Hello, I am
Viet Dung Nguyen
PhD Candidate @ RIT
Hello, I am Viet Dung Nguyen
PhD Candidate @ RIT
Viet Dung Nguyen is a Ph.D. Candidate at Rochester Institute of Technology. He is working with Dr. Alexander G. Ororbia in theNeural Adaptive Computing Lab (The NAC Lab), Dr. Reynold Bailey in theComputer Graphics & Applied Perception Lab., and Dr. Gabriel Diaz in thePerception For Movement (PerForM) Lab.
Research Interests
(Generative World Model + Active Inference) @ Autonomous Systems
My research interests encompass multimodal deep learning, computer vision, eye tracking, reinforcement learning, embodied robotics, and natural language processing. By integrating these interdisciplinary fields, my dissertation aims to develop an embodied neuro-robotic agent capable of solving real-world tasks and interacting with humans in daily life.
Active Inference
Active inference process theory is a mathematical framework which focus on modeling human perception and action (similar to model-based reinforcement learning and involves the use of a generative world model). I mainly focus on implementing and improving world model architecture to make active inference agents more robust. I also focus on integration of active inference in real-life autonomous systems such as robot arms, drones, and autonomous driving cars.
Keywords: Generative Model, World Model, Reinforcement Learning, Model-Based, Autonomous Systems, Robotics, Multimodal
Computer Vision
Compute vision encapsulates many different sub-fields such as image segmentation, recognition, and generation. I mainly focus on domain adaptation problem -- improving models' real-world inference accuracy while replacing real-world samples with synthetic samples.
Keywords: Sim2Real, Segmentation, GAN, Diffusion Models
Natural Language Processing
I mainly focus on the construction of robotic agents/autonomous systems that act based the instruction/guidance in the form of natural language. Besides that, I focus on improving hate-speech detection models through the use of large language model.
Keywords: LLM, Hate-Speech Detection, Multimodal, Transformer, Multimodal
Highlights
y = x @ a + b
Neural approaches for addressing domain gap between real and synthetic eye-tracking data
Develop novel CycleGAN-variant methodologies to perform domain transfer while keeping the image structure. Minimized the domain shift across different eye segmentation datasets. Best Paper Award at ETRA 2024.
Keywords: CycleGAN, Pytorch
Deep multimodal active inference for robotic arm control
Developing a novel deep multimodal approach to the torque control system of robotic arms, motivated by the framework of active inference.
Keywords:
Robust active inference for continuous partially observable Markov decision processes
Developing an agent that solves robotics vision tasks through the integration of active inference process theory.
Keywords:
Deep multimodal generative model for robotic agents with language guided instruction
Keywords: Pytorch, Computer Vision, Transformer, LLM
News
Up-to-date
[Dec 1, 2024] Research Visit at VinUniversity
I'm happy to share that I will be doing research visit in VinUniversity, Vietnam.
[Nov 13, 2024] Talk: Dynamic Prior Preference Learning for Scalable, Robust Deep Active Inference
I'm happy to share my talk at the 4th Active Inference Symposium. The talk is about Dynamic Prior Preference Learning for Scalable, Robust Deep Active Inference. n this talk, we specifically analyze this issue and discuss a potential resolution for scaling the instrumental signal inherent to AIF by introducing the “contrastive recurrent state prior preference” (CRSPP) model learning framework. This methodology frames AIF agents in terms of progressively constructing and adapting a prior preference at each time step, facilitating the dynamic emission of a useful, dense instrumental signal.
[Oct 1, 2024] Research Visit at University College Dublin
I'm happy to share that I will be doing research visit in University College Dublin, Ireland.
[Aug 21, 2024] New Research Up on ArXiv
I'm happy to share that our work is now available on ArXiv! This research presents a novel active inference (AIF) framework designed to solve sparse-reward, image-based reinforcement learning (RL) tasks. We demonstrate that our agent outperforms state-of-the-art RL and AIF baselines.
Zhizhuo Yang
,Christopher L. Buckley
,Alexander Ororbia
[Jun 9, 2024] Machine Learning Intern
Congratulation Viet Dung Nguyen on starting new intern position at Petanux GmbH.
[Jun 7, 2024] Best Paper Award
Congratulation Viet Dung Nguyen and his team on his best paper award at ETRA 2024.
[Jan 17, 2024] Paper Accepted to ETRA 24
Congratulation Viet Dung Nguyen and his team on their paper accepted to ETRA 2024. Project funded by: US National Science Foundation and Meta Reality Labs.
Reynold Bailey
,Gabriel J. Diaz
,Chengyi Ma
,Alexander Fix
,Alexander Ororbia
[Oct 10, 2023] Seed Grant Accepted
Congratulation Viet Dung Nguyen on securing grant proposal in multimodal. Title: 'Deep Multimodal Active Inference for Robotic Arm Control.' Funding: $1,080
[April 15, 2023] Physics Olympiad
Congratulation Viet Dung Nguyen and his team on achieving Second College Physics Olympiad at RIT 2023
[Mar 26, 2023] Paper Accepted to AAAI 23
Congratulation Viet Dung Nguyen on paper accepted to AAAI 23.
Quan H. Nguyen
,Richard G. Freedman