Searcharxiv⌕ Search

arXiv subjects

Duy Nguyen

Publications and source records attributed to Duy Nguyen.

At least 19 recordsLinked to original sources

Exact and Approximate Maximum Likelihood Estimation for Multivariate Diffusions under High-Frequency Sampling

We establish the asymptotic theory of the exact maximum likelihood estimator (MLE) for discretely observed multivariate ergodic diffusions under high-frequency, long-span sampling, solving an open problem posed by \citet{YCW2019} and \citet{Li2013}. The drift may grow linearly, and the volatility is bounded away from zero and either bounded or proportional to $1+|x|$. When the sampling interval decays polynomially, the exact MLE is consistent, asymptotically normal and efficient, with rates $\sqrt{nΔ_n}$ for drift and $\sqrt n$ for diffusion parameters. Under an additional condition on ergodic averages, and when the time span $nΔ_n$ also grows polynomially, its moments converge as well. The proof never analyzes the exact likelihood directly: it shows that a closed-form density expansion of sufficiently high order is uniformly accurate in relative terms at every observed transition, so that the two log-likelihoods differ by $o_P(1)$ uniformly in the parameter. The same argument yields accuracy conditions under which an approximate likelihood is asymptotically equivalent to the exact one; in particular, the approximate MLE of Yang, Chen and Wan converges to the true parameter. For the delta-expansion estimator of fixed order $K\ge1$, we derive the leading bias explicitly. It shows that $nΔ_n^{K+1}\to0$ cannot be weakened when the bias constant is nonzero, and that the bias falls on the drift parameters for even $K$ and on the diffusion parameters for odd $K$. Monte Carlo experiments confirm the predicted biases.

math.ST↗

Multi-Rate Flow Matching for Point Cloud Upsampling

Flow-matching point-cloud upsamplers refine sparse inputs into dense point clouds by integrating a learned velocity field with a fixed temporal discretization. However, local reconstruction requirements vary substantially across a shape, while a shared solver budget applies the same temporal resolution everywhere. We propose Multi-Rate Flow Matching (MRFM), a patch-adaptive framework that selects an integration budget for each local patch at inference time while keeping the learned velocity field frozen. MRFM relaxes the discrete budget into a continuous variable, optimizes it online through differentiable soft rollouts using an input-derived geometric fidelity objective, and projects the result back to a discrete budget for the final solver rollout. Experiments on PU1K and PUGAN show strong reconstruction quality, while matched-budget permutation and multi-factor analyses demonstrate that the benefit depends on patch-specific allocation rather than on the step distribution alone. MRFM is accuracy-oriented and intentionally uses additional test-time optimization to obtain more suitable local temporal resolutions.

cs.CV↗

Spin-Charge Conversion in an Epitaxial Kagome Flat-Band Metal

The formation of topological flat bands near the Fermi level in kagome-structured metals has generated substantial research interest. One key consequence of this electronic structure is the strong anisotropy in charge transport, such as the significantly higher in-plane resistivity relative to the out-of-plane direction. However, the spin transport properties of these flat-band kagome metals are still largely unexplored. In this work, we investigate spin-charge conversion in CoSn, a representative flat-band kagome metal. Scanning tunneling microscopy is employed to image the CoSn surface with atomic resolution, and $dI/dV$ spectroscopy shows clear signatures of the flat bands. Using magneto-optical techniques, we quantify the spin-orbit torque (SOT) in bilayers of ferromagnetic nickel (Ni) and c-plane CoSn thin films. We find a large effective spin Hall angle of $θ_{SH} = 0.484 \pm 0.005$ within the CoSn layers, significantly exceeding the value measured for Pt of $θ_{SH} = 0.0436 \pm 0.0003$ in Ni/Pt bilayers. Density functional theory (DFT) calculations provide a microscopic understanding of the intrinsic spin Hall effect in CoSn, elucidating the sources of Berry curvature in momentum space. These findings establish kagome metals as promising materials for efficient spin-charge conversion and low-power spintronics.

