image

Thinh Hoang Dinh, Ph.D. (Jensen Hoang)

I am a postdoctoral researcher at the French Civil Aviation University (ENAC) Optimization Laboratory. This website hosts my personal blog, where I share my research, projects, and thoughts on various topics. It also serves a Graph RAG powered by GPT 4o mini.

See more.

Contact

Research Interests

My doctoral dissertation focused on function approximations and signal processing, specifically applied to vehicle trajectory data. I’ve also explored sparse optimization, large language models, and theoretical reinforcement learning.

For the moment, I’m current working on the SESAR’s DeepFlow Project under guidance of Pr. Daniel DELAHAYE. Our aim is to develop a transparent and interpretable machine learning method to enhance air traffic management.

Highlight Work

image

Helmholtz Model

The project attempts to leverage the sparse activation patterns of optimization constraints in order to simplify a large optimization problem into many smaller subproblems. This creates an alternative approach to solving the Hamilton-Jacobi-Bellman equation in reinforcement learning. As a result, the Helmholtz model can be used to align the model without the cost of generating a large amount of synthetic data.

image

StratoEye (Montéllama Codex + Web Platforms)

This project aims to leverage the power of large language models in air traffic commands comprehension and bridge the gap between Language Understanding (LU) and Trajectory Signal Processing (TSP).

This work is realized under guidance of Pr. Daniel Delahaye (ENAC), Pr. Pierre Maréchal (IMT), Laurent Lapasset (ENAC) and in collaboration with Huijuan Yang (Civil Aviation Flight University of China).

Long demonstration video

Short demonstration video

Project Description

image

Spherical Codec

The codec reduces the Channel Busy Ratio (CBR) for CAM messages exchange by up to 2.5 times. A European Patent has been filed since October ‘22.

image

Real-time Anomaly Detection

Perform real-time estimation of the “probability of anomaly” for early warning of dangerous scenarios in Air Traffic Control.

image

Leisure Project: A Productivity Tool

Using React Native, React and Firebase, I designed a small Raspberry Pi-powered smart display that shows me the to-do list with Pomodoro functions. By keeping track of the time until the next calendar event, I am also prevented from forgetting to attend meetings. By doing so, I am able to increase my productivity and stay on top of dozens of hundred tasks throughout the day.

Give it a spin: Ordoscope Beta

Recent Posts

  • Welcome to Jekyll!