Hadi Niknam

Department of Computer Science

Tehran, Iran

niknam@aut.ac.ir

Welcome to my personal website! I am Mohammad Hadi Niknam, an undergraduate Computer Science student at Amirkabir University of Technology (Tehran Polytechnic). My academic goal is to build theoretically grounded, reliable machine learning systems at the intersection of reinforcement learning, control theory, and high-impact medical applications.

Currently, my research focuses on:

Reinforcement Learning (RL): RL Theory, LLM Alignment & Preference Modeling, Decision Making Under Uncertainty, and Multi-Agent RL.
Medical Robotics: Robot learning, human-adaptive compliance control, and ultrasound-guided robotic arms.
Medical AI: Medical imaging, foundation models for healthcare, and clinically robust decision-making.

For my BSc Thesis, I developed HAS-RL: Human-Adaptive Safe Reinforcement Learning for Clinician-Configured Force Safety in Robotic Ultrasound Spinal Scanning (see the bilingual thesis post or download the research paper and AUT defense thesis).

Current Research Engagements

Dr. Javad Alirezaie Lab (Toronto Metropolitan University, Canada — Remote): Serving as Research Project Coordinator and Research Assistant. Leading undergraduate research on lightweight super-resolution architectures, neural architecture search via reinforcement learning (SRIQL, URLSR), and low-precision network quantization (QuantaSR).
Shafa Yahyaian Hospital, Iran University of Medical Sciences: Research Assistant working on fairness-aware operating-room scheduling under capacity constraints, AI diagnostics for osteonecrosis of the jaw, and foundation model adaptation (MedSAM/BTXRD).
CAMP Lab, Technical University of Munich (TUM, Germany — Remote): Collaborating with Dr. Nassir Navab on ultrasound foundation models.
Independent RL Research: Developing adaptive tail-weighting policy gradient formulations (TailRL) for return-percentile credit assignment.

Feel free to explore my Curriculum Vitae, view my publications and projects, or reach out to me via email. You can also download my latest resume in PDF format or view the live version on Overleaf.

latest posts

Oct 02, 2026	HAS-RL: Clinician-Configured Force Safety in Robotic Ultrasound (BSc Thesis)
Apr 15, 2024	a post with pseudo code

selected publications

ICASSP

SRIQL: Implicit Q Learning for Ultralight-weight Image Super Resolution

Farzan Niknejad, Mohammad Hadi Niknam, and Javad Alirezaie

In IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2026

Abs Code

Designed a reinforcement learning-based neural architecture search over lightweight super-resolution blocks using Implicit Q-Learning (IQL). Achieved state-of-the-art (SOTA) performance across benchmark datasets (DIV2K, Set14, Set5, Urban100) with a 1 dB PSNR gain over existing lightweight baselines.
BMC

URLSR: Efficient Image Super Resolution Network using Neural Operations and Reinforcement Learning

Farzan Niknejad, Mohammad Hadi Niknam, and Javad Alirezaie

BMC Medical Informatics and Decision Making, 2026

Under Review

Abs

Proposes URLSR, an efficient super-resolution architecture using neural operations and reinforcement learning designed for high-fidelity, resource-constrained medical image enhancement.
Under Review

QuantaSR: Efficient Image Super-Resolution through Quantization Error Optimization in Low-Precision Neural Networks

Farzan Niknejad, Mohammad Hadi Niknam, and Javad Alirezaie

2026

Under Review

Abs

Investigates quantization error optimization strategies in ultra-low precision neural networks to achieve efficient, accurate image super-resolution on edge computing platforms.
Dataset

Arena Human Preference 90K: Engineered Features and Correctness Annotations

Nazanin Mirzaei, Parsa Naderi, Mohammad Hadi Niknam, and 6 more authors

2026

Abs Website

A derived release built on lmarena-ai/arena-human-preference-140k ( 90,000 battles) providing interpretable structural and linguistic features across conversation turns and independent human correctness annotations for LLM evaluation and human preference modeling.
BMC

Fairness-Aware Operating-Room Sequencing for Acute Orthopaedic Trauma: A Retrospective Capacity-Calibrated Stress Test Using MOVER

Mohammad Hadi Niknam, Saeed Vosoughi, Ghazaleh Shakiba Maram, and 4 more authors

BMC Medical Informatics and Decision Making, 2026

Under Review

Abs

Introduces a fairness-aware scheduling framework for operating-room sequencing in acute orthopaedic trauma cases, validated through retrospective capacity-calibrated stress testing on the MOVER dataset.
BSc Thesis
HAS-RL: Human-Adaptive Safe Reinforcement Learning for Clinician-Configured Force Safety in Robotic Ultrasound Spinal Scanning

Mohammad Hadi Niknam

Amirkabir University of Technology (Tehran Polytechnic), 2026

Abs Bib PDF Supp Slides Website

BSc Thesis introducing HAS-RL, an operator-conditioned safe reinforcement learning framework for robotic ultrasound spinal scanning. Delivers zero-shot force safety under dynamic runtime limits, the Smoothed Bandit-Lagrangian (SBL) multiplier method, anticipatory trajectory tracking, and two-sided acoustic window constraints.
@thesis{thesis2026, title = {HAS-RL: Human-Adaptive Safe Reinforcement Learning for Clinician-Configured Force Safety in Robotic Ultrasound Spinal Scanning}, author = {Niknam, Mohammad Hadi}, school = {Amirkabir University of Technology (Tehran Polytechnic)}, year = {2026}, }
Project

SuperGradio: A general API for all huggingface Spaces

Mohammad Hadi Niknam

2025

Abs Code Website

SuperGradio is a lightweight framework that enables developers to connect, customize, and deploy Hugging Face Spaces with minimal configuration. It drastically reduces setup time and supports real-time inference, model chaining, and UI customization out of the box. Built for researchers and engineers who want to prototype AI demos faster.