cond-mat.mtrl-sci↗

An efficient stochastic generalized method of moments estimator for nonlinear moment models

We study the asymptotic properties of the nonlinear stochastic GMM algorithm proposed by Chen, Lee, Liao, Seo, Shin and Song for recursively estimating parameters defined by possibly nonlinear and overidentified moment restrictions $\E[g(X,β^*)]=0$. Each observation updates estimates of the moment Jacobian, inverse covariance matrix and Gauss--Newton preconditioner at a cost independent of sample size. For linear and nonlinear instrumental-variable moments, we derive an exact rank-two Sherman--Morrison--Woodbury update. The algorithm combines preconditioned stochastic approximation with iterate averaging. Under a preconditioned monotonicity condition, satisfied globally in linear models and locally around a consistent pilot estimate in general, we establish almost sure convergence and rates, as well as a functional central limit theorem with the efficient GMM variance. These results require four moments of the moment function and its Jacobian when the step exponent exceeds $2/3$. The inverse Gauss--Newton matrix consistently estimates this variance, and a projected linearised statistic enables online testing of overidentifying restrictions. We extend the results to $L$-dependent data with long-run variance weights under stronger moment assumptions. Simulations show that the averaged estimator approaches offline two-step GMM in accuracy and coverage at $10^6$ observations, while requiring a fraction of its computing time and fixed memory at up to $10^7$ observations. The online test achieves correct size and comparable power. In an application estimating the effect of a third child on weeks worked using $254{,}654$ women from the 1980 US Census, single-pass estimates agree with offline GMM within a fraction of a standard error on average across data orderings, and both methods yield the same overidentification-test conclusion in most orderings.

stat.ME↗

VEPA: Verified Evolving Formal Policy for Lifelong LLM Agent Guardrails

LLM agents increasingly operate over sensitive resources, such as private documents and mutable system states, which can create serious safety and security risks. Runtime guardrails should therefore make inspectable, reproducible, and deterministic allow or refuse decisions before potentially irreversible agent actions are executed. However, safety and security requirements are rarely complete before deployment, so guardrails need to keep adapting as new risks emerge. Existing adaptive guardrails learn from deployment experience, but they usually apply safety policies through LLM inference, which can lead to inconsistent decisions and ambiguous policy interpretations. Formal guardrails, on the other hand, enforce policies deterministically, but they usually rely on pre-deployment or task-specific policies and have no mechanism to adapt to newly observed risks. We introduce VEPA, a framework for the verified evolution of formal agent guardrails. VEPA represents the guardrail as a formal policy of executable symbolic rules over typed agent events. During deployment, we use an LLM guided by deployment feedback to propose policy patches that fix newly observed failures while preserving previously correct behavior. Then, a deterministic verifier only accepts policies that are structurally valid and conflict-free. Finally, an evidence-aware authority layer activates a verified rule only when it has enough supporting evidence, so uncertain updates cannot directly affect guardrail decisions. Across ASSEBench-Safety, ASSEBench-Security, and ATBench, VEPA consistently outperforms baselines while expanding the scope of deterministic formal-policy enforcement. These gains also hold under more challenging deployment conditions, including mixed streams, corrupted feedback, and longer horizons.

cs.AI↗

Efficient Record-and-Replay Arithmetic for Quantum Elliptic-Curve Point Addition

We study reversible secp256k1 point-addition circuits developed through ECDSA.Fail for Shor's elliptic-curve discrete-logarithm algorithm. Two complementary constructions improve record-and-replay GCD arithmetic: Jump-2 groups binary-GCD steps and compresses their decisions using base-5 encoding, while ping-pong uses fixed register alternation and one-bit decisions to avoid full-width comparisons and data-dependent swaps. Fused replay combines doubling and signed addition into one modular correction. Both constructions support quantum-addressed window selection with measurement-based lookup cleanup. We compare three circuits on 100,000 fresh inputs across nine lookup-table configurations. A separately tested repair uses 1,419 qubits and 1.356 million mean executed Toffolis, with no detected failures on another 100,000 inputs. Structured supported-input counterexamples remain, so these tests do not establish all-input correctness. We also provide conditional coherent-error analysis and reversible safegcd comparisons. Under the stated window allowance, repaired-circuit resources lie below Google's low-gate caps and Schrottenloher's low-gate estimates, but differing accounting and correctness evidence preclude formal dominance. The results concern individual window-selected additions, not complete Shor computations.

