Kian Shokraneh

Bachelor of Science in Computer Engineering

• CGPA: 3.96/4 (18.75/20)
• GPA over the past two semesters: 4/4 (19.25/20)

High School Diploma in Mathematics and Physics

• CGPA: 4/4 (19.33/20)

Supervisor: Mohammad Sabokrou, MLDS Unit

• Conduct research aimed at ensuring computer vision models focus on the primary object in images rather than background or irrelevant elements.
• Develop and implement adversarial approaches to improve the robustness and accuracy of models.
• Advance methods for creating more reliable and trustworthy machine learning systems.

Supervisor: Saeedeh Momtazi, Computer Engineering Department

• Exploring RAG and LLMS for Persian question answering.
• Fine-tuning XLM-RoBERTa on the PQuAD dataset to enhance accuracy and relevance in real-time responses.
• Developing a hybrid system that combines retrieval-based and generation-based models.
• Generating optimized indexes to improve retrieval efficiency.

• Advanced Programming (Java), Instructor: Dr. Fatemeh Ziaeetabar Fall 2023
• Applied Linear Algebra, Instructor: Dr. Maryam Amirmazlaghani Spring 2024
• Software Engineering, Instructor: Dr. Faezeh Gohari Fall 2024
• Software Testing, Instructor: Dr. Faezeh Gohari Fall 2024

• Specialized in backend development using Django and Django REST Framework (DRF).
• Designed project architecture and database schemas.
• Deployed and managed applications on Linux servers.
• Collaborated on frontend implementation using HTML, CSS, and Bootstrap as a secondary focus.

• Regularization and Robustness Evaluation Using SHAP GitHub
• Implemented SHAP-based regularization to enhance MNIST model robustness.
• Improved model resilience via FGSM adversarial training.
• Captum Attribution Metrics Analysis GitHub
• Conducted metric-based analysis of attribution methods using Captum on CIFAR-10.
• Evaluated robustness with infidelity and sensitivity metrics.
• Captum Interpretability Demonstrations GitHub
• Applied Grad-CAM, Integrated Gradients, and Gradient SHAP to interpret ResNet18 predictions.
• Enhanced model transparency through visual explanations.
• Background Importance Analysis Using Gradient SHAP GitHub
• Performed Gradient SHAP analysis on ResNet18 to evaluate background influence on predictions.
• Visualized attribution values to highlight decision-making processes.
• Adversarial FGSM Comparisons GitHub
• Compared FGSM across multiple adversarial attack libraries on ResNet18 (CIFAR-10).
• Analyzed Attack Success Rate (ASR) to assess model robustness.

• Developed an advanced information retrieval system with a focus on document indexing, querying, and relevance ranking.
• Implemented TF-IDF weighting and vector space models to enhance the precision and recall of document retrieval.
• Optimized search algorithms and relevance feedback mechanisms to improve retrieval effectiveness and user satisfaction.

• Implemented a fuzzy logic system for simulating self-driving car decision-making in various conditions.
• Developed and optimized fuzzy rules for obstacle avoidance and speed control.

• Developed AI using Genetic Algorithms to optimize Super Mario’s behavior and strategy.
• Evolved neural network parameters for improved navigation of complex levels.
• Implemented fitness functions to enhance adaptation and performance.

• CS 188 Pac-Man AI Projects at UC Berkeley, applying advanced AI algorithms in a gaming context.
• Implemented A*, DFS, and BFS for optimized Pac-Man navigation.
• Applied Expectimax and Minimax algorithms for adversarial agent development.
• Integrated reinforcement learning to enhance adaptability and performance.