Margin-based Progress Measures for Grokking
Ngnawé, J., & Tikeng Notsawo, P. J. (2026). Margin-based Progress Measures for Grokking. NeurIPS 2026 Workshop on Neural Network Artifacts as a New Data Modality.
CV
Education
- PhD in Computer Science (Fall 2021 - Ongoing)
- Mila and Intelligence and Data Institute (IID), Laval University, Quebec
- Visiting PhD student at Stanford University
- Focus: Adversarial Robustness, Distribution Shifts and Uncertainty of Deep Learning Models
- Research: Jonas Ngnawé - Google Scholar
- DEEL project’s page: DEpendable and Explainable Learning - DEEL-QUEBEC - DEEL-FRANCE
- Master’s in Machine Learning (2018-2019)
- AIMS African Master’s of Machine Intelligence, Kigali, Rwanda
- Funded by Facebook and Google
- Courses: Link complete list of courses (Fundamentals, Vision, NLP etc.)
- Project: Scalable Inference with Distributed Gaussian Processes using the Wasserstein Barycenter.
- Master’s in Mathematical Sciences (2016-2017)
- African Institute for Mathematical Sciences, Limbe, Cameroon
- Project: k-Clustering Problem-Application to a Field Technician Allocation Optimization Problem.
- Master’s Degree in Computer Engineering (2011-2016)
- Ecole Nationale Supérieure Polytechnique de Yaoundé, Cameroon
- Other: Machine Learning Summer School (MLSS), London 2019
Work experience
- Google AI resident (Oct 2019 - Sep 2021)
- Google Brain Team Accra, Ghana
- Development of robust methods for deep neural networks
- Compression of deep neural networks by quantization and pruning
- Tools: Python/Tensorflow 2.0, Google Cloud Platform (GCP)
- Publication: Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Networks (Accepted at NeurIPS 2022 Workshop on Distribution Shifts: Connecting Methods and Applications)
- Computer Engineer (Jan - Aug 2018)
- GroupOne Holding Company, Limbe, Cameroun
- Website development
- Fuel consumption prediction projects (Fraud Detection, Supply prediction)
- Business plan for startup incubator creation
- Industrial Maintenance (Refinery) of the Honeywell Experion System
- IT Support
- Tools: HTML5/CSS3/JavaScript/JQuery/JQueryUI/Ajax/PHP, MySQL, Méthodologie PMI, Business Model Canvas
- Freelance Software Developer (2017-2018)
- Digitization of administrative procedures for various ministries in Cameroon
- Development and maintenance of WordPress plugins (WooCommerce)
- Tools: Electronic Document Management (EDM)/Electronic Archiving(MARCH RM), PHP/Woocommerce, MySQL Workbench, Microsoft Access, Git/Gitlab/Github, Java, Netbeans, SCRUM
- Software Engineer Intern (Feb-August 2016)
- Bolloré Africa Logistics, Port of Douala, Cameroon
- Automation and Analysis of the Review of Access Rights to the Information System
- Detection of abnormal access rights
- Tools: Talend Open Studio, Qlikview, C#, ASP.Net, Microsoft SQL Server, Visual Studio, UML/Merise, BPMN 2.0, PMI Methodology
Additional Skills
- Data Science/AI: Pytorch, Scikit-Learn, Pandas, Weight&Biases, Matlab, Octave, Docker etc.
Publications
RobustGenBench: A Benchmark for Robust Generalization to Adversarial and Common Perturbations, with Applications to Vision and Vision-Enabled Large Language Models
Heuillet, M., Ngnawé, J., Bhagwatkar, R., Pequignot, Y., Larouche, A., Gagné, C., Rish, I., Ahmad, O., & Durand, A. (2026). RobustGenBench: A Benchmark for Robust Generalization to Adversarial and Common Perturbations, with Applications to Vision and Vision-Enabled Large Language Models. NeurIPS 2026, Evaluations and Datasets Track.
