

Sharif University of Technology, Tehran, Iran

Robust and Interpretable Machine Learning Lab

Focused on developing robust and interpretable machine learning models.

2024–Present
Artificial Intelligence and Scientific Computing Lab

Tehran, Iran

Artificial Intelligence and Scientific Computing Lab

Worked under the supervision of Prof. Parand on modeling disease progression using differential equations and Physics-Informed Neural Networks (PINN).

2023–Present
Institute for Research in Fundamental Sciences (IPM)

Institute for Research in Fundamental Sciences (IPM), Tehran, Iran

Deep Learning and Neuroscience Intern Researcher

Conducted M/EEG data analysis using advanced deep-learning techniques under the supervision of Dr. Rezvani.

2023–2024

Projects

Physics-Informed Neural Networks

Artificial Intelligence

Physics-Informed Neural Networks for Disease Progression Modeling

Developed a PINN model to predict disease progression using differential equations, contributing to advancements in AI-driven healthcare.

M/EEG Data Analysis

Artificial Intelligence

Machine Learning for M/EEG Data Analysis

Applied deep learning techniques to analyze M/EEG data, advancing research in computational neuroscience.

AI Model Security

Artificial Intelligence

AI Model Security

Contributed to research on backdooring and scanning trojaned models, resulting in multiple publications and submissions to NeurIPS 2024.

Fashion MNIST Classification

Neural Networks

Fashion MNIST Classification

Implemented a PyTorch-based model for classifying Fashion MNIST dataset using multi-layer perceptrons, focusing on dropout, batch normalization, and regularization.

CIFAR-10 Classification

Neural Networks

CNN for CIFAR-10 Classification

Developed a CNN for CIFAR-10 image classification, optimizing architecture with pooling layers and batch normalization for maximum accuracy.

EfficientNet, ResNext, and Inception-ResNet

Neural Networks

Exploring EfficientNet, ResNext, and Inception-ResNet

Explored the EfficientNet, ResNext, and Inception-ResNet architectures for image classification on the Intel Image Classification dataset, incorporating transfer learning techniques.

Variational Autoencoders

Neural Networks

Variational Autoencoders for Image Generation

Conducted a study on Variational Autoencoders (VAE) and implemented a network capable of generating CIFAR-10 images using novel architectures.

Persian Language Model

Neural Networks

Persian Language Model using RNNs/LSTM

Created a Persian language model using RNNs/LSTM cells, trained on the Persian Wikipedia dataset for sequence prediction, achieving optimized perplexity scores.

Neuron Model Simulation

Computational Neuroscience

Neuron Model Simulation

Implemented a neuron model to simulate the effects of a neurological disease and a drug treatment using PyTorch. This project also involved analyzing the role of the myelin sheath and the function of auditory hair cells.

Neuron Behavior Replication

Computational Neuroscience

Neuron Behavior Replication

Developed and evaluated neuron models that replicate real neuron behavior in response to various input currents, including predicting neuron spikes and analyzing different connectivity patterns between neurons using PyTorch and SynapticFlow.

Population Behavior Simulation

Computational Neuroscience

Population Behavior Simulation

Simulated the behavior of neuron populations with different connectivity patterns under various input stimuli, creating raster plots, analyzing the interaction between inhibitory and excitatory populations, and designing encoding schemes for neural signals.

Spiking Neural Network

Computational Neuroscience

Spiking Neural Network for Digit Recognition

Designed and trained a spiking neural network for digit recognition using reward-modulated spike-timing-dependent plasticity (STDP). The network was trained to recognize binary digits and output their decimal equivalents.

Facial Feature Manipulation

Image Processing

Facial Feature Manipulation

Developed a program using Python libraries (e.g., OpenCV, skimage) to apply linear and nonlinear transformations to detect and manipulate facial features in images, such as creating caricatures.

Image Index Compression

Image Processing

Image Index Compression using K-means

Implemented a function for image index compression using K-means clustering to reduce color space, with performance comparison across different color spaces.

Edge Detection and Morphological Operations

Image Processing

Edge Detection and Morphological Operations

Designed and implemented algorithms for edge detection and morphological operations in fingerprint image classification, working with different classes of fingerprint patterns.

Texture Synthesis and Hole Filling

Image Processing

Texture Synthesis and Hole Filling

Built a system for texture synthesis and hole filling in images, focusing on synthesizing textures and filling missing image data using advanced techniques.

JPEG Compression Techniques

Image Processing

JPEG Compression Techniques

Worked on JPEG compression techniques, including parameter tuning for compression levels and comparing the impact on image quality before and after compression.