MSc Informatics @ TUM • Generative Vision Researcher

I am a Machine Learning researcher currently pursuing my MSc in Informatics at the Technical University of Munich (TUM), specializing in Machine Learning and Computer Vision, and supported by the relAI Scholarship at the Konrad Zuse School of Excellence in Reliable AI.

Before starting my master's, I was already actively involved in research across multiple machine learning domains including reinforcement learning, explainable AI, federated learning, optimization, and medical AI systems. My earlier research experience included publications, large-scale experimental pipelines, and presenting research internationally at ACM SIGSPATIAL 2024 as an SRC Finalist.

My current work focuses on generative vision and modern representation learning systems, particularly:

• Diffusion Models
• Flow Matching
• Representation Learning
• Foundation Models
• Image & Video Generation
• 3D Generative Vision
• Vision-Language & Multimodal Models

My research interests center around building scalable and expressive foundation models capable of understanding and generating visual worlds.

research_interests = {
    "Foundation Models": [
        "Vision Foundation Models",
        "Multimodal Foundation Models",
        "Generative Foundation Models"
    ],

    "Generative Models": [
        "Diffusion Models",
        "Flow Matching",
        "Score-Based Generative Models",
        "Latent Diffusion"
    ],

    "Generative Vision": [
        "Image Generation",
        "Video Generation",
        "3D Generative Vision",
        "Neural Rendering"
    ],

    "Representation Learning": [
        "Self-Supervised Learning",
        "Multimodal Representation Learning",
        "Vision-Language Models"
    ],

    "Machine Learning Systems": [
        "Large-Scale Training",
        "Efficient Inference",
        "Reproducible ML Pipelines"
    ]
}

Student Research Assistant

Conducting research on diffusion models and generative vision with a focus on:

• Representation learning
• Foundation models
• Generative modeling
• Experimental infrastructure for diffusion systems
• Large-scale training and evaluation pipelines

Guided Research

Working on:

• 3D generation
• Diffusion-based 3D modeling
• Flow-matching approaches for generative vision
• 3D-aware representation learning

Worked on reinforcement learning and optimization systems including:

• RL for scheduling and combinatorial optimization
• Large-scale simulation environments
• Reliability-aware AI systems
• LLM-based reasoning and optimization pipelines

Built reproducible experimental infrastructure using PyTorch, Stable-Baselines3, and OpenAI Gym.

Worked on:

• Federated learning under heterogeneous client settings
• LLM-based feature extraction
• Generative AI applications for healthcare systems

Co-authored research on feature extraction using LLMs across heterogeneous federated environments.

First author and presenter at ACM SIGSPATIAL 2024 Student Research Competition.

Worked on:

• Explainable AI for localization systems
• Spatial reasoning systems
• Interpretable machine learning pipelines
• Experimental evaluation frameworks

Worked on medical AI systems involving:

• Graph Neural Networks (GNNs)
• Clinical prediction systems
• Vision-based medical AI
• Large-scale medical data preprocessing pipelines

Built machine learning backend systems for recommending optimal database configurations from system telemetry and performance data.

Worked on backend engineering, ML infrastructure, and production-oriented software systems.

MSc in Informatics
Specialization in Machine Learning & Computer Vision

relAI Scholar — Konrad Zuse School of Excellence in Reliable AI

BSc in Computer Engineering (AI Concentration)
Minor in Mathematics
Summa Cum Laude — GPA: 3.886

Enhancing Trust in Localization Systems: An XAI Approach

Presented research at ACM SIGSPATIAL 2024 in Atlanta, USA.

Improving Mixed-Criticality Scheduling with Reinforcement Learning

Research on reinforcement learning for optimization and scheduling systems.

Federated Learning Meets LLMs: Feature Extraction From Heterogeneous Clients

Research exploring LLM-based feature extraction across heterogeneous federated systems.

• Python
• C++
• Rust
• Java
• SQL

• PyTorch
• TensorFlow
• Hugging Face
• Scikit-learn
• Diffusers

• Diffusion Models
• DDPM / DDIM
• Latent Diffusion
• Flow Matching
• Vision-Language Models
• Foundation Models
• Representation Learning

• Docker
• Linux
• Git
• AWS
• GCP

🏆 relAI Scholar — Konrad Zuse School of Excellence in Reliable AI
🏆 ACM SIGSPATIAL 2024 SRC Finalist
🏆 Summa Cum Laude — AUC
🏆 RoboCup Egypt 2023 — 1st Place
🏆 RoboCup France 2023 — People's Choice Award

• LinkedIn: https://www.linkedin.com/in/muhammadahmedelmahdy/
• Email: muhammad.elmahdy@tum.de
• Email: muhammadahmedelmahdy@aucegypt.edu