Learn more about me
I’m a Master’s student in Computer Software Engineering at Sharif University of Technology. Alongside my studies, I’m actively involved in research as a Research Assistant at the RIML lab, where I focus on Compositional Visual Generation problems in Machine Learning. This allows me to blend theory with real-world applications, pushing the boundaries of what AI can create. My research centers around teaching machines to generate images by combining different elements in creative ways—a field that merges AI, visual computation, and problem-solving. I’m passionate about finding innovative solutions and advancing what AI can achieve. I love collaborating on projects that challenge me to think differently, and I’m always open to connecting with others who share an interest in machine learning, AI, or tech in general.
Projects
Licenses & Certifications
Skills
Languages
Experiences
Interests
My Education
Sharif University of Technology
GPA: 4
Grade: 18.68
University of Guilan
GPA: 3.98
Grade: 19.58
Dr. Moein
Field: Mathematics & Physics
GPA: 4
Grade: 19.94
Begher Al'Olum
GPA: 4
Grade: 20
My Honors & Awards
My Publications
Seyed Amir Kasaei, Ali Aghayari, Arash Marioriyad, Niki Sepasian, Shayan Baghayi Nejad, MohammadAmin Fazli, Mahdieh Soleymani Baghshah, Mohammad Hossein Rohban
Accepted at Transactions on Machine Learning Research (TMLR), 2026.
A reward-guided optimization framework for initial noise in diffusion models, combining discrete exploration and
continuous refinement for compositional alignment.
Arian Komaei Koma, Seyed Amir Kasaei, Ali Aghayari, Aida Aryafar, Matin Ghiasi, Amirhossein Souri, Mohammad Mosayyebi, AmirMahdi Sadeghzadeh, Mohammad Hossein Rohban
Under review at ICML 2026.
Adaptive noise sampling for unlearning in text-to-image diffusion models.
Arian Komaei Koma, Seyed Amir Kasaei, Amirhossein Arefzadeh, Roham Izadidoost, AmirMahdi Sadeghzadeh, Mohammad Hossein Rohban
Under review at CVPR 2026.
A black-box embedding-aware adversarial prompting framework for attacking unlearned text-to-image diffusion models.
Seyed Amir Kasaei, Ali Aghayari, Arash Marioriyad, Niki Sepasian, MohammadAmin Fazli, Mahdieh Soleymani Baghshah, Mohammad Hossein Rohban
NeurIPS 2025 Workshop GenProCC: Generative and Protective AI for Content Creation.
A correlation study of compositional text-to-image metrics, benchmarking category-specific and cross-category
agreement with human evaluations.
Seyed Amir Kasaei, Mohammad Hossein Rohban
NeurIPS 2025 Workshop GenProCC: Generative and Protective AI for Content Creation.
Interprets hallucination as an upper bound for assessing semantic faithfulness in text-to-image systems.
Arian Komaei Koma, Seyed Amir Kasaei, Ali Aghayari, AmirMahdi Sadeghzadeh, Mohammad Hossein Rohban
Under review at ICLR 2026 Workshop Trustworthy AI.
Evaluating unlearned diffusion models through the lens of compositionality.
Seyed Amir Kasaei, Arash Marioriyad, Mahbod Khaleti, MohammadAmin Fazli, Mahdieh Soleymani Baghshah, Mohammad Hossein Rohban
Under review at ICLR 2026 Workshop HCAIR.
RebusBench is a benchmark of 1,164 rebus puzzles for evaluating abstract cognitive visual reasoning in LVLMs.
My Skills
My Experiences
RIML Lab (Sharif Univeristy of Technology)
Supervisor: Dr M H Rohban
Skills: Machine Learning . Deep Learning . NLP . Computer Vision . Compositional Generation . Medical Image Analysis
Univeristy of Guilan
Supervisor: Dr S A Mirroshandel
Skills: Machine Learning . Deep Learning . NLP . Computer Vision . Machine Translation . Medical Image Analysis
Instructor: Dr MA Fazli
Skills: Deep Learning · Data Anlaysis · Data Mining
Instructor: Dr M H Rohban, Dr M Soleymani
Skills: Compositional Generation · Inferenece-Time Optimization · LLM · Reasoning
Instructor: Dr M Soleymani
Skills: Optimization · Diffusion · Image Generation
Instructor: Dr M H Rohban
Skills: Deep Learning · Medical Image Analysis · Trustworthy . Interpretability
My Certificates
