arXiv subjects

Hien Tran

Publications and source records attributed to Hien Tran.

11 recordsLinked to original sources

arXiv2026

Zero-shot Transfer of Reinforcement Learning Control Policies for the Swing-Up and Stabilization of a Cart-Pole System

Reinforcement learning (RL) is a powerful and convenient tool to modernize controller design. In this work, we study the zero-shot transfer of RL-based control policies from simulation to hardware for cart-pole swing-up and stabilization. The two policies are trained independently, and the handoff is implemented in Simulink via switching logic. We apply a first-order action smoothing filter to prevent hardware damage from high-frequency oscillatory actuation. Pairing this bandwidth-aware filtering with sensitivity-guided domain randomization (DR) and a simple linear curriculum learning (CL) schedule, we obtain a swing-up policy that in all of our experiments injects sufficient energy for handoff into the stabilizer's region of attraction. The stabilization policy rejects disturbances within the tested range, and the swing-up policy can re-engage after larger perturbations and restores the pendulum to the inverted position.

cs.RO↗

arXiv2025

Control Synthesis with Reinforcement Learning: A Modeling Perspective

Controllers designed with reinforcement learning can be sensitive to model mismatch. We demonstrate that designing such controllers in a virtual simulation environment with an inaccurate model is not suitable for deployment in a physical setup. Controllers designed using an accurate model is robust against disturbance and small mismatch between the physical setup and the mathematical model derived from first principles; while a poor model results in a controller that performs well in simulation but fails in physical experiments. Sensitivity analysis is used to justify these discrepancies and an empirical region of attraction estimation help us visualize their robustness.

eess.SY↗

arXiv2025

Detection of Suicidal Risk on Social Media: A Hybrid Model

Suicidal thoughts and behaviors are increasingly recognized as a critical societal concern, highlighting the urgent need for effective tools to enable early detection of suicidal risk. In this work, we develop robust machine learning models that leverage Reddit posts to automatically classify them into four distinct levels of suicide risk severity. We frame this as a multi-class classification task and propose a RoBERTa-TF-IDF-PCA Hybrid model, integrating the deep contextual embeddings from Robustly Optimized BERT Approach (RoBERTa), a state-of-the-art deep learning transformer model, with the statistical term-weighting of TF-IDF, further compressed with PCA, to boost the accuracy and reliability of suicide risk assessment. To address data imbalance and overfitting, we explore various data resampling techniques and data augmentation strategies to enhance model generalization. Additionally, we compare our model's performance against that of using RoBERTa only, the BERT model and other traditional machine learning classifiers. Experimental results demonstrate that the hybrid model can achieve improved performance, giving a best weighted $F_{1}$ score of 0.7512.

cs.CL↗

arXiv2023

An optimal control approach for the treatment of hepatitis C patients

In this article, the feasibility of using optimal control theory will be studied to develop control theoretic methods for personalized treatment of HCV patients. The mathematical model for HCV progression includes compartments for healthy hepatocytes, infected hepatocytes, infectious virions and noninfectious virions. Methodologies have been used from optimal control theory to design and synthesize an open-loop control based treatment regimen for HCV dynamics.

math.OC↗

arXiv2022

Predicting Mutual Funds' Performance using Deep Learning and Ensemble Techniques

Predicting fund performance is beneficial to both investors and fund managers, and yet is a challenging task. In this paper, we have tested whether deep learning models can predict fund performance more accurately than traditional statistical techniques. Fund performance is typically evaluated by the Sharpe ratio, which represents the risk-adjusted performance to ensure meaningful comparability across funds. We calculated the annualised Sharpe ratios based on the monthly returns time series data for more than 600 open-end mutual funds investing in listed large-cap equities in the United States. We find that long short-term memory (LSTM) and gated recurrent units (GRUs) deep learning methods, both trained with modern Bayesian optimization, provide higher accuracy in forecasting funds' Sharpe ratios than traditional statistical ones. An ensemble method, which combines forecasts from LSTM and GRUs, achieves the best performance of all models. There is evidence to say that deep learning and ensembling offer promising solutions in addressing the challenge of fund performance forecasting.

q-fin.ST↗

arXiv2022

A Federated Deep Learning Framework for Privacy Preservation and Communication Efficiency

Deep learning has achieved great success in many applications. However, its deployment in practice has been hurdled by two issues: the privacy of data that has to be aggregated centrally for model training and high communication overhead due to transmission of a large amount of data usually geographically distributed. Addressing both issues is challenging and most existing works could not provide an efficient solution. In this paper, we develop FedPC, a Federated Deep Learning Framework for Privacy Preservation and Communication Efficiency. The framework allows a model to be learned on multiple private datasets while not revealing any information of training data, even with intermediate data. The framework also minimizes the amount of data exchanged to update the model. We formally prove the convergence of the learning model when training with FedPC and its privacy-preserving property. We perform extensive experiments to evaluate the performance of FedPC in terms of the approximation to the upper-bound performance (when training centrally) and communication overhead. The results show that FedPC maintains the performance approximation of the models within $8.5\%$ of the centrally-trained models when data is distributed to 10 computing nodes. FedPC also reduces the communication overhead by up to $42.20\%$ compared to existing works.

