Research, Reproducibility & Trustworthy Synthetic Cybersecurity Data
I build reproducible research infrastructure and evaluation methodologies for synthetic cybersecurity data, malware-image classification, and trustworthy machine learning. My work emphasizes downstream utility, probabilistic reasoning, external conditioning audits, uncertainty, calibration, and systems-aware evaluation that scales into reliable research and engineering infrastructure.
Research Focus
My research program examines how synthetic cybersecurity data can be generated, evaluated, selected, and deployed through reproducible, utility-first, and uncertainty-aware methods.
GenCyberSynth Benchmarking
Reproducible comparison of synthetic malware-image generators under matched datasets, preprocessing, budgets, seeds, baseline models, and held-out downstream evaluation.
Synthetic Data Utility & Policy
Systematic study of when synthetic augmentation helps or harms malware classification, and how samples should be budgeted, filtered, ranked, accepted, or allocated.
External Conditioning Audits
Independent testing of whether class-conditional generators produce samples that are externally recognized as their requested malware classes.
Trustworthy & Probabilistic Evaluation
Reliability-focused evaluation of cybersecurity models through uncertainty estimation, calibration, robustness, abstention, and risk-aware decision policies.
Research Projects by Status
A living view of completed studies, active research, and planned directions across trustworthy cybersecurity machine learning, synthetic-data evaluation, probabilistic decision-making, and systems-oriented research infrastructure.
Completed
-
Masked Autoencoder for Data Recovery in Polymer Research
-
Reliable and Interference-Aware CSI Feedback with Bayesian Neural Networks
-
GenCyberSynth Phase-1: Reproducible Synthetic Malware Benchmarking
-
External Conditioning Audits for Class-Conditional Synthetic Malware Images
Pending / Active
-
When Does Synthetic Data Help Malware Classification?
-
Selective Synthetic Data Policies for Malware Classification
-
Trustworthy Synthetic Malware-Image Augmentation
-
GenCyberSynth 2.0: Scalable and Artifact-Backed Evaluation on HPC
Not Started
-
Robustness and Calibration Under Distribution Shift for Malware Classification
-
Uncertainty-Aware Security Decisions: Risk, Abstention, and Cost-Sensitive Classification
-
Classwise Reliability and Fair Utility in Synthetic Malware Augmentation
-
Compiler/Runtime-Informed Research Infrastructure for Scalable ML Evaluation
Selected Publications & Research Outputs
Peer-reviewed publications, active manuscripts, benchmark frameworks, and reproducible research artifacts spanning data integrity, Bayesian uncertainty, synthetic malware evaluation, and trustworthy cybersecurity machine learning.
-
Masked Autoencoder for Data Recovery in Polymer Research: Mitigating Data Integrity Threats
Introduces a Transformer-based masked-autoencoder method with dynamic re-masking for recovering corrupted or missing polymer research data while preserving complex feature relationships.
Autoencoders Data Recovery Data Integrity Transformers -
Towards Reliable and Interference-Aware CSI Feedback with Bayesian Neural Network
Develops a Bayesian CSI-feedback framework that reconstructs channel-state information while using predictive uncertainty to identify interference and trigger corrective action.
Bayesian Neural Networks Uncertainty CSI Feedback Interference Detection -
GenCyberSynth Phase-1: A Repaired, Reproducible, Seed-Aware Comparative Benchmark for Synthetic Malware Images Across Two Datasets
Establishes an artifact-backed benchmark comparing seven generative-model families through matched preprocessing, fixed budgets, three-seed evaluation, held-out real-data utility, KID, and MS-SSIM.
Synthetic Malware Images Benchmarking Downstream Utility Reproducibility -
GenCyberSynth Reproducible Evaluation Framework
A modular research framework for training, sampling, evaluating, aggregating, and auditing synthetic malware-image generators through standardized manifests, summaries, seed-aware metrics, and paper-grade evidence artifacts.
Research Software HPC Workflows Artifact Provenance Evaluation Infrastructure
Research Roadmap
A staged progression from reproducible synthetic-data benchmarking to trustworthy evaluation and scalable, systems-oriented research infrastructure.
Foundational Benchmarking
Establish realistic, leakage-safe, and reproducible synthetic-malware benchmarks through standardized datasets, preprocessing, model families, budgets, seeds, metrics, and downstream evaluation protocols.
Established foundationUtility & Policy
Determine when synthetic augmentation improves classification, where it fails, and how data should be allocated, filtered, ranked, selected, or rejected under constrained budgets.
Active research programExternal Conditioning & Causality
Audit whether conditional generators satisfy requested labels, diagnose class leakage and collapse, and extend observational audits toward randomized and causal analyses of external influences.
Audit foundation establishedTrustworthy Evaluation
Advance calibration, uncertainty estimation, robustness, abstention, risk-sensitive metrics, distribution-shift analysis, and reliable security decision-making.
Planned next research stageSystems & Research Infrastructure
Build reproducible, artifact-backed, performance-aware research tooling and gradually connect machine-learning evaluation with systems, compilers, runtimes, memory, parallel execution, and high-performance computing.
Long-term systems directionLooking Ahead
My long-term vision is to connect trustworthy machine-learning research with systems-level impact. I aim to build reproducible research infrastructure, performance-aware tooling, and scalable evaluation frameworks that make security models easier to audit, reproduce, optimize, and deploy reliably. As my systems expertise deepens, this work will increasingly connect with memory, parallel execution, compilers, runtimes, and high-performance computing.