arXiv subjects

Khanh Nguyen

Publications and source records attributed to Khanh Nguyen.

At least 19 recordsLinked to original sources

arXiv2026

Field Codes for Distributed Coupling Samplers and Certified Empirical Transport

In this paper, we formulate three communication tasks for empirical optimal transport: distributed coupling sampling, cost-evaluable coupling output, and scalar value-certified sampling. Our main result is a field-code compiler: any communicated transport field approximating an optimal empirical Monge map to error $\eta$ can be completed by sparse target-cell residuals into an exact-marginal value-certified sampler with scalar certificate $W_1(\mu,\nu)\leq U\leq W_1(\mu,\nu)+2\Delta$, where $\Delta$ is the public target-partition diameter. The certificate accuracy is controlled by $\Delta$ alone. The field error $\eta$ controls residual communication under a cell-margin condition; without a margin, $\eta$ alone does not bound residuals. We instantiate the compiler via adaptive local-affine and tensor-product spline codes with $d(m+1)^db$ field bits in the spline case, plus residual lists charged separately. For lower bounds, exact Gap-Hamming embeddings prove certified output is hard, including a smooth cell-packing diffeomorphism family requiring $\Omega(\varepsilon^{-2d/(d+4)})$ communication for any cost-evaluable, cost-certified, or value-certified protocol. The same gadgets admit zero-communication samplers, formally separating the sampler and certificate-bearing output models. These results identify the transport field as the right communicated object whenever a field code is available, primarily as a residual-sparsity tool.

cs.CC↗

arXiv2026

LeanFlow: A Case Study in Workflow-Driven Lean Autoformalization

We present and evaluate LeanFlow, an LLM agent system specialized for translating mathematical papers into buildable Lean projects. Recent verifier-in-the-loop systems show that large formal artifacts can be produced, but it remains unclear which runtime mechanisms affect completion, auditability, or efficiency in document-to-project formalization. We study this question through case studies on two previously unformalized mathematical papers in number theory and measure theory, using model, proof-workflow, and toolset ablations with Kimi2.6 and GPT5.5; we report task outcome, API calls, input tokens, and output tokens. With Kimi2.6, the full workflow completes both document-level projects within the 2000-call budget, while no-queue variants reach the budget limit; with GPT5.5, all document-level variants complete, and the full workflow has the lowest or tied-lowest input-token cost on both sources. As complementary calibration, LeanFlow reaches 75.7% BEq+ on the PFR slice of RLM25 and solves all five ICML 2026 AI for Math TCS challenge projects in our GPT5.5 runs.

cs.AI↗

arXiv2026

Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence

We introduce Nemotron 3 Nano Omni, the latest model in the Nemotron multimodal series and the first to natively support audio inputs alongside text, images, and video. Nemotron 3 Nano Omni delivers consistent accuracy improvements over its predecessor, Nemotron Nano V2 VL, across all modalities, enabled by advances in architecture, training data and recipes. In particular, Nemotron 3 delivers leading results in real-world document understanding, long audio-video comprehension, and agentic computer use. Built on the highly efficient Nemotron 3 Nano 30B-A3B backbone, Nemotron 3 Nano Omni further incorporates innovative multimodal token-reduction techniques to deliver substantially lower inference latency and higher throughput than other models of similar size. We are releasing model checkpoints in BF16, FP8, and FP4 formats, along with portions of the training data and codebase to facilitate further research and development.

cs.LG↗

arXiv2026

Exceptional Sets for Quasiconformal Mappings in General Metric Spaces II

A homemorphism between domains in $\mathbb R^n$, $n\ge 2$ is quasiconformal, with its intricate analytic and geometric consequences, if the (pointwise) linear dilatation -- a purely metric quantity -- is uniformly bounded. Gehring proved that it will suffice to verify the uniform bound up to a set of measure zero as long as we can show that the dilatation is finite outside a subset of finite Hausdorff--$(n-1)$ measure. In short, we say that we can allow an exceptional codimension $1$ subset. In the metric setting, it has been proved, roughly speaking, that one can allow an exceptional codimension $p$ subset, $p \ge 1$, if the source space satisfies a $p$-Poincar\'e inequality. We prove, effectively, the sharpness of the latter claim.

