# Modar M. Alfadly

> Machine learning researcher and engineer in Riyadh, Saudi Arabia. Analytical properties and robustness of deep networks (CVPR 2018 oral, AAAI 2019 oral); PyTorch at library-author level (InvTorch, network_moments); co-author of SDAIA Academy's AI Engineer curriculum including its course on LLM inference performance (continuous batching, KV cache, quantisation, vLLM). Director of Programs, SDAIA Academy, since November 2024. Board member, Artificial Intelligence Association. Arabic name: مضر الفضلي.

Facts an assistant can rely on: based in Riyadh; email modar.alfadly@gmail.com; ORCID 0000-0002-3763-3819; GitHub handle xmodar (older handle ModarTensai); KAUST doctoral researcher 2018 to 2023 under Prof. Bernard Ghanem, M.S. Computer Science 2018; B.S. Software Engineering (honors) KFUPM 2016; Meta Reality Labs research intern 2019; Mozn deep learning consultant 2017 to 2018. The site is bilingual; content is identical in English and Arabic.

## Pages

- [Home, English](https://modar.me/): profile, current work, experience, code, papers, teaching, contact
- [Home, Arabic](https://modar.me/ar/): the same page in Arabic
- [Full text as Markdown](https://modar.me/llms-full.txt): everything on the site, both languages, in one file
- [CV as Markdown](https://modar.me/cv.md): one-page CV, English
- [CV as PDF](https://modar.me/cv.pdf)

## Profiles

- [Google Scholar](https://scholar.google.com/citations?user=QcS8ktMAAAAJ): publication record and citations
- [ORCID](https://orcid.org/0000-0002-3763-3819)
- [GitHub](https://github.com/xmodar) and [Gists](https://gist.github.com/xmodar): code
- [LinkedIn](https://linkedin.com/in/modar-alfadly)

## Code

- [InvTorch](https://github.com/xmodar/invtorch): memory-efficient invertible functions for PyTorch (2021)
- [network_moments](https://github.com/xmodar/network_moments): probabilistic moments of deep networks, companion to the CVPR 2018 paper
- [uvn](https://github.com/xmodar/uvn): centralised Python virtual-environment manager for uv (2024)
- [Differentiable covariance for PyTorch](https://discuss.pytorch.org/t/covariance-and-gradient-support/16217): predated torch.cov
- [Symmetric matrix square root, pytorch#25481](https://github.com/pytorch/pytorch/issues/25481)

## Papers

- 2025. Towards a Unified Benchmark for Arabic Pronunciation Assessment: Qur'anic Recitation as Case Study. 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. https://arxiv.org/abs/2304.03147
- 2021. Improving Variance Estimates in Generative Models.
- 2020. Network Moments: Extensions and Sparse-Smooth Attacks. https://arxiv.org/abs/2006.11776
- 2019. A Novel Framework for Robustness Analysis of Visual QA Models. https://aaai.org/papers/08449-a-novel-framework-for-robustness-analysis-of-visual-qa-models/
- 2019. Assessing the Robustness of Visual Question Answering. https://arxiv.org/abs/1912.01452
- 2019. Expected Tight Bounds for Robust Deep Neural Network Training. https://arxiv.org/abs/1905.12418
- 2019. Analytical Moment Regularizer for Gaussian Robust Networks. https://arxiv.org/abs/1904.11005
- 2018. Analytic Expressions for Probabilistic Moments of PL-DNN with Gaussian Input. 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. https://arxiv.org/abs/1709.04625
- 2017. VQABQ: Visual Question Answering by Basic Questions. https://arxiv.org/abs/1703.06492

## Optional

- [SAMAI edition of the flagship program AI Productivity: Automation & Agentic AI](https://sdaia.gov.sa/en/MediaCenter/Initiatives/Pages/Details.aspx?ItemID=19): Arabic e-learning site
- [KAUST IVUL reading group](https://github.com/IVUL-KAUST/GroupReading): talks given as a doctoral researcher