arXiv subjects

Thanh Nguyen

Publications and source records attributed to Thanh Nguyen.

At least 37 records · Page 2Linked to original sources

arXiv2026

Automatic Identification of Compounds in Molecular Mixtures from Liquid-Phase Infrared Spectra

Interpreting spectroscopy data is a critical bottleneck in automating chemical research and industrial characterization. Particularly within infrared (IR) spectroscopy, identifying compounds in complex, liquid-phase chemical mixtures largely relies on expert knowledge, as variable peak assignment, broadening, and shifts hinder data-driven methods. Here, we show that an algorithmic approach can identify components in both simulated and experimental mixture spectra with high accuracy despite nonlinearities in liquid-phase IR data. The method is comprehensively benchmarked with a dataset of over 44,000 simulated liquid-phase IR spectra for mixtures and achieves up to 90% accuracy in identifying molecular components across a dataset of binary and ternary liquid mixtures. Our strategy is robust to perturbation of spectra, and its accuracy is capped by near-identical liquid-phase IR spectra that limit the resolution of chemical identification, imposing theoretical limits on achieving perfect accuracy in structure identification. Finally, we apply the method to automatically interpret IR spectra in experimental settings, correctly identifying the components of nearly all samples within a blind study. This work provides tools and data to advance automated chemical laboratories through algorithmic interpretation of liquid-phase IR spectra of mixtures.

cond-mat.mtrl-sci↗

arXiv2026

Uncertainty-Aware Rank-One MIMO Q Network Framework for Accelerated Offline Reinforcement Learning

Offline reinforcement learning (RL) has garnered significant interest due to its safe and easily scalable paradigm. However, training under this paradigm presents its own challenge: the extrapolation error stemming from out-of-distribution (OOD) data. Existing methodologies have endeavored to address this issue through means like penalizing OOD Q-values or imposing similarity constraints on the learned policy and the behavior policy. Nonetheless, these approaches are often beset by limitations such as being overly conservative in utilizing OOD data, imprecise OOD data characterization, and significant computational overhead. To address these challenges, this paper introduces an Uncertainty-Aware Rank-One Multi-Input Multi-Output (MIMO) Q Network framework. The framework aims to enhance Offline Reinforcement Learning by fully leveraging the potential of OOD data while still ensuring efficiency in the learning process. Specifically, the framework quantifies data uncertainty and harnesses it in the training losses, aiming to train a policy that maximizes the lower confidence bound of the corresponding Q-function. Furthermore, a Rank-One MIMO architecture is introduced to model the uncertainty-aware Q-function, \TP{offering the same ability for uncertainty quantification as an ensemble of networks but with a cost nearly equivalent to that of a single network}. Consequently, this framework strikes a harmonious balance between precision, speed, and memory efficiency, culminating in improved overall performance. Extensive experimentation on the D4RL benchmark demonstrates that the framework attains state-of-the-art performance while remaining computationally efficient. By incorporating the concept of uncertainty quantification, our framework offers a promising avenue to alleviate extrapolation errors and enhance the efficiency of offline RL.

cs.LG↗

arXiv2026

Systematic Trend-Following with Adaptive Portfolio Construction: Enhancing Risk-Adjusted Alpha in Cryptocurrency Markets

Cryptocurrency markets exhibit pronounced momentum effects and regime-dependent volatility, presenting both opportunities and challenges for systematic trading strategies. We propose AdaptiveTrend, a multi-component algorithmic trading framework that integrates high-frequency trend-following on 6-hour intervals with monthly adaptive portfolio construction and asymmetric long-short capital allocation. Our framework introduces three key innovations: (1) a dynamic trailing stop mechanism calibrated to intra-day volatility regimes, (2) a rolling Sharpe-ratio-based asset selection procedure with market-capitalization-aware filtering, and (3) a theoretically motivated asymmetric 70/30 long-short allocation scheme grounded in the empirical positive drift of crypto markets. Through extensive out-of-sample backtesting across 150+ cryptocurrency pairs over a 36-month evaluation window (2022-2024), AdaptiveTrend achieves an annualized Sharpe ratio of 2.41, a maximum drawdown of -12.7%, and a Calmar ratio of 3.18, significantly outperforming benchmark trend-following strategies (TSMOM, time-series momentum) and equal-weighted buy-and-hold portfolios. We further conduct rigorous robustness analyses including parameter sensitivity, transaction cost modeling, and regime-conditional performance decomposition, demonstrating the strategy's resilience across bull, bear, and sideways market conditions.