Mind the Spike: Mechanisms and Brittleness of Visual Massive Activations in Large Vision–Language Models
Ngnawé, J., Pequignot, Y., Sahoo, S., Gagné, C., Precioso, F., & Koyejo, S. (2026). Mind the Spike: Mechanisms and Brittleness of Visual Massive Activations in Large Vision–Language Models. arXiv preprint arXiv:2609.32808.
A Stochastic–Geometric Theory of Scaling Laws in Grokking
Luo, R., Gagné, C., Ngnawé, J., Ullah, I., & Morrissey, K. (2026). A Stochastic–Geometric Theory of Scaling Laws in Grokking. arXiv preprint arXiv:2606.30388.
Robust Fine-Tuning from Non-Robust Pretrained Models: Mitigating Suboptimal Transfer With Epsilon-Scheduling
Ngnawé, J., Heuillet, M., Sahoo, S., Pequignot, Y., Ahmad, O., Durand, A., Precioso, F., Gagné, C. (2026). Robust Fine-Tuning from Non-Robust Pretrained Models: Mitigating Suboptimal Transfer With Epsilon-Scheduling. The Fourteenth International Conference on Learning Representations (ICLR 2026).
A Guide to Robust Generalization: The Impact of Architecture, Pre-training, and Optimization Strategy
Heuillet, M., Bhagwatkar, R., Ngnawé. J., Pequignot, Y., Larouche, A., Gagné, C., Rish, I., Ahmad, O., Durand, A. (2025). A Guide to Robust Generalization: The Impact of Architecture, Pre-training, and Optimization Strategy. arXiv preprint arXiv:2508.14079.
A Layer Selection Approach to Test Time Adaptation
Sahoo, S., ElAraby, M., Ngnawe, J., Pequignot, Y. B., Precioso, F., & Gagné, C. (2025, April). A layer selection approach to test time adaptation. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 39, No. 19, pp. 20237-20245).
Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers.
Ngnawé, J., Sahoo, S., Pequignot, Y., Precioso, F., & Gagné, C. (2024). Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers. The Thirty-eighth Annual Conference on Neural Information Processing Systems.
Robustmix: Improving Robustness by Regularizing the Frequency Bias of Deep Nets
Ngnawe, J., NJIFON, M. A., Heek, J., and Dauphin, Y. Robustmix: Improving robustness by regularizing the frequency bias of deep nets. In NeurIPS 2022 Workshop on Distribution Shifts: Connecting Methods and Applications, 2022. URL https://openreview.net/forum?id=Na64z0YpOx.
Talks
Galsen AI Reading Group Robust Fine-Tuning from Non-Robust Pretrained Models: Mitigating Suboptimal Transfer With Epsilon-Scheduling Talk at Galsen AI Youtube Channel, Online
Galsen AI Reading Group Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers Talk at Galsen AI Youtube Channel, Online
Neptune.ai Challenge AI Research Paper Overview: Detecting Brittle Decisions For Free Talk at Neptune.ai Youtube Channel, Online
Teaching
Activities
- Host (2024)
- Keynote speaker (Dec. 27, 2022)
- Deep Learning IndabaX-Cameroon: Robustness in AI
- Organizing committee member
- Founding member (Since 2020)
- Kmer-AI, an association for the promotion and vulgarization of AI and Machine Learning in Cameroon.
- Co-organizer (Since 2020)
- Machine Learning Projects Competition (MLPC), Polytechnique Yaoundé
- Co-organizer (September 2019)
- Two (02) week introductory course in Machine Learning, Polytechnique Yaoundé
- Co-organizer (April 2019)
- First Deep Learning IndabaX Rwanda, a conference on Machine Learning and AI
- Reviewer (2022, 2023, 2024)
- Reviews: Algorithmic Fairness through the Lens of Causality and Privacy (2022), Montreal AI Symposium, Black in AI Workshop @NeurIPS, Deep Learning Indaba.