quant-ph↗

Turning Safety into Competence: Minimally Exploitable Robot Policies via Safety-Filtered Reinforcement Learning

Robots deployed for competitive tasks must outmaneuver their opponents without sacrificing safety. Existing approaches, including safe reinforcement learning (RL), train a single policy to achieve task success and avoid failures simultaneously. This coupling can complicate training and leave the learned policy exploitable by deliberate attacks. We propose Safety to Competence (S2C), a two-stage RL framework that separates safety synthesis from competitive task learning. We formulate competitive interactions as safety-critical Markov games and prove that perfect filtering preserves policy non-exploitability when all players commit to safe maneuvers. S2C learns a robust safety filter via adversarial RL, embeds it in the environment during task policy training, and retains the same filter at deployment. In simulated touchdown games, S2C outperforms eight safe RL baselines, achieving the highest win rate and Elo rating, and the lowest exploitability. Hardware stress tests against a human opponent confirm S2C's competence.

cs.RO↗

Broken site symmetry of Fe adatoms on Bi$_2$Te$_3$

We report a combined scanning tunneling microscopy and atomistic theoretical study of Fe adatoms on the Bi$_2$Te$_3$(111) surface. Topographic imaging at $4.5$~K shows that Fe adatoms in fcc and hcp hollow sites exhibit a threefold-symmetric contrast, consistent with the $C_{3v}$ symmetry of the adsorption site. However, simultaneously acquired differential conductance ($dI/dV$) maps reveal a pronounced reduction in symmetry, evidenced by differential contrast observed at nearest-neighbor Te sites. Density functional theory calculations show that the Fe/Bi$_2$Te$_3$ system undergoes a static Jahn--Teller distortion, reducing the adsorption symmetry from $C_{3v}$ to $C_{1v}$, with the distorted configuration favored by $72.5$~meV. Orbital-projected density of states calculations show that the occupied states near the Fermi level are dominated by $d_{xz}$ and $d_{yz}$ orbitals, whereas the unoccupied states are primarily of $d_{z^2}$, $d_{x^2-y^2}$, and $d_{xy}$ character. The local density of states from these orbitals is in good qualitative agreement with experimental $dI/dV$ spectra. Furthermore, simulated local-density-of-states maps using a tight-binding Green's function approach are consistent with experimental $dI/dV$ maps, confirming that the reduced symmetry originates from the $C_{1v}$ structural distortion.

cond-mat.mes-hall↗

Multi-Attribute Steering of Language Models via Targeted Intervention

Inference-time intervention (ITI) has emerged as a promising method for steering large language model (LLM) behavior in a particular direction (e.g., improving helpfulness) by intervening on token representations without costly updates to the LLM's parameters. However, existing ITI approaches fail to scale to multi-attribute settings with conflicts, such as enhancing helpfulness while also reducing toxicity. To address this, we introduce Multi-Attribute Targeted Steering (MAT-Steer), a novel steering framework designed for selective token-level intervention across multiple attributes. MAT-Steer learns steering vectors using an alignment objective that shifts the model's internal representations of undesirable outputs closer to those of desirable ones while enforcing sparsity and orthogonality among vectors for different attributes, thereby reducing inter-attribute conflicts. We evaluate MAT-Steer in two distinct settings: (i) on question answering (QA) tasks where we balance attributes like truthfulness, bias, and toxicity; (ii) on generative tasks where we simultaneously improve attributes like helpfulness, correctness, and coherence. MAT-Steer outperforms existing ITI and parameter-efficient fine-tuning approaches across both task types (e.g., 3% average accuracy gain across QA tasks and 55.82% win rate against the best ITI baseline).

