Mahdi Zakizadeh
I'm an NLP enthusiast with more than two years of research experience in the field, and a passionate gamer with a love for 4X games. Feel free to reach out at mzakizadeh@proton.me.
-
Latest Research Blind Men and the Elephant: Diverse Perspectives on Gender Stereotypes in Benchmark Datasets
Latest research on the reliability of bias benchmarks revealing that current gender stereotype benchmarks provide a fragmented picture.
-
GitHub Project KARA Toolkit: Optimized RAG Chunking
KARA is a toolkit for efficient Retrieval-Augmented Generation (RAG) chunking. It models chunking as a DAG and uses Dijkstra's algorithm to find optimal splits, boosting retrieval and generation.
Updates
- [Sep '25]
I will be serving as a program committee member of ICLR 2026 and LREC 2026 this year.
- [Aug '25]
Our paper, Blind Men and the Elephant: Diverse Perspectives on Gender Stereotypes in Benchmark Datasets, was accepted at EMNLP 2025.
- [May '25]
Our paper, Gender Encoding Patterns in Pretrained Language Model Representations, was accepted at TrustNLP 2025.
- [Jan '25]
The draft of our paper, Blind Men and the Elephant: Diverse Perspectives on Gender Stereotypes in Benchmark Datasets, is now publicly available. In this work, we investigate why gender bias metrics often fail to align and delve into the factors contributing to these discrepancies.
- [Aug '24]
Excited to share that I will be serving as a program committee member for COLING 2025 and ICLR 2025.