~/research // publications & patents

research.log

Most of my research lives in the gap between how secure a system looks and how secure it really is. Twelve peer-reviewed papers and two granted patents came out of that work, published through IEEE, MDPI and other international journals. The three I'd lead with are below; everything else is sealed in the archive. My core research focus areas are OT security, IoT security, satellite security, cyber threat intelligence, and Windows exploitation.

12
PUBLICATIONS
02
GRANTED PATENTS
02
FIRST-AUTHOR
$17K
RESEARCH GRANT
$ cat ~/research/selected.md
Journal of Cybersecurity and Privacy (MDPI) · 2025

Threat Intelligence Extraction Framework (TIEF) for TTP Extraction

Cyber threat reports are written for humans, not machines. TIEF is a DistilBERT pipeline that reads those unstructured reports and extracts attacker tactics and techniques as clean STIX 2.1 data. It maps to 560 MITRE ATT&CK classes and reaches an F1 of 0.933 while running on 40% fewer parameters than a full BERT model.

F1 0.933 560 ATT&CK classes DistilBERT STIX 2.1 BERTopic
Institute of Electrical and Electronics Engineers (IEEE) · 2025

From Simulation to Application: The Role of CTF Competitions in Cybersecurity Training

You do not learn to break into systems by reading about it. This paper makes the case for Capture the Flag competitions as real training, breaking down how good challenges are built and why hands-on problem solving sticks far better than lectures. It follows CTFs from the classroom all the way to large-scale contests.

CTF Cybersecurity Education Applied Pedagogy
International Journal of Intelligent Systems and Applications in Engineering · 2024

Detecting Deepfakes: Exploring Machine Learning Models for Audio, Video, and Image Analysis

A comparative study of machine-learning models for deepfake detection across audio, video and image. It benchmarks Random Forests, Gradient Boosting, SVMs, neural networks and CNNs, and weighs ensemble and ResNet approaches for future work.

audio · video · image Ensemble CNN SVM
$ ls -la ~/publications · 12 records

Threat Intelligence Extraction Framework (TIEF) for TTP Extraction

F1 0.933

Journal of Cybersecurity and Privacy (MDPI) · 2025

A DistilBERT pipeline that converts unstructured threat reports into structured STIX 2.1 intelligence. It classifies attacker behaviour across 560 MITRE ATT&CK techniques and sub-techniques, reaching an F1 of 0.933 on a tough multi-label task.

DistilBERT STIX 2.1
read paper →

From Simulation to Application: The Role of CTF Competitions in Cybersecurity Training

education

Institute of Electrical and Electronics Engineers (IEEE) · 2025

Argues that Capture the Flag competitions are an effective way to teach real cybersecurity skill. It breaks down how challenges are designed and deployed, from classrooms to job training to major contests, and why active problem solving beats passive study.

CTF Security Education
read paper →

ResNet50 DeepFake Detector: Unmasking Reality

97% accuracy

Indian Journal of Science and Technology · 2024

A ResNet50 detector for video deepfakes trained on CelebDF and FaceForensics++. An image-scraping stage removes irrelevant regions before inference, pushing accuracy to 97% with under 5% error across face-swap and lip-sync forgeries.

ResNet50 FaceForensics++
read paper →

Securing Visual Integrity: EfficientNetB4 with Attention Layers and Siamese Training

94% accuracy

International Journal of Intelligent Systems and Applications in Engineering · 2024

Extends EfficientNetB4 with attention layers and Siamese training to spot manipulated faces in video. Evaluated on the FaceForensics and DFDC benchmarks, the model reaches 94% accuracy against modern face-manipulation methods.

EfficientNetB4 Siamese Nets Attention
read paper →

Advancements in Video Deepfake Detection: ResNet50, EfficientNetB7 and EfficientNetAutoAttB4

ensemble

International Journal of Intelligent Systems and Applications in Engineering · 2024

Combines ResNet50, EfficientNetB7 and EfficientNetAutoAttB4 into a single averaged ensemble. The design catches both still-image and full-video deepfakes, stays light on compute, and is built with real-world deployment in mind.

ResNet50 EfficientNet Ensemble
read paper →

Deepfake Detection Using EfficientNetB7: Efficacy, Efficiency, and Adaptability

85% accuracy

International Journal of Intelligent Systems and Applications in Engineering · 2024

Studies how well EfficientNetB7's compound scaling transfers to video deepfake detection. It weighs accuracy against computational cost across several datasets, landing near 85% while staying efficient enough to deploy.

EfficientNetB7 CNN
read paper →

Harmonizing Algorithms: An Approach to Enhancing Audio Deepfake Detection

97.28% accuracy

International Journal of Intelligent Systems and Applications in Engineering · 2024

Targets audio deepfakes by running two detectors together: an MFCC-based SVM at 97.28% accuracy and an attention neural network at 91.04%. Cross-checking both models makes the authenticity call far more reliable than either alone.

MFCC SVM Attention
read paper →

Improving Deepfake Audio Detection: A Support Vector Machine Approach with MFCC

97.28% accuracy

International Journal of Intelligent Systems and Applications in Engineering · 2024

Trains a Support Vector Machine on MFCC features drawn from the 195,000-clip Fake-or-Real dataset. The streamlined model separates genuine speech from synthetic voices at 97.28% accuracy and sets a baseline for tougher audio forensics.

MFCC SVM FoR dataset
read paper →

Enhancing Audio Deepfake Detection using SVM and MFCC

97.28% accuracy

Journal of Harbin Engineering University · 2024

An SVM and MFCC pipeline for detecting synthetic speech, with a focus on how raw, unprocessed training data improves robustness. It distinguishes real from generated voices at 97.28% accuracy.

MFCC SVM
read paper →

Detecting Deepfakes: Exploring Machine Learning Models for Audio, Video, and Image Analysis

comparative study

International Journal of Intelligent Systems and Applications in Engineering · 2024

A comparative study of machine-learning models for deepfake detection across audio, video and image. It benchmarks Random Forests, Gradient Boosting, SVMs, neural networks and CNNs, and weighs ensemble and ResNet approaches for future work.

Ensemble CNN
read paper →

Deepfake Technology and Image Forensics: Advancements, Challenges, and Ethical Implications

review

International Journal of Intelligent Systems and Applications in Engineering · 2024

A broad review of how deepfakes are built with autoencoders, GANs and CNNs, and how image forensics fights back. It also digs into the ethical stakes of synthetic media and where detection research needs to go next.

GAN Image Forensics Ethics
read paper →

PixelTrace: A Forensic Investigation Tool

forensics tool

Journal of Emerging Technologies and Innovative Research · 2023

A digital-forensics tool that pairs Error Level Analysis with machine learning to flag tampered or morphed regions inside an image. Built to support investigators, journalists and platforms checking whether a photo is authentic.

Error Level Analysis Image Tampering
read paper →
$ ls ~/patents · 2 granted

Deepfake detection system and method for detecting misinformation in manipulated content

granted

A system for flagging manipulated audio, video and image content, identifying synthetic media used to spread misinformation.

Cybersecurity contest platform system and method

granted

A platform for hosting and running cybersecurity competitions — provisioning challenges, scoring submissions and managing participants.

$ want to see what the work has earned? cd ~/achievements →
$ fsck ~/archive · 1 corrupted record: ░░░░░_░░░░░░░.enc [recover]