cs.DC↗

arXiv2017

Nonlinear Kalman Filtering for Censored Observations

The use of Kalman filtering, as well as its nonlinear extensions, for the estimation of system variables and parameters has played a pivotal role in many fields of scientific inquiry where observations of the system are restricted to a subset of variables. However in the case of censored observations, where measurements of the system beyond a certain detection point are impossible, the estimation problem is complicated. Without appropriate consideration, censored observations can lead to inaccurate estimates. Motivated by the work of [1], we develop a modified version of the extended Kalman filter to handle the case of censored observations in nonlinear systems. We validate this methodology in a simple oscillator system first, showing its ability to accurately reconstruct state variables and track system parameters when observations are censored. Finally, we utilize the nonlinear censored filter to analyze censored datasets from patients with hepatitis C and human immunodeficiency virus.

math.DS↗

arXiv2016

The W. M. Keck Observatory infrared vortex coronagraph and a first image of HIP79124 B

An optical vortex coronagraph has been implemented within the NIRC2 camera on the Keck II telescope and used to carry out on-sky tests and observations. The development of this new L'-band observational mode is described, and an initial demonstration of the new capability is presented: a resolved image of the low-mass companion to HIP79124, which had previously been detected by means of interferometry. With HIP79124 B at a projected separation of 186.5 mas, both the small inner working angle of the vortex coronagraph and the related imaging improvements were crucial in imaging this close companion directly. Due to higher Strehl ratios and more relaxed contrasts in L' band versus H band, this new coronagraphic capability will enable high-contrast small-angle observations of nearby young exoplanets and disks on a par with those of shorter-wavelength extreme adaptive optics coronagraphs.

astro-ph.SR↗

arXiv2001

WFPC2 Images of the Central Regions of Early-Type Galaxies - I. The Data

We present high resolution R-band images of the central regions of 67 early-type galaxies obtained with the Wide Field and Planetary Camera 2 (WFPC2) aboard the Hubble Space Telescope (HST). Our sample strikingly confirms the complex morphologies of the central regions of early-type galaxies. In particular, we detect dust in 43 percent of all galaxies, and evidence for embedded stellar disks in a remarkably large fraction of 51 percent. In 14 of those galaxies the disk-like structures are misaligned with the main galaxy, suggesting that they correspond to stellar bars in S0 galaxies. We analyze the luminosity profiles of the galaxies in our sample, and classify galaxies according to their central cusp slope. To a large extent we confirm the clear dichotomy found in previous HST surveys: bright, boxy ellipticals with shallow inner cusps (`core' galaxies) on one hand and faint, disky ellipticals with steep central cusps (`power-law' galaxies) on the other hand. The advantages and shortcomings of classification schemes utilizing the extrapolated central cusp slope are discussed, and it is shown that this cusp slope might be an inadequate representation for galaxies whose luminosity profile slope changes smoothly with radius rather than resembling a broken power-law. In fact, we find evidence for an `intermediate' class of galaxies, that cannot unambiguously be classified as either core or power-law galaxies, and which have central cusp slopes and absolute magnitudes intermediate between those of core and power-law galaxies.

astro-ph↗

arXiv2000

The Physics of Blazar Optical Emission Regions I: Alignment of Optical Polarization and the VLBI Jet

We collected optical and near IR linear polarization data obtained over 20--30 years for a sample of 51 blazars. For each object, we calculated the probability that the distribution of position angles was isotropic. The distribution of these probabilities was sharply peaked, with 27 blazars showing a probability < 15% of an isotropic distribution of position angles. For these 27 objects we defined a preferred position angle. For those 17 out of 27 blazars showing a well-defined radio structure angle (jet position angle) on VLBI scales (1--3mas), we looked at the distribution of angle differences -- the optical polarization relative to the radio position angles. This distribution is sharply peaked, especially for the BL Lac objects, with alignment better than 15 degrees for half the sample. Those blazars with preferred optical position angles were much less likely to have bent jets on 1--20mas scales. These results support a shock-in-jet hypothesis for the jet optical emission regions.

astro-ph↗

arXiv2000

The Physics of Blazar Optical Emission Regions II: Magnetic Field Orientation, Viewing Angle and Beaming

For a sample of 51 blazars with extensive optical polarization data, we used circular statistics to calculate the scatter among the polarization position angles for each object. We found that this scatter is correlated with the radio core dominance. We compared this relationship with the predictions of a simple transverse shock model. The result suggests that blazar jets are likely to cover a wide range of speeds consistent with those derived from observations of superluminal motion.

astro-ph↗