cs.CL↗

GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs

Inference-time steering methods offer a lightweight alternative to fine-tuning large language models (LLMs) and vision-language models (VLMs) by modifying internal activations at test time without updating model weights. However, most existing approaches rely on fixed, global intervention vectors, overlook the causal influence of individual input tokens, and fail to leverage informative gradients from the model's logits, particularly in multimodal settings where visual and textual inputs contribute unevenly. To address these limitations, we introduce GrAInS, an inference-time steering approach that operates across both language-only and vision-language models and tasks. GrAInS uses contrastive, gradient-based attribution via Integrated Gradients to identify the top-k most influential tokens, both positively and negatively attributed based on their contribution to preferred versus dispreferred outputs. These tokens are then used to construct directional steering vectors that capture semantic shifts from undesirable to desirable behavior. During inference, GrAInS adjusts hidden activations at transformer layers guided by token-level attribution signals, and normalizes activations to preserve representational scale. This enables fine-grained, interpretable, and modular control over model behavior, without retraining or auxiliary supervision. Empirically, GrAInS consistently outperforms both fine-tuning and existing steering baselines: it achieves a 13.22% accuracy gain on TruthfulQA using Llama-3.1-8B, reduces hallucination rates on MMHal-Bench from 0.624 to 0.514 with LLaVA-1.6-7B, and improves alignment win rates on SPA-VL by 8.11%, all while preserving the model's fluency and general capabilities.

cs.CL↗

OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers

Diffusion transformers (DiTs) achieve state-of-the-art image and video generation, but their multi-step sampling and growing parameter count make inference expensive. Post-training quantization (PTQ) is the natural remedy, yet DiT activations shift across timesteps, prompts, and guidance branches, forcing prior methods to re-fit calibration data for every new checkpoint or modality. We present OrbitQuant, a data-agnostic weight-activation quantizer that bypasses range estimation by quantizing in a normalized, rotated basis. In this basis, a randomized permuted block-Hadamard (RPBH) rotation concentrates each coordinate around one fixed, known marginal regardless of the input, so a single Lloyd-Max codebook serves all timesteps, prompts, and layers of a given input dimension. We extend the same quantizer to weight rows offline, absorbing the rotation into the weights so that it cancels inside each linear layer and only a forward rotation on the activations remains at runtime. The same recipe transfers from image to video with no per-modality tuning. Across FLUX.1, Z-Image-Turbo, Wan 2.1, and CogVideoX, it sets the state of the art for PTQ at several low-bit settings. It also pushes PTQ of image diffusion transformers to W2A4 with usable generation quality.

cs.CV↗

CodeGraphVLP: Code-as-Planner Meets Semantic-Graph State for Non-Markovian Vision-Language-Action Models

Vision-Language-Action (VLA) models promise generalist robot manipulation, but are typically trained and deployed as short-horizon policies that assume the latest observation is sufficient for action reasoning. This assumption breaks in non-Markovian long-horizon tasks, where task-relevant evidence can be occluded or appear only earlier in the trajectory, and where clutter and distractors make fine-grained visual grounding brittle. We present CodeGraphVLP, a hierarchical framework that enables reliable long-horizon manipulation by combining a persistent semantic-graph state with an executable code-based planner and progress-guided visual-language prompting. The semantic-graph maintains task-relevant entities and relations under partial observability. The synthesized planner executes over this semantic-graph to perform efficient progress checks and outputs a subtask instruction together with subtask-relevant objects. We use these outputs to construct clutter-suppressed observations that focus the VLA executor on critical evidence. On real-world non-Markovian tasks, CodeGraphVLP improves task completion over strong VLA baselines and history-enabled variants while substantially lowering planning latency compared to VLM-in-the-loop planning. We also conduct extensive ablation studies to confirm the contributions of each component.