cs.CE↗

arXiv2025

Sharing with Frictions: Limited Transfers and Costly Inspections

The radio spectrum suitable for commercial wireless services is limited. A portion of the radio spectrum has been reserved for institutions using it for non-commercial purposes such as federal agencies, defense, public safety bodies and scientific institutions. In order to operate efficiently, these incumbents need clean spectrum access. However, commercial users also want access, and granting them access may materially interfere with the existing activity of the incumbents. Conventional market based mechanisms for allocating scarce resources in this context are problematic. Allowing direct monetary transfers to and from public or scientific institutions risks distorting their non-commercial mission. Moreover, often only the incumbent knows the exact value of the interference it experiences, and, likewise, only commercial users can predict accurately the expected monetary outcome from sharing the resource. Thus, our problem is to determine the efficient allocation of resources in the presence of private information without the use of direct monetary transfers. The problem is not unique to spectrum. Other resources that governments hold in trust share the same feature. We propose a novel mechanism design formulation of the problem, characterize the optimal mechanism and describe some of its qualitative properties.

econ.TH↗

arXiv2025

A Gray Literature Study on Fairness Requirements in AI-enabled Software Engineering

Today, with the growing obsession with applying Artificial Intelligence (AI), particularly Machine Learning (ML), to software across various contexts, much of the focus has been on the effectiveness of AI models, often measured through common metrics such as F1- score, while fairness receives relatively little attention. This paper presents a review of existing gray literature, examining fairness requirements in AI context, with a focus on how they are defined across various application domains, managed throughout the Software Development Life Cycle (SDLC), and the causes, as well as the corresponding consequences of their violation by AI models. Our gray literature investigation shows various definitions of fairness requirements in AI systems, commonly emphasizing non-discrimination and equal treatment across different demographic and social attributes. Fairness requirement management practices vary across the SDLC, particularly in model training and bias mitigation, fairness monitoring and evaluation, and data handling practices. Fairness requirement violations are frequently linked, but not limited, to data representation bias, algorithmic and model design bias, human judgment, and evaluation and transparency gaps. The corresponding consequences include harm in a broad sense, encompassing specific professional and societal impacts as key examples, stereotype reinforcement, data and privacy risks, and loss of trust and legitimacy in AI-supported decisions. These findings emphasize the need for consistent frameworks and practices to integrate fairness into AI software, paying as much attention to fairness as to effectiveness.

cs.SE↗

arXiv2025

Talyxion: From Speculation to Optimization in Risk Managed Crypto Portfolio Allocation

Cryptocurrency trading has attracted tremendous attention from both retail and institutional investors. However, most traders fail to scale their assets under management due to fragile strategies that collapse during adverse markets. The primary causes are oversized leverage, speculative position sizing, and the absence of robust risk management or hedging mechanisms. This paper introduces Talyxion, an end to end framework for crypto portfolio allocation that shifts the paradigm from speculation to optimization. The proposed pipeline consists of four stages: universe selection, alpha backtesting, volatility aware portfolio optimization, and dynamic drawdown based risk management. By combining operations research techniques with practical risk controls, Talyxion enables scalable crypto portfolios that can withstand market downturns. In live 30 day trading on Binance Futures, the framework achieved a return on investment (ROI) of +16.68%, with the Sharpe ratio reaching 5.72 and the maximum drawdown contained at just 4.56%, demonstrating strong downside risk control. The system executed 227 trades, of which 131 were profitable, resulting in a win rate of 57.71% and a PnL of +1,137.49 USDT. Importantly, these results outperformed the buy and hold baseline (Sharpe 1.79, ROI 4.36%, MDD 4.96%) as well as several top leader copy trading bots on Binance, highlighting both the competitiveness and scalability of Talyxion in real world trading environments.

