Mabon Ninan
Texas A&M University, advised by Dr. Marcus Botacin. Machine learning, cybersecurity, and trustworthy AI.
I build security systems that still work when conditions are not ideal — when the threat landscape drifts, the signal is noisy, the hardware is tiny, or someone is actively trying to fool the model. That means malware detection you can interpret, side-channel analysis that transfers between real devices, neural networks small enough for a microcontroller, and clinical AI that holds up on the patients it was not trained on.
- 9 Publications
- 33 Citations OpenAlex
- 3 h-index OpenAlex
- 1 Best Paper HOST 2024
- 5 Students Mentored
- 7 Talks Given
Citation figures as of September 2026.
Research
-
Robust & Explainable Malware Detection
Detectors that stay reliable as malware evolves, and that can say why they flagged something — a requirement for anyone who has to act on the alert.
-
Agentic & LLM-Assisted Security
Self-healing and adaptive pipelines for security workflows, including where language models help an analyst and where they quietly mislead one.
-
Side-Channel Attacks & Defenses
How physical leakage behaves across real hardware, and why attacks that look strong in one lab setup often fail on the next device.
-
Tiny Neural Networks
Compressing models far enough to run on constrained and embedded devices — which changes what a resource-limited attacker or defender can do.
-
Multimodal Medical AI
Interpretable models for pediatric chest radiography and clinical report understanding, evaluated on the populations that actually get scanned.
Selected work
A Second Look at the Portability of Deep Learning Side-Channel Attacks over EM Traces
27th International Symposium on Research in Attacks, Intrusions and Defenses (RAID '24)
Earlier portability results for EM side-channel attacks were established on easy targets such as 8-bit microcontrollers. This paper re-runs them on harder ones — 32-bit microcontrollers and traces with random delay — under domain shifts from hardware variation, different keys, and inconsistent probe placement. Pre-processing and unsupervised domain adaptation do help, but which method wins depends on the target and probe location, and none dominates: results from easy targets do not generalize. The paper also identifies two evaluation pitfalls that make cross-device attacks look better than they are, and releases a 3-million-trace public dataset.
Abstract
Deep learning side-channel attacks can recover encryption keys on a target by analyzing power consumption or electromagnetic (EM) signals. However, they are less portable when there are domain shifts between training and test data. While existing studies have shown that pre-processing and unsupervised domain adaptation can enhance the portability of deep learning side-channel attacks given domain shifts over EM traces, the findings are limited to easy targets (e.g., 8-bit microcontrollers). In this paper, we investigate the portability of deep learning side-channel attacks over EM traces acquired from more challenging targets, including 32-bit microcontrollers and EM traces with random delay. We study domain shifts introduced by the combination of hardware variations, distinct keys, and inconsistent probe locations between two targets. In addition, we perform comparative analyses of multiple existing (and new) pre-processing and unsupervised domain adaptation methods. We conduct a series of comprehensive experiments and derive three main observations. (1) Pre-processing and unsupervised domain adaptation methods can enhance the portability of deep learning side-channel attacks over more challenging targets. (2) The effectiveness of each method, however, varies depending on the target and probe locations in use. In other words, observations of a method on easy targets do not necessarily generalize to challenging targets. (3) None of the methods can constantly outperform others. Moreover, we highlight two types of pitfalls that could lead to over-optimistic attack results in cross-device evaluations. We also contribute a large-scale public dataset (with 3 million EM traces from 9 probe locations over multiple targets) for benchmarking and reproducibility of side-channel attacks tackling domain shifts over EM traces.
Artifact. Contributes a public dataset of 3 million EM traces across 9 probe locations and multiple targets.
TinyPower: Side-Channel Attacks with Tiny Neural Networks
2024 IEEE International Symposium on Hardware Oriented Security and Trust (HOST)
Deep-learning side-channel attacks are usually assumed to need a well-resourced attacker. TinyPower shows the attack still works after aggressive pruning and quantization, with models compact enough to run on microcontrollers, which lowers the practical bar for mounting one. Awarded Best Student Paper at HOST 2024.
TinyRadio: Tiny Neural Networks for Fingerprinting Radio Frequency Signals
IEEE National Aerospace and Electronics Conference (NAECON 2025)
Radio-frequency fingerprinting identifies an individual transmitter from imperfections in the signal it emits. TinyRadio carries the tiny-neural-network approach into that setting, building classifiers small enough to run on embedded hardware rather than requiring a workstation to make the identification.
Education, research experience, service, and technical skills are on the full CV — for collaborations or questions, reach out at ninanmm@tamu.edu.