cs.RO↗

AVSD: Adaptive-View Self-Distillation by Balancing Consensus and Teacher-Specific Privileged Signals

Self-distillation enables language models to learn on-policy from their own trajectories by using the same model as both student and teacher, with the teacher being conditioned on privileged information unavailable to the student. Such information can come in different types or views, such as solutions, demonstrations, feedback, or final answers. This setup provides dense token-level feedback without relying on a separate external model, but creates a fundamental asymmetry: the teacher may rely on view-specific information that the student cannot access at inference time. Moreover, the best type of privileged information is often task-dependent, making it difficult to choose a single teacher view. In this work, we address both these challenges jointly by introducing AVSD (Adaptive-View Self-Distillation), a novel method of self-distillation with multiple privileged-information views, which reconstructs token-level supervision by separating stable cross-view consensus from view-specific residual signals. AVSD identifies the consensus signal shared across views, which provides a reliable update direction, and then selectively adds the view-specific residual signal to adjust the update magnitude when it both aligns with the consensus direction and remains proportionate to the consensus signal. Experiments on math competition benchmarks (AIME24, AIME25, and HMMT25) show that AVSD consistently outperforms both single-view self-distillation baselines and GRPO, achieving average Avg@8 gains of 3.1% and 2.2% over the strongest baselines on Qwen3-8B and Qwen3-4B, respectively. Moreover, on code-generation benchmarks (Codeforces, LiveCodeBench v6) using Qwen3-8B, AVSD outperforms the single-view self-distillation baseline by 2.4% on average.

cs.LG↗

Self-Supervised Learning via Flow-Guided Neural Operator on Time-Series Data

Self-supervised learning (SSL) is a powerful paradigm for learning from unlabeled time-series data. However, popular methods such as masked autoencoders (MAEs) rely on reconstructing inputs from a fixed, predetermined masking ratio. Instead of this static design, we propose treating the corruption level as a new degree of freedom for representation learning, enhancing flexibility and performance. To achieve this, we introduce the Flow-Guided Neural Operator (FGNO), a novel framework combining operator learning with flow matching for SSL training. FGNO learns mappings in functional spaces by using Short-Time Fourier Transform to unify different time resolutions. We extract a rich hierarchy of features by tapping into different network layers and flow times that apply varying strengths of noise to the input data. This enables the extraction of versatile representations, from low-level patterns to high-level global features, using a single model adaptable to specific tasks. Unlike prior generative SSL methods that use noisy inputs during inference, we propose using clean inputs for representation extraction while learning representations with noise; this eliminates randomness and boosts accuracy. We evaluate FGNO across three biomedical domains, where it consistently outperforms established baselines. Our method yields up to 35% AUROC gains in neural signal decoding (BrainTreeBank), 16% RMSE reductions in skin temperature prediction (DREAMT), and over 20% improvement in accuracy and macro-F1 on SleepEDF under low-data regimes. These results highlight FGNO's robustness to data scarcity and its superior capacity to learn expressive representations for diverse time series.

cs.LG↗

Conflict-Resolving and Sharpness-Aware Minimization for Generalized Knowledge Editing with Multiple Updates

Large language models (LLMs) rely on internal knowledge to solve many downstream tasks, making it crucial to keep them up to date. Since full retraining is expensive, prior work has explored efficient alternatives such as model editing and parameter-efficient fine-tuning. However, these approaches often break down in practice due to poor generalization across inputs, limited stability, and knowledge conflict. To address these limitations, we propose the CoRSA (Conflict-Resolving and Sharpness-Aware Minimization) training framework, a parameter-efficient, holistic approach for knowledge editing with multiple updates. CoRSA tackles multiple challenges simultaneously: it improves generalization to different input forms and enhances stability across multiple updates by minimizing loss curvature, and resolves conflicts by maximizing the margin between new and prior knowledge. Across three widely used fact editing benchmarks, CoRSA achieves significant gains in generalization, outperforming baselines with average absolute improvements of 12.42% over LoRA and 10% over model editing methods. With multiple updates, it maintains high update efficacy while reducing catastrophic forgetting by 27.82% compared to LoRA. CoRSA also generalizes to the code domain, outperforming the strongest baseline by 5.48% Pass@5 in update efficacy.