cs.OS↗

arXiv2025

Sharpe-Driven Stock Selection and Liquidiy-Constrained Portfolio Optimization: Evidence from the Chinese Equity Market

This paper develops and empirically evaluates a Sharpe-driven stock selection and liquidity-constrained portfolio optimization framework designed for the Chinese equity market. The proposed methodology integrates three sequential stages: Sharpe-ratio-based universe selection, liquidity-adjusted mean-variance optimization, and multi-layered risk management implemented within an automated trading bot. Using daily price and volume data from 2023 to 2025 across the A-share universe, the framework dynamically identifies stocks exhibiting strong risk-adjusted performance while accounting for trading frictions and liquidity asymmetries that are common in emerging markets. Empirical backtests reveal that the proposed strategy achieves an annualized return of 25 percent, a Sharpe ratio of 1.71, and a maximum drawdown of 8.2 percent. These results significantly outperform the Buy-and-Hold benchmark, which records an annualized return of 21 percent, a Sharpe ratio of 1.62, and a drawdown of 7.6 percent over the same period. The superior performance demonstrates that incorporating risk-adjusted selection and liquidity-aware constraints enhances both profitability and stability, enabling the portfolio to capture upside potential while maintaining drawdown resilience. Beyond its empirical success, this study contributes methodologically by bridging classical mean-variance theory with practical liquidity adjustments and dynamic Sharpe-based screening. The resulting system not only improves the tradability of optimized portfolios but also provides a scalable and adaptive framework for quantitative asset allocation in liquidity-sensitive markets, offering new evidence that disciplined risk-return optimization can outperform passive investment strategies in the post-2023 Chinese equity landscape.

cs.OS↗

arXiv2025

Language Models for Adult Service Website Text Analysis

Sex trafficking refers to the use of force, fraud, or coercion to compel an individual to perform in commercial sex acts against their will. Adult service websites (ASWs) have and continue to be linked to sex trafficking, offering a platform for traffickers to advertise their victims. Thus, organizations involved in the fight against sex trafficking often use ASW data when attempting to identify potential sex trafficking victims. A critical challenge in transforming ASW data into actionable insight is text analysis. Previous research using ASW data has shown that ASW ad text is important for linking ads. However, working with this text is challenging due to its extensive use of emojis, poor grammar, and deliberate obfuscation to evade law enforcement scrutiny. We conduct a comprehensive study of language modeling approaches for this application area, including simple information retrieval methods, pre-trained transformers, and custom transformer models. We demonstrate that characteristics of ASW text data allow efficient custom transformer models to be trained with relatively small GPU resources and used efficiently for inference on consumer hardware. Our custom models outperform fine-tuned variants of well-known encoder-only transformer models, including BERT-base, RoBERTa, and ModernBERT, on accuracy, recall, F1 score, and ROC AUC. We demonstrate the use of our best-performing custom configuration on three tasks related to ASW data analysis: (i) decomposing the giant component in a graph representation of ASW data, (ii) clustering ASW ad text, and (iii) using the learned token embeddings to understand the use of emojis in the illicit context we study. The models we develop represent a significant advancement in ASW text analysis, which can be leveraged in a variety of downstream applications and research.

cs.CL↗

arXiv2025

Spatiotemporal Mapping of Anisotropic Thermal Transport in GaN Thin Films via Ultrafast X-ray Diffraction

