W2C Research Projects

Adversarial attacks pose a critical threat to natural language processing systems by subtly manipulating input text to mislead classifiers. Emotion classification models are particularly vulnerable due to the overrepresentation of neutral labels in widely used datasets, making small perturbations disproportionately impactful. This work explores robust emotion classification using text autoencoders, large language models, and unsupervised learning to create fair and dependable systems.

This project develops a multimodal situational awareness framework for underground mine disaster environments using LiDAR point clouds, thermal imagery, low-light RGB images, gaseous sensor readings, and textual reports. The framework supports 3D mapping, localization, segmentation, hazardous gas monitoring, and emergency response decision-making.

This research proposes robust out-of-distribution detection methods using internal neural network representations. By constructing prototypes across multiple layers and computing cosine similarities, anomalous inputs can be reliably distinguished from in-distribution samples for applications including autonomous driving and medical imaging.

This research develops machine learning techniques for UAV detection, swarm formation analysis, and threat classification using visual imagery and electromagnetic signatures. The framework aims to improve situational awareness and reliable UAV identification under complex environmental conditions.

This project develops stereo vision and landmark recognition frameworks for localization and navigation in GPS-denied environments. The system utilizes deep learning-based landmark recognition and optimization techniques for reliable positioning and safe path planning in safety-critical scenarios.

This research develops reinforcement learning frameworks for intelligent and energy-efficient wireless sensor networks in underground mines. The project includes battery prediction, adaptive duty-cycle optimization, and rescue path planning to improve long-term monitoring and miner safety.

This project develops a privacy-preserving federated learning framework for detecting and mitigating backdoor attacks in mining environments. The framework integrates secure aggregation and robust adversarial defense mechanisms using datasets collected from the Missouri S&T experimental mine to improve fairness, reliability, and operational safety.

This research focuses on secure communication frameworks between Digital Twins and Additive Manufacturing systems. The framework combines lightweight encryption, CP-ABE access control, blockchain-based key management, and resilient communication protocols to defend against cyber threats in real-time manufacturing environments.

This research investigates payment mechanism design for photo crowdsensing applications using mobile devices. The proposed mechanisms incentivize worker participation while ensuring budget feasibility and maximizing contribution quality for applications such as 3D photo mapping and infrastructure monitoring.