Mohammadreza Samadi

My name is Mohammadreza, and I am an AI Researcher at Huawei Canada, specializing in optimizing Generative AI models—such as Large Language Models (LLMs) and Diffusion Models—for efficient deployment. Passionate about making AI accessible and efficient, I focus on developing advanced technologies that are lightweight and fast for real-world applications.

Before joining Huawei, I worked as a Machine Learning intern at the Alberta Machine Intelligence Institute (Amii), where I created customized AI solutions to meet the unique needs of various companies.

I hold an M.Sc. in Computing Science from the University of Alberta, where I was advised by Prof. Davood Rafiei. My research focused on Natural Language Processing and Information Retrieval, particularly on query performance prediction in multi-hop question answering. Prior to that, I completed my Bachelor’s in Computer Engineering at Amirkabir University of Technology (Tehran Polytechnic), where I developed a fake news detection system for Persian news under the guidance of Prof. Saeede Momtazi.

Selected Publications

2025

FunEditor: Achieving Complex Image Edits via Function Aggregation with Diffusion Models
Mohammadreza Samadi, Fred X Han, Mohammad Salameh, Hao Wu, and 3 more authors

The 39th Annual AAAI Conference on Artificial Intelligence

Abstract arXiv DOI Project

Diffusion models have shown exceptional performance in generative tasks, making them suitable for image editing. Recent research highlights their ability to apply edits based on textual instructions, but two key challenges persist. First, these models struggle to perform multiple edits simultaneously, causing computational inefficiencies due to sequential processing. Second, using textual prompts to specify editing regions can result in unintended changes to the image. We introduce FunEditor, an efficient diffusion model designed to learn basic editing functions and combine them to perform complex edits. This method allows tasks like object movement by merging multiple functions and applying them simultaneously to designated areas. Our experiments show that FunEditor significantly outperforms recent inference-time optimization techniques and fine-tuned models, both quantitatively across various metrics and through visual comparisons, on complex tasks like object movement and object pasting. Moreover, with only four inference steps, FunEditor achieves a 5-24x speedup compared to existing popular methods. Codes and datasets are available online.
2025

PixelMan: Consistent Object Editing with Diffusion Models via Pixel Manipulation and Generation
Liyao Jiang, Negar Hassanpour, Mohammad Salameh, Mohammadreza Samadi, and 3 more authors

The 39th Annual AAAI Conference on Artificial Intelligence

Abstract arXiv DOI

Recent research explores the potential of Diffusion Models (DMs) for consistent object editing, which aims to modify object position, size, and composition, etc., while preserving the consistency of objects and background without changing their texture and attributes. Current inference-time methods often rely on DDIM inversion, which inherently compromises efficiency and the achievable consistency of edited images. Recent methods also utilize energy guidance which iteratively updates the predicted noise and can drive the latents away from the original image, resulting in distortions. In this paper, we propose PixelMan, an inversion-free and training-free method for achieving consistent object editing via Pixel Manipulation and generation, where we directly create a duplicate copy of the source object at target location in the pixel space, and introduce an efficient sampling approach to iteratively harmonize the manipulated object into the target location and inpaint its original location, while ensuring image consistency by anchoring the edited image to be generated to the pixel-manipulated image as well as by introducing various consistency-preserving optimization techniques during inference. Experimental evaluations based on benchmark datasets as well as extensive visual comparisons show that in as few as 16 inference steps, PixelMan outperforms a range of state-of-the-art training-based and training-free methods (usually requiring 50 steps) on multiple consistent object editing tasks.
2023

Performance Prediction for Multi-hop Questions
Mohammadreza Samadi and Davood Rafiei

arXiv preprint arXiv:2308.06431

Abstract arXiv

We study the problem of Query Performance Prediction (QPP) for open-domain multi-hop Question Answering (QA), where the task is to estimate the difficulty of evaluating a multi-hop question over a corpus. Despite the extensive research on predicting the performance of ad-hoc and QA retrieval models, there has been a lack of study on the estimation of the difficulty of multi-hop questions. The problem is challenging due to the multi-step nature of the retrieval process, potential dependency of the steps and the reasoning involved. To tackle this challenge, we propose multHP, a novel pre-retrieval method for predicting the performance of open-domain multi-hop questions. Our extensive evaluation on the largest multi-hop QA dataset using several modern QA systems shows that the proposed model is a strong predictor of the performance, outperforming traditional single-hop QPP models. Additionally, we demonstrate that our approach can be effectively used to optimize the parameters of QA systems, such as the number of documents to be retrieved, resulting in improved overall retrieval performance.
2021

Deep contextualized text representation and learning for fake news detection
Mohammadreza Samadi, Maryam Mousavian, and Saeedeh Momtazi

Information Processing & Management

Abstract DOI

In recent years, due to the widespread use of social media and broadcasting agencies around the world, people are extremely exposed to being affected by false information and fake news, all of which have negative impacts on both collective thoughts and governments’ policies. In recent years, the great success of pre-trained models for embedding contextual information from texts motivates researchers to utilize these embeddings in different natural language processing tasks. However, in a complex task like fake news detection, it is not determined which contextualized embedding can assist the classifier with more valuable features. Due to the lack of a comparative study about utilizing different contextualized pre-trained models besides distinct neural classifiers, we aim to dive into a comparative study about using different classifiers and embedding models. In this paper, we propose three classifiers with different pre-trained models for embedding input news articles. We connect Single-Layer Perceptron (SLP), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN) after the embedding layer which consists of novel pre-trained models such as BERT, RoBERTa, GPT2, and Funnel Transformer in order to benefit from deep contextualized representation provided by those models as well as deep neural classifications. We evaluate our proposed models on three well-known fake news datasets: LIAR (Wang, 2017), ISOT (Ahmed et al., 2017), and COVID-19 Patwa et al. (2020). The results on these three datasets show the superiority of our proposed models for fake news detection compared to the state-of-the-art models. The results show 7% and 0.1% improvements in classification accuracy compared to the proposed model by Goldani et al. (2021) on LIAR and ISOT, respectively. We also achieved 1% improvement compared to the proposed model by Shifath et al. (2021) on the COVID-19 dataset.