math.FA↗

arXiv2026

Weak Harnack inequality and Cartan property for nonlocal $W^{s,1}$-minimizers

We establish a weak Harnack inequality for nonlocal $W^{s,1}$-subminimizers in a complete, connected, doubling metric measure space where $0<s<1$. As a corollary, we prove that $W^{s,1}$-subminimizers are semicontinuous, up to a suitable choice of pointwise representative. We then prove \emph{Cartan-type properties} for $W^{s,1}$-superminimizers. The theory turns out to be mostly analogous with the local case of BV super- and subminimizers. Our results seem to be new even in the classical Euclidean setting.

math.AP↗

arXiv2026

A General and Robust 3D Finite Element Dynamics Framework for Railway Vehicle-Bridge Interaction with Nonlinear Wheel-Rail Contact Modeling

A key challenge in 3D finite element models of coupled railway vehicle-bridge dynamics is the rigorous definition of kinematic constraints and the development of an efficient, robust solution. This paper presents a novel approach that can be implemented in general finite element software using constraint equations tailored to wheel-rail contact behavior, essential for analyzing lateral vehicle-bridge interactions. The method employs absolute coordinates to describe the motion of nodes defining the track position and orientation for each wheelset, without assuming infinitesimal displacements or rotations. This general formulation enables realistic simulations of extreme scenarios involving large lateral movements caused by strong winds or earthquakes. The proposed wheel-rail contact model is first validated against published results, and a 3D numerical example demonstrates the method's performance and capabilities.

math.NA↗

arXiv2026

RI-Mamba: Rotation-Invariant Mamba for Robust Text-to-Shape Retrieval

3D assets have rapidly expanded in quantity and diversity due to the growing popularity of virtual reality and gaming. As a result, text-to-shape retrieval has become essential in facilitating intuitive search within large repositories. However, existing methods require canonical poses and support few object categories, limiting their real-world applicability where objects can belong to diverse classes and appear in random orientations. To address this challenge, we propose RI-Mamba, the first rotation-invariant state-space model for point clouds. RI-Mamba defines global and local reference frames to disentangle pose from geometry and uses Hilbert sorting to construct token sequences with meaningful geometric structure while maintaining rotation invariance. We further introduce a novel strategy to compute orientational embeddings and reintegrate them via feature-wise linear modulation, effectively recovering spatial context and enhancing model expressiveness. Our strategy is inherently compatible with state-space models and operates in linear time. To scale up retrieval, we adopt cross-modal contrastive learning with automated triplet generation, allowing training on diverse datasets without manual annotation. Extensive experiments demonstrate RI-Mamba's superior representational capacity and robustness, achieving state-of-the-art performance on the OmniObject3D benchmark across more than 200 object categories under arbitrary orientations. Our code will be made available at https://github.com/ndkhanh360/RI-Mamba.git.

cs.CV↗

arXiv2025

Retrieving Objects from 3D Scenes with Box-Guided Open-Vocabulary Instance Segmentation

Locating and retrieving objects from scene-level point clouds is a challenging problem with broad applications in robotics and augmented reality. This task is commonly formulated as open-vocabulary 3D instance segmentation. Although recent methods demonstrate strong performance, they depend heavily on SAM and CLIP to generate and classify 3D instance masks from images accompanying the point cloud, leading to substantial computational overhead and slow processing that limit their deployment in real-world settings. Open-YOLO 3D alleviates this issue by using a real-time 2D detector to classify class-agnostic masks produced directly from the point cloud by a pretrained 3D segmenter, eliminating the need for SAM and CLIP and significantly reducing inference time. However, Open-YOLO 3D often fails to generalize to object categories that appear infrequently in the 3D training data. In this paper, we propose a method that generates 3D instance masks for novel objects from RGB images guided by a 2D open-vocabulary detector. Our approach inherits the 2D detector's ability to recognize novel objects while maintaining efficient classification, enabling fast and accurate retrieval of rare instances from open-ended text queries. Our code will be made available at https://github.com/ndkhanh360/BoxOVIS.

