Amirhossein Souri
Research in machine learning and intelligent systems, alongside software engineering that connects ideas with practical applications.
About Me
Research and engineering at Sharif University of Technology
I'm a Computer Science and Engineering student at Sharif University of Technology and a research assistant at RIML Lab. I currently study visual reasoning in vision-language models, from question-conditioned visual grounding to evaluating whether model answers respond to visual evidence.
My interests span machine and deep learning, computer vision and multimodal learning, and language and vision-language models. I'm also interested in ML systems, generative models, and trustworthy ML. These interests extend to reinforcement learning, multi-agent systems, and robotics. I also enjoy building software systems; at Hamravesh, I worked across APIs, interfaces, and Kubernetes integration for a cloud application marketplace.
Experience
Technical, research, and teaching experience
Research
Research Assistant
RIML Lab, Sharif University of Technology · Visual Reasoning
- PI: Dr. Mohammad Hossein Rohban.
- Developed the project's initial question-conditioned visual-grounding study and evaluated its effect on downstream VQA with Qwen2.5-VL-3B-Instruct.
- Investigating training and evaluation methods for evidence-sensitive reasoning in vision-language models.
Research
Research Assistant
RIML Lab, Sharif University of Technology · Diffusion Model Unlearning
- PI: Dr. Mohammad Hossein Rohban.
- Investigated noise-robust concept unlearning for text-to-image diffusion models under adversarial and distributional variation.
- Contributed to adaptive noise sampling and experimental evaluation.
Technical
Software Engineer
Hamravesh
- Developed and deployed a marketplace service for one-click cloud applications.
- Implemented backend APIs and frontend interfaces for the marketplace.
- Integrated the application launch workflow with Kubernetes.
Artificial Intelligence
SUT • Dr. Mahdieh Soleymani
Fall 2025
Artificial Intelligence
SUT • Mr. Samiei • Mr. Fereydooni
Spring 2025
Machine Learning
SUT • Dr. Abolfazl Motahari
Spring 2025
Probability and Statistics
SUT • Dr. Amir Najafi
Spring 2024
Advanced Programming (Java)
SUT • Dr. MohammadAmin Fazli
Spring 2024 • Spring 2026
Fundamentals of Programming (Python)
SUT • Mr. Kazemi
Spring 2024
Fundamentals of Programming (Python)
Sharif MicroMaster • Dr. Hamid Zarrabi-Zadeh
Winter 2024 • Summer 2024 • Winter 2025 • Summer 2025 • Winter 2026
Programming for Data Analysis
Sharif MicroMaster • Dr. AmirMahdi Sadeghzadeh
Winter 2024
Data Structures and Algorithms
Sharif MicroMaster • Dr. MohammadAli Abam
Winter 2024
Fundamentals of Programming (C)
SUT • Dr. MohammadAmin Fazli
Fall 2023 • Fall 2024 • Fall 2025
Fundamentals of Programming (Python)
SUT • Mr. Malekzadeh
Fall 2023
Machine Learning Course Materials
Introductory slides and notebooks on machine learning fundamentals. Materials in Persian.
Git Workshop
Workshop slides covering version control and collaboration, with a Persian recording.
Preprint
Research shared on arXiv and currently under review
Weeding Out Bad Seeds: Initial-Noise-Robust Unlearning for Text-to-Image Diffusion Models
Studies how unlearned concepts re-emerge under different noise initializations in text-to-image diffusion models. Adaptive, concept-conditioned noise sampling improves the robustness of concept erasure.
Projects
Selected research, software, and coursework projects
Question-Conditioned Visual Grounding
Trained an attention-based model on CLIP patch and question embeddings to localize answer-relevant image regions. Converted predicted heatmaps into bounding boxes and evaluated their effect on downstream VQA with Qwen2.5-VL-3B-Instruct.
Modern Information Retrieval Project
Built a Goodreads search engine with BM25 and ranking evaluation, plus hybrid multimodal product search with dense and sparse retrieval and reranking.
Deep Learning Practical Assignments
Completed 15 notebooks spanning neural networks, LoRA/QLoRA, RAG, generative models, and self-supervised vision methods.
Machine Unlearning and Robustness Exercises
Implemented class unlearning in a conditional VAE using Fisher information and evaluated recovery attacks.
Hardwar Website
Built the frontend, backend, and infrastructure for the Hardwar event website at SUT.
BugsBuzzy Website
Contributed to the backend of the BugsBuzzy event website.
Education
My academic journey
B.Sc. in Computer Science and Engineering
Sharif University of Technology
GPA: 18.99/20
Diploma of Mathematics and Physics
Shahid Beheshti High School
Skills & Technologies
A concise readout of the stacks, tools, and skills I rely on
Machine Learning & GenAI
Machine Learning & GenAI: 15 skills
Programming Languages
Programming Languages: 8 skillsData Science
Data Science: 9 skills
DevOps & Infrastructure
DevOps & Infrastructure: 11 skills
Backend Development
Backend Development: 6 skills
Frontend Development
Frontend Development: 8 skills
Other Technical
Other Technical: 3 skillsAcademic Service & Leadership
Voluntary contributions to the academic community
Public Relations Officer
Student's Scientific Chapter • CE, SUT
Senior Editor
Byte Publication • CE, SUT
President of CE Department • University Media Lead
Student's Guild Council • SUT
Technical Staff
BugsBuzzy • CE, SUT
Technical & Graphical Design Staff
Emeet • EE, SUT
Vice President & Technical Lead
Hardwar • CE, SUT
Technical Staff
ICPC • CE, SUT
Executive Staff
CodoCodile • CE, SUT
Social Media Lead
Rayan AI Contest • CE, SUT
Marketing & Executive Staff
S4 • CE, SUT
Content Lead & Social Media Staff
WSS • CE, SUT
Social Media Staff
CodoCodile • CE, SUT
Executive Staff
ICPC • CE, SUT
Get in Touch
Feel free to reach out for collaborations and questions