Modarمضر
العربية
Portrait of Modar M. Alfadly

Modar M. Alfadly

Machine learning researcher and engineer.

I work on the parts of machine learning that have to be right: the mathematics of how deep networks behave, the library code others build on, and the systems that serve models at scale. Trained at KAUST, published at CVPR and AAAI, and now building Saudi Arabia's AI engineering curriculum at SDAIA.

Based in
Riyadh, Saudi Arabia
Currently
Director of Programs, SDAIA Academy

Download CVGitHubGoogle ScholarLinkedIn

Figure from the CVPR 2018 paper: panda images with Gaussian noise of growing variance enter a neural network; the expected output is confident for the mean input and flattens as the noise grows.
From the CVPR 2018 paper: a Gaussian input of growing variance, the network Φ, and the expected output. Feeding the mean input gives the confident top row; the true expectation flattens as σ grows, and the paper derives its mean and variance in closed form.

Now

SDAIA Academy, since November 2024

At the Saudi Data & AI Authority I direct the Academy's programs and its capability-building work. The technical core of that is the AI Engineer curriculum.

  • AI Engineer curriculum

    Co-authored the Academy's course catalogue across foundation, data-science and AI-engineering tracks: PyTorch, fine-tuning, retrieval-augmented generation, agentic systems with LangGraph and MCP, MLOps and LLMOps, security and red-teaming. Its expert course on inference performance and cost covers latency and throughput profiling, continuous batching, KV cache and PagedAttention, prefix and semantic caching, quantisation, distillation, model routing and GPU autoscaling, with labs on vLLM, ONNX Runtime and TensorRT.

  • AI Productivity: Automation & Agentic AI

    The Academy's flagship program, delivered nationally through SDAIA's SAMAI initiative: what AI applications, agents and models can do; prompt engineering; process automation and generative coding; responsible use and output verification.

  • Professional Badges

    Helped launch SDAIA's badges for AI engineers and data scientists at the Assistant, Associate and Expert levels.

  • National standards

    Co-developed three national Data & AI standards (academic, occupational, and the higher-education curriculum), launched at ICAN 2026 and recognised by the Council of Ministers during the Year of AI.

  • Artificial Intelligence Association

    Board member since 2025.

Work

  • Saudi Data & AI Authority (SDAIA)

    Director, SDAIA Academy Programs & Capability Building ConsultantNovember 2024 to present

    Built the Academy's programs and delivery team from the ground up and trained more than 200 professionals across government entities.

  • Artificial Intelligence Association (AIA)

    Board Member2025 to present

    National AI ecosystem initiatives and the strategic direction of the association.

  • KAUST, Image and Video Understanding Lab (IVUL)

    Doctoral Researcher, advised by Prof. Bernard Ghanem2018 to 2023

    Analytical characterisation of deep networks under input noise, adversarial robustness, and the robustness of visual question answering: CVPR 2018 oral, AAAI 2019 oral, further preprints. Released the code behind the papers and built InvTorch. Teaching Assistant for PhD-level courses from 2018 to 2021. M.S. in Computer Science, 2018, with a thesis on the analytic treatment of deep networks under additive Gaussian noise.

  • Meta (Facebook Reality Labs)

    Research Scientist InternSummer 2019

    Active learning for semantic segmentation in large-scale vision systems.

  • Mozn Systems

    Deep Learning Consultant2017 to 2018

    Joined at the company's inception and built its first data-driven solutions.

  • King Fahd University of Petroleum and Minerals (KFUPM)

    B.S. in Software Engineering, with honors2011 to 2016

    First place in the First National Programming Contest (NC3, 2016); first and second places in the KFUPM Hackathon (2015, 2016). Research intern at KAUST in summer 2014, on GPU methods for 3D neuron skeletonisation in CUDA.

Code

Public code, oldest habit first: writing the utility properly and giving it away.

  • invtorch2021

    Memory-efficient invertible functions for PyTorch: an invertible checkpoint that frees a module's inputs during the forward pass and reconstructs them in backward, with an InvertibleModule abstraction and inverse checks.

  • network_moments2018

    A toolkit for computing probabilistic moments of deep networks, the companion to the CVPR 2018 paper and its extensions.

  • gaussian-regularizer and ptb2019

    Official implementations of the Gaussian moment regularizer and the probabilistically true and tight bounds for robust training.

  • uvn2024

    A centralised Python virtual-environment manager for uv: a conda-style workflow on uv's speed.

  • PyTorch community2018 to 2021

    A differentiable covariance implementation on the PyTorch forums that predated torch.cov, and a symmetric matrix square root for the linear-algebra module on the issue tracker (pytorch#25481).

  • Gists2019 to now

    PyTorch and NumPy utilities: custom autograd functions, invertible BatchNorm, deconvolution, positional encodings for point clouds, RNG state management; and a Lambert W function in TypeScript adopted by a third-party library.

Papers

Eleven papers and preprints, 2017 to 2025. The full record is on Google Scholar and ORCID.

  1. 2025
  2. 2023
  3. 2021

    Improving Variance Estimates in Generative Models

    Technical report

  4. 2020
  5. 2019
  6. 2019
  7. 2019
  8. 2019
  9. 2018
  10. 2018
  11. 2017

Teaching

  • KAUST, 2018 to 2021

    Teaching Assistant for the PhD-level courses Deep Learning for Visual Computing and Introduction to Computer Vision. Designed the courses' hands-on deep learning labs and ran the lab's PyTorch tutorial and reading-group talks on transfer learning, capsule networks and the probabilistic analysis of deep networks.

    Lab materials Reading group

  • SDAIA Academy, 2024 to now

    The AI Engineer curriculum and the flagship program described above, plus keynotes and talks at national AI events.

    SAMAI edition of the flagship program

Contact

Write to me about inference systems, the robustness of deep networks, or building AI capability in Saudi Arabia.