Publications

Amirhossein Dabiriaghdam's peer-reviewed publications and preprints.

Check out my Google Scholar profile for a complete list of publications.

Peer-reviewed Publications

2025

  1. EMNLP 2025
    SimMark: A robust sentence-level similarity-based watermarking algorithm for large language models
    Amirhossein Dabiriaghdam, and Lele Wang
    EMNLP Main, 2025 Oral (325/8174 = 3.98%)

    Introduces a black-box sentence-level watermark based on semantic similarity that remains detectable after paraphrasing while preserving text quality.

  2. EMNLP 2025
    SOI Matters: Analyzing Multi-Setting Training Dynamics in Pretrained Language Models via Subsets of Interest
    Shayan Vassef*, Amirhossein Dabiriaghdam*, Mohammadreza Bakhtiari*, and Yadollah Yaghoobzadeh (* Equal contribution)
    In Proceedings of the 5th Workshop on Multilingual Representation Learning (MRL 2025) Nov, 2025 · Co-located with EMNLP 2025

    Introduces Subsets of Interest to track example-level learning dynamics across multi-task, multilingual, and multi-source training and guide targeted fine-tuning.

2024

  1. NAACL 2024
    BcAmirs at semeval-2024 task 4: Beyond words: A multimodal and multilingual exploration of persuasion in memes
    Amirhossein Abaskohi*, Amirhossein Dabiriaghdam*, Lele Wang, and Giuseppe Carenini (* Equal contribution)
    In Proceedings of the 18th International Workshop on Semantic Evaluation (SemEval-2024), 2024 · Co-located with NAACL 2024

    Combines generated image captions with RoBERTa text and CLIP image representations to classify persuasive techniques in multilingual memes.

    Presentation video ↗

2023

  1. ICASSP 2023
    Targeted Adversarial Attacks Against Neural Machine Translation
    Sahar Sadrizadeh, Amirhossein Dabiriaghdam, Ljiljana Dolamic, and Pascal Frossard
    ICASSP, 2023

    Introduces a targeted white-box attack that inserts chosen keywords into neural machine translations while preserving source-sentence similarity.

Preprints

2026

  1. arXiv
    UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs
    Amirhossein Abaskohi*, Amirhossein Dabiriaghdam*, Liang Luo, Ellie Dingqiao Wen, Lele Wang, Giuseppe Carenini, and Peter West (* Equal contribution)
    arXiv preprint arXiv:2606.06622, 2026

    Introduces a benchmark and KS@N metric for testing whether language models reproduce target probability distributions when sampled repeatedly.

  2. arXiv
    VAMPS: Visual-Assisted Mathematical Problem Solving Benchmark
    Amirhossein Dabiriaghdam, Shayan Vassef, Mohammadreza Bakhtiari, Yasamin Medghalchi, Ilker Hacihaliloglu, Mesrob Ohannessian, Lele Wang, and Giuseppe Carenini
    arXiv preprint arXiv:2606.04244, 2026

    Introduces a bilingual multimodal benchmark for studying whether vision-language models can use graphs and visual tools to solve mathematical problems.

  3. arXiv
    When Minor Edits Matter: LLM-Driven Prompt Attack for Medical VLM Robustness in Ultrasound
    Yasamin Medghalchi, Milad Yazdani, Amirhossein Dabiriaghdam, Moein Heidari, Mojan Izadkhah, Zahra Kavian, Giuseppe Carenini, Lele Wang, Dena Shahriari, and Ilker Hacihaliloglu
    arXiv preprint arXiv:2603.21047, 2026

    Evaluates how realistic prompt rewrites, typos, shorthand, and ambiguity expose robustness gaps in medical vision-language models for ultrasound question answering.

Master's Thesis

  1. M.A.Sc.
    Combating Disinformation in the Age of Generative AI: From Watermarking LLMs to Persuasion Analysis of Memes
    Amirhossein Dabiriaghdam
    Master's thesis, The University of British Columbia, 2025