🎓 Student Researcher | Human–AI Teams | Deep Reinforcement Learning | Digital Twins | Computer Vision

I’m an student researcher exploring how AI systems learn, adapt, and collaborate with humans in complex, uncertain environments. I investigate how intelligent systems learn, adapt, and collaborate with humans in complex, real-world environments. My work integrates Deep Learning, Reinforcement Learning, Large Language Models (LLMs), and Digital Twin simulations to design adaptive, explainable, and scalable AI systems.

• Deep Reinforcement Learning (DRL): Decision-making under uncertainty and delayed rewards
• Digital Twins & XR Simulations: Real-time integration of virtual and physical systems (Unity, HoloLens 2)
• Computer Vision: Object detection, segmentation, and non-destructive defect analysis (CNNs, Vision Transformers)
• Human–AI Collaboration: Trust, explainability, and adaptive teaming with autonomous systems
• Large Language Models (LLMs): Reasoning, instruction-following, and multimodal integration with control systems

A high-fidelity digital twin system designed for real-time scheduling in U-shaped automated container terminals.
Integrates predictive models, hybrid optimization, and adaptive control mechanisms to achieve collaborative task scheduling, reduced system latency, and improved operational efficiency within dynamic port environments.

An XR-based digital twin environment integrating real-time industrial sensor data and physics-based models.
Explores synchronization between real and virtual systems for predictive control, maintenance, and training simulations.

A deep convolutional neural network trained on the Fashion-MNIST dataset with transfer learning and data augmentation.
Demonstrates model optimization, feature visualization, and robust generalization across visual classes.

An implementation of parameter-space noise for deep reinforcement learning.
This method perturbs network parameters directly, enabling richer exploration and more stable policy learning across continuous-control tasks.

A complete Twin Delayed Deep Deterministic Policy Gradient (TD3) implementation for the BipedalWalker-v2 environment.
Demonstrates stable learning using twin critics, delayed policy updates, and target smoothing to train a walking agent efficiently.

A research project applying Deep Learning to automatic defect detection in X-ray imagery.
Targets high-precision detection of inclusions, pores, and cracks in industrial materials using explainable AI (Grad-CAM).

Languages: Python, C++, C#, MATLAB
Frameworks: PyTorch, TensorFlow, Unity ML-Agents, ROS
Libraries: OpenCV, NumPy, Pandas, Scikit-learn, Matplotlib
Platforms: HoloLens 2, Unity, Jupyter, GitHub Actions

Below is a selection of publications, research papers, and technical outputs aligned with my academic and applied research in Artificial Intelligence.

Shah, S., & Yao, N. (2023). Deep Reinforcement Learning for Unpredictability-Induced Rewards to Handle Spacecraft Landing.
Proceedings of the 13th International Conference on Information Science and Technology (ICIST).

Explores robust policy optimization for dynamic and stochastic control scenarios using DQN, Double DQN, Duel DQN and domain randomization.

M. Khalid, B Chen & S. Shah, (2025). Attention-Guided Feature Fusion with MobileNetV3 for Real-Time Vehicle Classification.
IET Computer Vision (under review).

Proposes an attention-guided feature fusion architecture using MobileNetV3 for real-time vehicle classification under varying illumination and occlusion. Achieves high inference speed and improved accuracy with lightweight model design suitable for embedded systems.

(Full publication list available upon request or via Google Scholar.)

• AI-Driven Digital Twins: Predictive analytics and control through cyber-physical synchronization
• RL under Uncertainty: Safe and robust decision-making in dynamic environments
• Explainable & Collaborative AI: Trust and interpretability in human-AI teaming

📫 LinkedIn: linkedin.com/in/salman-shah-ai
🔗 GitHub: github.com/salman-shah-ai
📄 Google Scholar: Scholar Profile

“AI is not merely automating intelligence — it is extending the boundaries of how humans and machines reason together.”
— Salman Shah