# Modar M. Alfadly

Riyadh, KSA · modar.alfadly@gmail.com · [github.com/xmodar](https://github.com/xmodar) · [linkedin.com/in/modar-alfadly](https://linkedin.com/in/modar-alfadly) · [modar.me](https://modar.me) · [Google Scholar](https://scholar.google.com/citations?user=QcS8ktMAAAAJ)

## Summary

Machine learning researcher and engineer with a decade across academic research (KAUST; CVPR and AAAI oral papers on the analytical and robustness properties of deep networks), industry (Meta Reality Labs, Mozn) and national AI programs (helped redefine SDAIA Academy and three national Data & AI standards/frameworks). PyTorch at library-author level.

## Technical profile

- **Deep learning engineering**, PyTorch at library-author level: [InvTorch](https://github.com/xmodar/invtorch) (memory-efficient invertible functions), [network_moments](https://github.com/xmodar/network_moments) (probabilistic moments of DNNs), official code for two robustness papers; PyTorch community contributions (a differentiable [covariance](https://discuss.pytorch.org/t/covariance-and-gradient-support/16217) that predated torch.cov, a [symmetric matrix square root](https://github.com/pytorch/pytorch/issues/25481)); NumPy, TorchVision, ONNX, CUDA.
- **LLM inference and serving**, as co-author of the Academy's expert course on inference performance and cost: latency/throughput profiling, continuous batching, KV cache and PagedAttention, prefix and semantic caching, quantisation, distillation, model routing, GPU autoscaling and TCO, with labs on vLLM, ONNX Runtime and TensorRT.
- **Systems and tooling**: Linux, Git, Docker, FastAPI and REST services, SQL, Bash; author of [uvn](https://github.com/xmodar/uvn), a centralised virtual-environment manager for uv (2024); C/C++ and CUDA for GPU code, TypeScript for tooling.

## Experience

**Saudi Data & AI Authority (SDAIA)**, Director, SDAIA Academy Programs & Capability Building Consultant, November 2024 to present

- Co-authored the SDAIA Academy catalogue of instructor-ready courses across Foundation, Data Scientist and AI Engineer tracks, from Python and PyTorch to fine-tuning (PEFT/TRL), RAG, agentic systems (LangGraph, MCP), LLMOps, MLOps, inference performance and cost optimisation, and AI security and red-teaming.
- Helped launch SDAIA's Professional Badges for AI Engineers and Data Scientists (Assistant, Associate, Expert).
- Led the flagship program *AI Productivity: Automation & Agentic AI*, delivered nationally through SDAIA's [SAMAI initiative](https://sdaia.gov.sa/en/MediaCenter/Initiatives/Pages/Details.aspx?ItemID=19): agents and tool use, prompt engineering, automation and generative coding, responsible use.
- Built SDAIA Academy programs from the ground up: learning tracks, hired the core delivery team, trained 200+ professionals across government entities.
- Co-developed three national Data & AI standards (academic, occupational, higher-education curriculum), launched at ICAN 2026 and recognized by the Council of Ministers during the Year of AI; keynotes at national AI events.

**[Artificial Intelligence Association (AIA)](https://aia.org.sa/)**, Board Member, 2025 to present: national AI ecosystem initiatives and strategic direction of the association.

**KAUST, Image and Video Understanding Lab (IVUL)**, Doctoral Researcher (advisor: Prof. Bernard Ghanem); Teaching Assistant 2018 to 2021, 2018 to March 2023

- Research on the analytical characterization of deep networks under input noise, adversarial robustness, and VQA robustness: CVPR 2018 oral, AAAI 2019 oral, further preprints; code open-sourced.
- Teaching Assistant for PhD-level *Deep Learning for Visual Computing* and *Introduction to Computer Vision*; designed the courses' [hands-on deep learning labs](https://drive.google.com/drive/folders/13Xx6dBr-j57z1qlgMEw8r-TgY7Qa1HM3?usp=drive_link); ran the lab's PyTorch tutorial and [reading-group talks](https://github.com/IVUL-KAUST/GroupReading).

**Meta (Facebook Reality Labs)**, Research Scientist Intern, 2019: active learning for large-scale semantic segmentation.

**Mozn Systems**, Deep Learning Consultant, 2017 to 2018: joined at inception and built its first data-driven solutions.

**KAUST**, Research Intern, Summer 2014: GPU-based methods for 3D neuron skeletonization in CUDA.

## Education

- KAUST, doctoral studies in Computer Science, 2018 to 2023, KAUST Fellowship (full funding); did not complete.
- KAUST, M.S. Computer Science, 2018. Thesis: Analytic Treatment of Deep Neural Networks Under Additive Gaussian Noise.
- KFUPM, B.S. Software Engineering (Honors), 2016.

## Publications

- 2025. Towards a Unified Benchmark for Arabic Pronunciation Assessment: Qur'anic Recitation as Case Study. Interspeech 2025. https://www.isca-archive.org/interspeech_2025/elkheir25b_interspeech.html
- 2023. Improving Visual Question Answering Models through Robustness Analysis and In-Context Learning with a Chain of Basic Questions. arXiv preprint [second author]. https://arxiv.org/abs/2304.03147
- 2021. Improving Variance Estimates in Generative Models. Technical report.
- 2020. Network Moments: Extensions and Sparse-Smooth Attacks. arXiv preprint. https://arxiv.org/abs/2006.11776
- 2019. A Novel Framework for Robustness Analysis of Visual QA Models. AAAI 2019, oral [second author]. https://aaai.org/papers/08449-a-novel-framework-for-robustness-analysis-of-visual-qa-models/
- 2019. Assessing the Robustness of Visual Question Answering. arXiv preprint [second author]. https://arxiv.org/abs/1912.01452
- 2019. Expected Tight Bounds for Robust Deep Neural Network Training. arXiv preprint. https://arxiv.org/abs/1905.12418
- 2019. Analytical Moment Regularizer for Gaussian Robust Networks. arXiv preprint. https://arxiv.org/abs/1904.11005
- 2018. Analytic Expressions for Probabilistic Moments of PL-DNN with Gaussian Input. CVPR 2018, oral. https://openaccess.thecvf.com/content_cvpr_2018/html/Bibi_Analytic_Expressions_for_CVPR_2018_paper.html
- 2018. Robustness Analysis of Visual QA Models by Basic Questions. CVPR Workshop [second author]. https://arxiv.org/abs/1709.04625
- 2017. VQABQ: Visual Question Answering by Basic Questions. CVPR Workshop [second author]. https://arxiv.org/abs/1703.06492

## Service and awards

- Reviewer: CVPR, NeurIPS, ICLR, ICCV, AAAI
- 1st place, National Programming Contest
- KFUPM Hackathon winner