Research Scientist · PhD in Computer Vision
Mohamed Amine
Mohamed Amine
Kerkouri
Computer Vision / Visual Attention / LLM Agents
I build practical AI systems that connect academic research with the real world, from modeling human visual attention and medical imaging to retrieval-augmented LLM agents. Currently Research Scientist at F-Initiatives and AIOps Architect at neophi.ai.
Recent
News & highlights
2025
Work on knee-osteoarthritis grading and doctors' eye movements presented at EUSIPCO and ETRA 2025.
2024
Joined F-Initiatives as a Research Scientist and neophi.ai as an AIOps Architect.
Selected work
Research that defines my path
A few papers that best represent my contributions across attention, medical imaging, and language models.
SATSal: A Multi-Level Self-Attention Based Architecture for Visual Saliency Prediction
Automatic Diagnosis of Knee Osteoarthritis Severity Using Swin Transformer
SalyPath: A Deep-Based Architecture for Visual Attention Prediction
Focus areas
What I work on
Computer Vision
Deep learning for visual understanding, saliency, and image analysis.
Visual Attention
Modeling human gaze: saliency and scanpath prediction for images and art.
Medical Imaging
Diagnosis automation and quality assessment, e.g. knee osteoarthritis grading.
LLMs & NLP
Hallucination analysis, small language models, and multimodal reasoning.
AI Agents
Retrieval-augmented, tool-using agents and agentic workflows for real tasks.
AI for Science
Applying AI to accelerate scientific discovery, analysis, and evaluation.
Open source
Code & datasets
Reproducible research and tooling, released on GitHub.
SSLArtScanpath
5
Self-supervised scanpath prediction on paintings using a Barlow Twins objective, learning gaze dynamics without labeled fixations.
FDISS
Foveal Disc-IoU Scanpath Score: a perceptually grounded metric for comparing predicted and human scanpaths.
Enlighten-AI
A retrieval-augmented life-coaching assistant built on curated Dr. K (HealthyGamer) transcripts.
Writing
From the blog
June 27, 2026
ThinkProbe: Beyond Accuracy, Structural Profiling of LLM Reasoning Traces
When we evaluate a reasoning model, we almost always look at one thing: was the final answer right? In ThinkProbe, my collaborators...
May 27, 2026
Closing the Foveal Gap: Perceptually Grounded Scanpath Comparison with Disc IoU
Fixations are usually treated as exact points on an image, but human vision does not work that way. Each fixation gathers information...
May 26, 2026
What They Saw, Not Just Where They Looked: Semantic Scanpath Similarity via VLMs and NLP Metrics
Classic scanpath metrics tell us where two people looked and how well their gaze paths overlap in space. In this work, presented...