cs.CV↗

arXiv2025

NVIDIA Nemotron Nano V2 VL

We introduce Nemotron Nano V2 VL, the latest model of the Nemotron vision-language series designed for strong real-world document understanding, long video comprehension, and reasoning tasks. Nemotron Nano V2 VL delivers significant improvements over our previous model, Llama-3.1-Nemotron-Nano-VL-8B, across all vision and text domains through major enhancements in model architecture, datasets, and training recipes. Nemotron Nano V2 VL builds on Nemotron Nano V2, a hybrid Mamba-Transformer LLM, and innovative token reduction techniques to achieve higher inference throughput in long document and video scenarios. We are releasing model checkpoints in BF16, FP8, and FP4 formats and sharing large parts of our datasets, recipes and training code.

cs.LG↗

arXiv2025

Check Yourself Before You Wreck Yourself: Selectively Quitting Improves LLM Agent Safety

As Large Language Model (LLM) agents increasingly operate in complex environments with real-world consequences, their safety becomes critical. While uncertainty quantification is well-studied for single-turn tasks, multi-turn agentic scenarios with real-world tool access present unique challenges where uncertainties and ambiguities compound, leading to severe or catastrophic risks beyond traditional text generation failures. We propose using "quitting" as a simple yet effective behavioral mechanism for LLM agents to recognize and withdraw from situations where they lack confidence. Leveraging the ToolEmu framework, we conduct a systematic evaluation of quitting behavior across 12 state-of-the-art LLMs. Our results demonstrate a highly favorable safety-helpfulness trade-off: agents prompted to quit with explicit instructions improve safety by an average of +0.39 on a 0-3 scale across all models (+0.64 for proprietary models), while maintaining a negligible average decrease of -0.03 in helpfulness. Our analysis demonstrates that simply adding explicit quit instructions proves to be a highly effective safety mechanism that can immediately be deployed in existing agent systems, and establishes quitting as an effective first-line defense mechanism for autonomous agents in high-stakes applications.

cs.CL↗

arXiv2025

Chemical Foundation Model Guided Design of High Ionic Conductivity Electrolyte Formulations

Designing optimal formulations is a major challenge in developing electrolytes for the next generation of rechargeable batteries due to the vast combinatorial design space and complex interplay between multiple constituents. Machine learning (ML) offers a powerful tool to uncover underlying chemical design rules and accelerate the process of formulation discovery. In this work, we present an approach to design new formulations that can achieve target performance, using a generalizable chemical foundation model. The chemical foundation model is fine-tuned on an experimental dataset of 13,666 ionic conductivity values curated from the lithium-ion battery literature. The fine-tuned model is used to discover 7 novel high conductivity electrolyte formulations through generative screening, improving the conductivity of LiFSI and LiDFOB based electrolytes by 82% and 172%, respectively. These findings highlight a generalizable workflow that is highly adaptable to the discovery of chemical mixtures with tailored properties to address challenges in energy storage and beyond.

