Professional Experience

Publications

• “Advancing AI: Exploring the Potentials of Multimodal Large Language Models” CSCI’24 — Published in Springer Nature (📄 View Publication)
  📖 Read Abstract

  Abstract: Recently, the advent of multimodal large language models (MLLMs) which integrate text with other modalities such as images, audio, and video, promise to revolutionize various fields ranging from natural language processing to computer vision and beyond. However, while MLLMs have shown remarkable performance across a wide range of tasks, their inner workings remain largely opaque, presenting significant challenges in terms of interpretability, robustness, and ethical considerations. This paper investigates the next frontier in artificial intelligence research: understanding multimodal large language models. We explore the architecture and applications of MLLMs, shedding light on their capabilities and limitations. By delving into the intricacies of multimodal large language models, this paper aims to show the potential use of LLM in biomedical and advanced machine learning algorithms to extract valuable features and improve the prediction accuracy of clinical analysis. Therefore, it will pave the way for future research directions and facilitate the development of more transparent, equitable, and trustworthy AI systems.

  🔽 Read Less

---

• “Reinforcing a Monitoring System of a Regulated River using LSTM” ICAI'23 and CSCE'23 — Published in IEEE Xplore (📄 View Publication)
  📖 Read Abstract

  Abstract— Accurate methods predicting flow in a river require high performance computing systems providing only short-term early warnings. To address these limitations, we propose an optimized Long Short Term Memory layer combined with a Fully connected layer within a six-layer architecture to estimate flows based on readings of three neighboring river gage heights. This study assessed the number of cells in the LSTM and the number of neurons in the Fully connected layer as well as the probability in the Dropout layer to avoid overfitting. Identification of the best settings by means of the RMSE evaluated six different input datasets to estimate flow. Our study proved that relying on the river gage height from neighboring sensors is possible to predict flow which is crucial when a sensor may fail and a reliable neural network at low computational cost can handle that.

  🔽 Read Less

---

• “Intelligent Upper Limbs Prosthetics with 1D Convolutional Neural Networks and Quick Training” BIOENG'23 and CSCE'23 — Published in IEEE Xplore (📄 View Publication)
  📖 Read Abstract

  Abstract— Upper limb amputation can severely restrict the mobility and ability of amputees to perform daily activities. In addressing this issue, deep learning algorithms and electromyography pattern recognition have emerged as promising clinical solutions for functional upper-limb prosthetics. This article presents the use of EMG sensors to capture muscle movement signals and applies the pattern recognition function of a 1D convolutional neural network to identify these signals and control the movement of prosthetics. Experimental results demonstrate that the convolutional neural network exhibits fast training and high-precision recognition capabilities enabling it to accurately identify muscle signals and effectively control prosthetic movements.

  🔽 Read Less

---

• “Defending Characteristics and Attribution Analysis for Phishing Attacks” CSCI’23 (CSCI-RTCW) — Published in IEEE Xplore (📄 View Publication)
  📖 Read Abstract

  Abstract— In recent years, phishing attacks have become widespread cyber threats affecting people, governments, private sectors, service providers, and internet users globally. Attackers collect personal information about the targeted victims through emails or fake websites, such as user account login details, bank card numbers, etc. The victim is manipulated using a combination of technical and social engineering techniques. This paper provides a review of recent existing research on various phishing attacks, including HTTPS phishing, spear phishing, whale phishing, pop-up phishing, and email spoofing hacking. It also explores how to identify such attacks using Artificial Intelligence (AI) techniques, including Machine Learning and Deep Learning. Additionally, the paper investigates the benefits and drawbacks of different defense approaches and provides an overview of the challenges related to phishing attack detection.

  🔽 Read Less

---

• “A Literature Survey and Analysis of Defending Cyber Attacks Targeting IoT in Critical Infrastructure” CSCI’23 (CSCI-RTCW) — Published in IEEE Xplore (📄 View Publication)
  📖 Read Abstract

  Abstract— In an era defined by the relentless advancement of technology, the critical infrastructure that sustains our daily lives has become increasingly reliant on complex digital systems like Internet of Things (IoT). While these complex digital systems have brought convenience and efficiency to critical infrastructure, they have also become prime targets for malicious parties seeking to exploit vulnerabilities, raising concerns across vital sectors like energy, healthcare, transportation, and beyond. Through systematic literature review, historical incident analysis, and examination of recent attacks, this paper aims to deepen our understanding of the evolving threat landscape, emphasizing the urgency of strengthening defenses, particularly in the IoT realm to secure the foundations of modern society. This literature survey encompasses a spectrum of defending technologies and methodologies, including Artificial Intelligence (AI)-based defenses, and anomaly-based intrusion detection systems. Analyzing variations techniques and evaluates their efficiency and complexity.

  🔽 Read Less

---

• “Analytical Study of Cybersecurity Attacks and Defenses Characteristics” CSCI’23 (CSCI-RTCW) — Published in IEEE Xplore (📄 View Publication)
  📖 Read Abstract

  Abstract— As technology continues to advance rapidly, society becomes increasingly reliant on digital systems. The cybersecurity landscape becomes progressively intricate and fraught with challenges. This paper conducts an in-depth exploration of the contemporary cybersecurity threat landscape, charting its evolution and examining the defensive strategies in play. By synthesizing insights from a broad spectrum of literature and presenting real-world cases, this work comprehensively analyzes the key characteristics of various threat categories, including malware, phishing, ransomware, and denial-of-service attacks, among others. As a conclusion, this paper strongly advocates the adoption of a multi-faceted and multi-layered defense approach to yield a high accuracy and efficient cyber defense.

  🔽 Read Less

Accomplishments

Projects

Research Experience

• AI-Powered Intrusion Detection & Prevention (IDS/IPS) Ongoing
  Developing real-time defensive models using ML/DL (Random Forest, LSTM, CNN) with SHAP-based explainability and Optuna hyperparameter tuning.
• AI Security for Public-Health Surveillance: Robust Detection & Defense on CDC Case Surveillance Ongoing
  Policy-simple, auditable pipeline on monthly CDC state×month counts: Per-state 5–95% winsorization Rolling MAD YoY Residual-MAD EWMA. Produces baseline / attacked / defended streams, heatmaps, |z| distributions, and sensitivity sweeps over W∈{4,6,8}, k∈{2.0,2.5,3.0}. Results (W=6, k=2.5): −24.5% alert load (220→166) with 74.1% sensitivity retention and p95(|z|) 3.665/3.688→3.031. FOIA-compatible run manifests; no PHI/PII (server-side aggregation via CDC Open Data).
• Multimodal Large Language Model (LLM) Research
  Exploring fusion of text, audio, and visual modalities for richer contextual understanding in AI systems.
• Cybersecurity in Critical Infrastructure
  Investigated human-induced risks and defense mechanisms (e.g., phishing, ransomware) using deep neural networks within critical infrastructure domains.
• Intelligent Flood Forecasting for Kanawha River Basin
  Built a long-term, feedback-enabled flood prediction system using LSTM, BiLSTM, GRU, and ARIMA models.
• AI-Driven Robotic Prosthetic Limbs
  Designed upper-limb prosthetics using 1D CNNs on embedded Coral Accelerator hardware; accelerated training with PyTorch and GPU.
• Groundwater Level Prediction
  Developed multivariate forecasting models for groundwater trends using LSTM, BiLSTM, GRU, and ARIMA.

Contact Me