Efficient thermal management is essential for the reliability of modern power electronics, where increasing device density leads to severe heat dissipation challenges. However, in thin-film systems, thermal transport is often compromised by interfacial resistance and microscale defects introduced during synthesis or transfer, which are difficult to characterize using conventional techniques. Here we present a non-contact, spatiotemporal-resolved ultrafast x-ray diffraction method to extract in-plane thermal conductivity and thermal boundary conductance, using GaN thin films on silicon as a model system. By tracking the pump-induced lattice strain, we reconstruct the lateral heat flow dynamics and quantitatively probe thermal transport near a wrinkle defect. We uncover pronounced asymmetric heat dissipation across the wrinkle, with a four-fold reduction in the local thermal conductivity near the wrinkle and a 25% drop in interfacial conductance. Our work demonstrates that ultrafast x-ray diffraction can serve as a precise thermal metrology tool for characterizing heat transport in multilayered thin-film structures for next-generation microelectronic devices.

cond-mat.mtrl-sci↗

arXiv2025

A few good choices

A Condorcet winning set addresses the Condorcet paradox by selecting a few candidates--rather than a single winner--such that no unselected alternative is preferred to all of them by a majority of voters. This idea extends to $α$-undominated sets, which ensure the same property for any $α$-fraction of voters and are guaranteed to exist in constant size for any $α$. However, the requirement that an outsider be preferred to every member of the set can be overly restrictive and difficult to justify in many applications. Motivated by this, we introduce a more flexible notion: $(t, α)$-undominated sets. Here, each voter compares an outsider to their $t$-th most preferred member of the set, and the set is undominated if no outsider is preferred by more than an $α$-fraction of voters. This framework subsumes prior definitions, recovering Condorcet winning sets when $(t = 1, α= 1/2)$ and $α$-undominated sets when $t = 1$, and introduces a new, tunable notion of collective acceptability for $t > 1$. We establish three main results: 1. We prove that a $(t, α)$-undominated set of size $O(t/α)$ exists for all values of $t$ and $α$. 2. We show that as $t$ becomes large, the minimum size of such a set approaches $t/α$, which is asymptotically optimal. 3. In the special case $t = 1$, we improve the bound on the size of an $α$-undominated set given by Charikar, Lassota, Ramakrishnan, Vetta, and Wang (STOC 2025). As a consequence, we show that a Condorcet winning set of five candidates exists, improving their bound of six.

cs.GT↗

arXiv2025

QSM-RimDS: A detection and segmentation tool for paramagnetic rim lesions in multiple sclerosis

Paramagnetic rim lesions (PRLs) are an emerging biomarker in multiple sclerosis (MS). Manual identification and rim segmentation of PRLs on quantitative susceptibility mapping (QSM) images are time-consuming. Deep learning-based QSM-RimNet can provide automated PRL detection, but this method does not provide rim segmentation for microglial density quantification and requires precise QSM lesion masks. The purpose of this study is to develop a U-Net-based QSM-RimDS method for joint PRL detection and rim segmentation using readily available T2-weighted (T2W) fluid-attenuated inversion recovery (FLAIR) lesion masks. Two expert readers performed PRL classification and rim segmentation as the reference. Dice similarity coefficient (DSC) was used to assess the agreement between rim segmentation obtained by QSM-RimDS and the manual expert segmentation. The PRL detection performances of QSM-RimDS and QSM-RimNet were evaluated using receiver operating characteristic (ROC) and precision-recall (PR) plots in a five-fold cross validation. A total of 260 PRLs (3.3\%) and 7720 non-PRLs (96.7\%) were identified by the readers. Compared to the expert rim segmentation, QSM-RimDS provided a mean DSC of 0.57 \pm 0.02 with moderate to high agreement (DSC \leq 0.5) in 73.8pm 5.7\% of PRLs over five folds. QSM-RimDS produced better and more consistent detection performance with a mean area under curve (AUC) of 0.754 \pm 0.037 vs. 0.514 \pm 0.121 by QSM-RimNet (46.7\% improvement) on PR plots, and 0.956 \pm 0.034 vs. 0.908 \pm 0.073 (5.3\% improvement) on ROC plots. In conclusion, QSM-RimDS improves PRL detection accuracy compared to QSM-RimNet and unlike QSM-RimNet can provide reasonably accurate rim segmentation.

cs.CV↗

arXiv2025

Accurate Crop Yield Estimation of Blueberries using Deep Learning and Smart Drones

