Nahom M. Birhan - AI Security Researcher

Nahom M. Birhan

AI Security Researcher & Machine Learning Engineer

Research Assistant at Old Dominion University investigating jailbreak and backdoor attacks and their defenses in large language, multimodal models and federated learning settings . 3+ years of industry experience building applied ML proof of concept (POC) projects, and enterprise software.

Ongoing specializing in AI robustness, alignment, and security-first model design, combining deep-learning techniques with Reinforcement Learning preference optimization techniques. Committed to ethical AI development and the creation of secure, resilient AI systems.

Research Interests

Ongoing specializing in AI robustness, alignment, and security-first model design, combining deep-learning techniques with Reinforcement Learning preference optimization techniques. Committed to ethical AI development and the creation of secure, resilient AI systems.

Research Areas

Reinforcement Learning for human alignmentLanguage Model RobustnessMultimodal Language Model RobustnessFederated Learning and PrivacyAnomaly/Imposter DetectionEdge Computing and security

Research Projects

Jailbreak and Backdoor Attacks in LLMs and Multimodal LMs

Actively working on this research, not published yet

Impostor Detection in IoT Edge Sensor Networks

Developed lightweight LSTM model for edge devices, optimized using TensorFlow Lite quantization

Technologies: Raspberry PI, Arduino nano 33 BLE, TensorFlow/Lite, Python, C/C++

Email Phishing Detection in Africa

Analyzed African-specific phishing patterns, implemented and compared five ML models, mitigated overfitting

Technologies: Scikit-learn, TensorFlow, Numpy

Blockchain-Based Healthcare Data Security

Proposed system integrating blockchain and IPFS for secure medical records, compared IPFS and Amazon S3 access times.

Technologies: IPFS, AWS (IAM, S3), Blockchain concepts

Prime and Composite Classification

Created dataset of 2 million numbers, developed models using sequence models and 1D-CNN, analyzed performance

Technologies: Python, TensorFlow, NumPy, Google Colab

Smart Irrigation System (BSc. research)

Designed automated system with XBee sensor communication, integrated motors, pumps, and sensors, developed control algorithms

Technologies: Proteus, XBee, Arduino, C/C++, Various sensors