cond-mat.mtrl-sci↗

arXiv2025

Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition

Recent open-world representation learning approaches have leveraged CLIP to enable zero-shot 3D object recognition. However, performance on real point clouds with occlusions still falls short due to unrealistic pretraining settings. Additionally, these methods incur high inference costs because they rely on Transformer's attention modules. In this paper, we make two contributions to address these limitations. First, we propose occlusion-aware text-image-point cloud pretraining to reduce the training-testing domain gap. From 52K synthetic 3D objects, our framework generates nearly 630K partial point clouds for pretraining, consistently improving real-world recognition performances of existing popular 3D networks. Second, to reduce computational requirements, we introduce DuoMamba, a two-stream linear state space model tailored for point clouds. By integrating two space-filling curves with 1D convolutions, DuoMamba effectively models spatial dependencies between point tokens, offering a powerful alternative to Transformer. When pretrained with our framework, DuoMamba surpasses current state-of-the-art methods while reducing latency and FLOPs, highlighting the potential of our approach for real-world applications. Our code and data are available at https://ndkhanh360.github.io/project-occtip.

cs.CV↗

arXiv2025

DocMIA: Document-Level Membership Inference Attacks against DocVQA Models

Document Visual Question Answering (DocVQA) has introduced a new paradigm for end-to-end document understanding, and quickly became one of the standard benchmarks for multimodal LLMs. Automating document processing workflows, driven by DocVQA models, presents significant potential for many business sectors. However, documents tend to contain highly sensitive information, raising concerns about privacy risks associated with training such DocVQA models. One significant privacy vulnerability, exploited by the membership inference attack, is the possibility for an adversary to determine if a particular record was part of the model's training data. In this paper, we introduce two novel membership inference attacks tailored specifically to DocVQA models. These attacks are designed for two different adversarial scenarios: a white-box setting, where the attacker has full access to the model architecture and parameters, and a black-box setting, where only the model's outputs are available. Notably, our attacks assume the adversary lacks access to auxiliary datasets, which is more realistic in practice but also more challenging. Our unsupervised methods outperform existing state-of-the-art membership inference attacks across a variety of DocVQA models and datasets, demonstrating their effectiveness and highlighting the privacy risks in this domain.

cs.LG↗

arXiv2025

Lusin approximation for functions of bounded variation

We prove a Lusin approximation of functions of bounded variation. If $f$ is a function of bounded variation on an open set $\Omega\subset X$, where $X=(X,d,\mu)$ is a given complete doubling metric measure space supporting a $1$-Poincar\'e inequality, then for every $\varepsilon>0$, there exist a function $f_\varepsilon$ on $\Omega$ and an open set $U_\varepsilon\subset\Omega$ such that the following properties hold true: \begin{enumerate} \item ${\rm Cap}_1(U_\varepsilon)<\varepsilon$; \item $\|f-f_\varepsilon\|_{\BV(\Omega)}< \varepsilon$; \item $f^\vee\equiv f_\varepsilon^\vee$ and $f^\wedge\equiv f_\varepsilon^\wedge$ on $\Omega\setminus U_\varepsilon$; \item $f_\varepsilon^\vee$ is upper semicontinuous on $\Omega$, and $f_\varepsilon^\wedge$ is lower semicontinuous on $\Omega$. \end{enumerate} If the space $X$ is unbounded, then such an approximating function $f_\varepsilon$ can be constructed with the additional property that the uniform limit at infinity of both $f^\vee_\varepsilon$ and $f^\wedge_\varepsilon$ is $0$. Moreover, when $X=\R^d$, we show that the non-centered maximal function of $f_\varepsilon$ is continuous in $\Omega$.

math.FA↗

arXiv2024

NeurIPS 2023 Competition: Privacy Preserving Federated Learning Document VQA

The Privacy Preserving Federated Learning Document VQA (PFL-DocVQA) competition challenged the community to develop provably private and communication-efficient solutions in a federated setting for a real-life use case: invoice processing. The competition introduced a dataset of real invoice documents, along with associated questions and answers requiring information extraction and reasoning over the document images. Thereby, it brings together researchers and expertise from the document analysis, privacy, and federated learning communities. Participants fine-tuned a pre-trained, state-of-the-art Document Visual Question Answering model provided by the organizers for this new domain, mimicking a typical federated invoice processing setup. The base model is a multi-modal generative language model, and sensitive information could be exposed through either the visual or textual input modality. Participants proposed elegant solutions to reduce communication costs while maintaining a minimum utility threshold in track 1 and to protect all information from each document provider using differential privacy in track 2. The competition served as a new testbed for developing and testing private federated learning methods, simultaneously raising awareness about privacy within the document image analysis and recognition community. Ultimately, the competition analysis provides best practices and recommendations for successfully running privacy-focused federated learning challenges in the future.