We present an AI pipeline that involves using smart drones equipped with computer vision to obtain a more accurate fruit count and yield estimation of the number of blueberries in a field. The core components are two object-detection models based on the YOLO deep learning architecture: a Bush Model that is able to detect blueberry bushes from images captured at low altitudes and at different angles, and a Berry Model that can detect individual berries that are visible on a bush. Together, both models allow for more accurate crop yield estimation by allowing intelligent control of the drone's position and camera to safely capture side-view images of bushes up close. In addition to providing experimental results for our models, which show good accuracy in terms of precision and recall when captured images are cropped around the foreground center bush, we also describe how to deploy our models to map out blueberry fields using different sampling strategies, and discuss the challenges of annotating very small objects (blueberries) and difficulties in evaluating the effectiveness of our models.

cs.CV↗

arXiv2024

Hydrogenation of acetaldehyde on interstellar ice analogs reveals limited destruction

We sought to determine which are the main hydrogenation paths of acetaldehyde (CH3CHO). As a partially unsaturated molecule, CH3CHO can have links with more hydrogenated species, like ethanol (C2H5OH) or with more unsaturated ones, like ketene (H2CCO). We used highly accurate quantum chemical calculations to determine the reaction rate constants for the CH3CHO + H/D reaction. Our theoretical results are confronted against our experiments on the hydrogenation and deuteration of CH3CHO ice. We find that acetaldehyde resists hydrogenation, with only a 10\% of conversion to products different than CH3CHO. This is due to a predominance of H-abstraction at the HCO moiety, with reaction rate constants up to four orders of magnitude higher than the next possible reaction channel, that is hydrogenation at the aldehydic carbon. The formed CH3CO radical experiences barrierless or nearly barrierless reactions in all possible reaction positions, reforming CH3CHO and creating a closed loop that protects the molecule against hydrogenation. We constrain the branching ratios for the second reaction from experiments. Our experiments agree with the calculations and from the combination of both we can explain the presence of H2CCO, CO, CH4, C2H5OH, H2CO or CH3OH as minor products at the end of the reaction. We provide recommendations for future modeling efforts. Our results show limited destruction of acetaldehyde, reinforcing the vision of this molecule as an abundant and resilient COM. From the experiments, we are not able to observe the reactive desorption of this molecule. Our results align with other modeling works, showing that the link between CH3CHO and C2H5OH is not direct. Finally, our results can explain the excess of CH3CDO found in prestellar cores.

astro-ph.GA↗

arXiv2024

Optimal operation of hole spin qubits

Hole spins in silicon or germanium quantum dots have emerged as a compelling solid-state platform for scalable quantum processors. Besides relying on well-established manufacturing technologies, hole-spin qubits feature fast, electric-field-mediated control stemming from their intrinsically large spin-orbit coupling [1, 2]. This key feature is accompanied by an undesirable susceptibility to charge noise, which usually limits qubit coherence. Here, by varying the magnetic-field orientation, we experimentally establish the existence of ``sweetlines'' in the polar-azimuthal manifold where the qubit is insensitive to charge noise. In agreement with recent predictions [3], we find that the observed sweetlines host the points of maximal driving efficiency, where we achieve fast Rabi oscillations with quality factors as high as 1200. Furthermore, we demonstrate that moderate adjustments in gate voltages can significantly shift the sweetlines. This tunability allows multiple qubits to be simultaneously made insensitive to electrical noise, paving the way for scalable qubit architectures that fully leverage all-electrical spin control. The conclusions of this experimental study, performed on a silicon metal-oxide-semiconductor device, are expected to apply to other implementations of hole spin qubits.

