___
 _____ ___ ___ ___|  _|___ ___
|     | . |_ -| .'|  _| -_|  _|
|_|_|_|___|___|__,|_| |___|_|

Independent ML/DL researcher | Network Engineer

Research interests

• AI safety and the internals of learning systems
• Uncertainty-aware deep reinforcement learning
• Anomaly detection in Multivariate Time Series and compromise detection
• Machine learning provenance and integrity

Publications and preprints

• From Latent Space to Jacobian Space: Measuring, Evading, and Training Against Safety-Content Accessibility.
  M. Mosafer. arXiv:2610.04316, 2026. [abs]
• GARSHA: Global Anomaly Representation for Series with Heterogeneous Attention: A tunable entropy approach for outlier detection.
  M. Mosafer, P. Mahmoudi-Nasr, R. Yousefpour. SSRN, 2026. [DOI]
• A Deep Ensemble Learning Method to Detect Insider Threats.
  M. Almoayed, P. Mahmoudi-Nasr, M. Mosafer. The ISC International Journal of Information Security, vol. 18, no. 2, 2025. [DOI]
• Stochastic Entropy-Regularized Deep Reinforcement Learning for TCP Congestion Control (SER-DQN).
  M. Mosafer, M. Bastam. Manuscript in preparation. [code]
• AI Terminology 101: A Plain-English Handbook to the Words Behind Artificial Intelligence.
  M. Mosafer, S. Bahri. Book, 2026. [Amazon]
• Looped transformers for semi-supervised reinforcement learning in multivariate time-series anomaly detection. In progress.

Code

• serdqn - deep RL controller for TCP congestion control (ns-3/OpenGym, PyTorch)
• rltsad - reinforcement learning for time-series anomaly detection
• adjudicator - checkpointing and integrity tooling for AI agent pipelines (Go)

Cybersecurity class for Undergraduate

Contact

mosafer at outlook dot com
github.com/mohsafer | google scholar | orcid.org/0009-0003-7148-6538

Last modified: 2026-11-02