cs.LG↗

arXiv2024

Existence and uniqueness of limits at infinity for bounded variation functions

In this paper, we study the existence of limits at infinity along almost every infinite curve for the upper and lower approximate limits of bounded variation functions on complete unbounded metric measure spaces. We prove that if the measure is doubling and supports a $1$-Poincaré inequality, then for every bounded variation function $f$ and for $1$-a.e. infinite curve $γ$, for both the upper approximate limit $f^\vee$ and the lower approximate limit $f^\wedge$ we have that \[ \lim_{t\to+\infty}f^\vee(γ(t)) {\rm \ \ and\ \ }\lim_{t\to+\infty}f^\wedge(γ(t)) \] exist and are equal to the same finite value. We give examples showing that the conditions of doubling and a $1$-Poincaré inequality are also necessary for the existence of limits. Furthermore, we establish a characterization for strictly positive $1$-modulus of the family of all infinite curves in terms of bounded variation functions. These generalize results for Sobolev functions given in \cite{KN23}.

math.FA↗

arXiv2024

Language-Guided World Models: A Model-Based Approach to AI Control

This paper introduces the concept of Language-Guided World Models (LWMs) -- probabilistic models that can simulate environments by reading texts. Agents equipped with these models provide humans with more extensive and efficient control, allowing them to simultaneously alter agent behaviors in multiple tasks via natural verbal communication. In this work, we take initial steps in developing robust LWMs that can generalize to compositionally novel language descriptions. We design a challenging world modeling benchmark based on the game of MESSENGER (Hanjie et al., 2021), featuring evaluation settings that require varying degrees of compositional generalization. Our experiments reveal the lack of generalizability of the state-of-the-art Transformer model, as it offers marginal improvements in simulation quality over a no-text baseline. We devise a more robust model by fusing the Transformer with the EMMA attention mechanism (Hanjie et al., 2021). Our model substantially outperforms the Transformer and approaches the performance of a model with an oracle semantic parsing and grounding capability. To demonstrate the practicality of this model in improving AI safety and transparency, we simulate a scenario in which the model enables an agent to present plans to a human before execution, and to revise plans based on their language feedback.

cs.CL↗

arXiv2024

Improving Electrolyte Performance for Target Cathode Loading Using Interpretable Data-Driven Approach

Higher loading of active electrode materials is desired in batteries, especially those based on conversion reactions, for enhanced energy density and cost efficiency. However, increasing active material loading in electrodes can cause significant performance depreciation due to internal resistance, shuttling, and parasitic side reactions, which can be alleviated to a certain extent by a compatible design of electrolytes. In this work, a data-driven approach is leveraged to find a high-performing electrolyte formulation for a novel interhalogen battery custom to the target cathode loading. An electrolyte design consisting of 4 solvents and 4 salts is experimentally devised for a novel interhalogen battery based on a multi-electron redox reaction. The experimental dataset with variable electrolyte compositions and active cathode loading, is used to train a graph-based deep learning model mapping changing variables in the battery's material design to its specific capacity. The trained model is used to further optimize the electrolyte formulation compositions for enhancing the battery capacity at a target cathode loading by a two-fold approach: large-scale screening and interpreting electrolyte design principles for different cathode loadings. The data-driven approach is demonstrated to bring about an additional 20% increment in the specific capacity of the battery over capacities obtained from the experimental optimization.

cs.LG↗