cond-mat.mes-hall↗

arXiv2024

Chemical pathways of SO2 with hydrogen atoms on interstellar ice analogues

Sulfur dioxide (SO2) is a sulfur-containing molecule expected to exist as a solid in the interstellar medium (ISM). In this study, we performed laboratory experiments and computational analyses on the surface reactions of solid SO2 with hydrogen atoms on amorphous solid water (ASW) at low temperatures. After 40 min of exposure of SO2 deposited on ASW to H atoms, approximately 80% of the solid SO2 was lost from the substrate at 10-40 K, and approximately 50% even at 60 K, without any definite detection of reaction products. Quantum chemical calculations suggest that H atoms preferentially add to the S atom of solid SO2, forming the HSO2 radical. Further reactions of the HSO2 radical with H atoms result in the formation of several S-bearing species, including HS(O)OH, the S(O)OH radical, HO-S-OH, HS-OH, and H2S. In codeposition experiments involving H and SO2, we confirmed the formation of H2S, HS(O)OH, and/or HO-S-OH. However, the yields of these S-bearing species were insufficient to account for the complete loss of the initial SO2 reactant. These findings suggest that some products desorbed into the gas phase upon formation. This study indicates that a portion of SO2 in ice mantles may remain unreacted, avoiding hydrogenation, while the remainder is converted into other species, some of which may be subject to chemical desorption.

astro-ph.GA↗

arXiv2024

Learning Code Preference via Synthetic Evolution

Large Language Models (LLMs) have recently demonstrated remarkable coding capabilities. However, assessing code generation based on well-formed properties and aligning it with developer preferences remains challenging. In this paper, we explore two key questions under the new challenge of code preference learning: (i) How do we train models to predict meaningful preferences for code? and (ii) How do human and LLM preferences align with verifiable code properties and developer code tastes? To this end, we propose CodeFavor, a framework for training pairwise code preference models from synthetic evolution data, including code commits and code critiques. To evaluate code preferences, we introduce CodePrefBench, a benchmark comprising 1364 rigorously curated code preference tasks to cover three verifiable properties-correctness, efficiency, and security-along with human preference. Our evaluation shows that CodeFavor holistically improves the accuracy of model-based code preferences by up to 28.8%. Meanwhile, CodeFavor models can match the performance of models with 6-9x more parameters while being 34x more cost-effective. We also rigorously validate the design choices in CodeFavor via a comprehensive set of controlled experiments. Furthermore, we discover the prohibitive costs and limitations of human-based code preference: despite spending 23.4 person-minutes on each task, 15.1-40.3% of tasks remain unsolved. Compared to model-based preference, human preference tends to be more accurate under the objective of code correctness, while being sub-optimal for non-functional objectives.

cs.LG↗

arXiv2024

Mitigating Adversarial Perturbations for Deep Reinforcement Learning via Vector Quantization

Recent studies reveal that well-performing reinforcement learning (RL) agents in training often lack resilience against adversarial perturbations during deployment. This highlights the importance of building a robust agent before deploying it in the real world. Most prior works focus on developing robust training-based procedures to tackle this problem, including enhancing the robustness of the deep neural network component itself or adversarially training the agent on strong attacks. In this work, we instead study an input transformation-based defense for RL. Specifically, we propose using a variant of vector quantization (VQ) as a transformation for input observations, which is then used to reduce the space of adversarial attacks during testing, resulting in the transformed observations being less affected by attacks. Our method is computationally efficient and seamlessly integrates with adversarial training, further enhancing the robustness of RL agents against adversarial attacks. Through extensive experiments in multiple environments, we demonstrate that using VQ as the input transformation effectively defends against adversarial attacks on the agent's observations.

cs.LG↗

arXiv2024

From Literature to Practice: Exploring Fairness Testing Tools for the Software Industry Adoption

In today's world, we need to ensure that AI systems are fair and unbiased. Our study looked at tools designed to test the fairness of software to see if they are practical and easy for software developers to use. We found that while some tools are cost-effective and compatible with various programming environments, many are hard to use and lack detailed instructions. They also tend to focus on specific types of data, which limits their usefulness in real-world situations. Overall, current fairness testing tools need significant improvements to better support software developers in creating fair and equitable technology. We suggest that new tools should be user-friendly, well-documented, and flexible enough to handle different kinds of data, helping developers identify and fix biases early in the development process. This will lead to more trustworthy and fair software for everyone.

cs.SE↗