cs.LG↗

VirDA: Reusing Backbone for Unsupervised Domain Adaptation with Visual Reprogramming

Existing UDA pipelines fine-tune already well-trained backbone parameters for every new source-and-target pair, resulting in the number of training parameters and storage memory growing linearly with each new pair, and also preventing the reuse of these well-trained backbone parameters. Inspired by recent implications that existing backbones have textural biases, we propose making use of domain-specific textural bias for domain adaptation via visual reprogramming, namely VirDA. Instead of fine-tuning the full backbone, VirDA prepends a domain-specific visual reprogramming layer to the backbone. This layer produces visual prompts that act as an added textural bias to the input image, adapting its "style" to a target domain. To optimize these visual reprogramming layers, we use multiple objective functions that optimize the intra- and inter-domain distribution differences when domain-adapting visual prompts are applied. This process does not require modifying the backbone parameters, allowing the same backbone to be reused across different domains. We evaluate VirDA on Office-31 and obtain 92.8% mean accuracy with only 1.5M trainable parameters. VirDA surpasses PDA, the state-of-the-art parameter-efficient UDA baseline, by +1.6% accuracy while using just 46% of its parameters. Compared with full-backbone fine-tuning, VirDA outperforms CDTrans and FixBi by +0.2% and +1.4%, respectively, while requiring only 1.7% and 2.8% of their trainable parameters. Relative to the strongest current methods (PMTrans and TVT), VirDA uses ~1.7% of their parameters and trades off only 2.2% and 1.1% accuracy, respectively.

cs.CV↗

Distributional Surgery for Language Model Activations

Language models, while capable of generating remarkably coherent and seemingly accurate text, can occasionally produce undesirable content, including harmful or toxic outputs. In this paper, we present a new two-stage approach to detect and mitigate undesirable content generations by rectifying activations. First, we train an ensemble of layerwise classifiers to detect undesirable content using activations by minimizing a smooth surrogate of the risk-aware score. Then, for detected undesirable contents, we propose layerwise distributional steering policies that transform the attention heads. These policies are computed through principled semidefinite programming, which aims to minimally perturb the attention distribution while probabilistically guaranteeing the effectiveness of the editions. Empirical evaluations across multiple language models and datasets show that our method outperforms baselines in reducing the generation of undesirable output.

cs.LG↗

LASeR: Learning to Adaptively Select Reward Models with Multi-Armed Bandits

Reward Models (RMs) are crucial to aligning large language models (LLMs), but the degree to which an RM specialized to one task (e.g. writing) generalizes to new tasks (e.g. math) is often not known a priori, often making using only one fixed RM to train LLMs suboptimal. However, optimizing LLMs with multiple RMs simultaneously can incur a prohibitively high computational cost and lead to conflicting signals from different RMs that may degrade performance. To address these challenges, we introduce LASeR (Learning to Adaptively Select Rewards), which frames reward model selection as a multi-armed bandit problem, efficiently and iteratively training LLMs using multiple RMs by selecting the most well-suited RM for each instance. On commonsense and math reasoning tasks, we show that LASeR boosts iterative LLM training, improving the absolute average accuracy of Llama-3-8B over three datasets by 2.67% over an ensemble of RM scores while also showing superior efficiency (e.g., a 2x speedup). Moreover, on WildChat (open-ended instruction-following tasks), LASeR leads to a 72.69% AlpacaEval win rate over the RM score ensemble baseline. Extending to long-context generation, LASeR improves by 2.96 F1 points (avg.) on single-document QA tasks and 2.97 F1 points on few-shot learning over the RM score ensemble baseline with best-of-n sampling.

cs.CL↗