arXiv subjects

Xinyang Lu

Publications and source records attributed to Xinyang Lu.

16 recordsLinked to original sources

arXiv2026

De-attribute to Forget for LLM Unlearning

The rapid development of large language models (LLMs) has raised concerns on the use of inappropriate data for training, which has led to a growing interest in LLM unlearning. Many existing LLM unlearning approaches rely on optimizing prediction loss(es), such as maximizing the loss on the forget set, but often face critical issues like over-forgetting and poor model utility. To address them, this paper novelly frames the optimization objective for LLM unlearning as one of zeroing out data attribution instead. In particular, we propose the first LLM unlearning framework based on data attribution rewards called DareU that performs reinforcement learning to update the LLM by reducing the attribution score of its generated responses (i.e., de-attributing) to the forget data owners. Empirical evaluation using an LLM classifier as an efficient approximation of attribution shows that DareU outperforms existing baselines by achieving effective unlearning while balancing forget quality and model utility well.

cs.LG↗

arXiv2025

WaterDrum: Watermarking for Data-centric Unlearning Metric

Large language model (LLM) unlearning is critical in real-world applications where it is necessary to efficiently remove the influence of private, copyrighted, or harmful data from some users. Existing utility-centric unlearning metrics (based on model utility) may fail to accurately evaluate the extent of unlearning in realistic settings such as when the forget and retain sets have semantically similar content and/or retraining the model from scratch on the retain set is impractical. This paper presents the first data-centric unlearning metric for LLMs called WaterDrum that exploits robust text watermarking to overcome these limitations. We introduce new benchmark datasets (with different levels of data similarity) for LLM unlearning that can be used to rigorously evaluate unlearning algorithms via WaterDrum. Our code is available at https://github.com/lululu008/WaterDrum and our new benchmark datasets are released at https://huggingface.co/datasets/Glow-AI/WaterDrum-Ax.

cs.LG↗

arXiv2025

Decorated phases in triblock copolymers: zeroth- and first-order analysis

We study a two-dimensional inhibitory ternary system characterized by a free energy functional which combines an interface short-range interaction energy promoting micro-domain growth with a Coulomb-type long-range interaction energy which prevents micro-domains from unlimited spreading. Here we consider a scenario in which two species are dominant and one species is vanishingly small. In this scenario two energy levels are distinguished: the zeroth-order energy encodes information on the optimal arrangement of the dominant constituents, while the first-order energy gives the shape of the vanishing constituent. This first-order energy also shows that, for any optimal configuration, the vanishing phase must lie on the boundary between the two dominant constituents and form lens clusters also known as vesica piscis.

math.AP↗

arXiv2024

Global-to-Local Support Spectrums for Language Model Explainability

Existing sample-based methods, like influence functions and representer points, measure the importance of a training point by approximating the effect of its removal from training. As such, they are skewed towards outliers and points that are very close to the decision boundaries. The explanations provided by these methods are often static and not specific enough for different test points. In this paper, we propose a method to generate an explanation in the form of support spectrums which are based on two main ideas: the support sets and a global-to-local importance measure. The support set is the set of training points, in the predicted class, that ``lie in between'' the test point and training points in the other classes. They indicate how well the test point can be distinguished from the points not in the predicted class. The global-to-local importance measure is obtained by decoupling existing methods into the global and local components which are then used to select the points in the support set. Using this method, we are able to generate explanations that are tailored to specific test points. In the experiments, we show the effectiveness of the method in image classification and text generation tasks.

cs.LG↗

arXiv2024

TRACE: TRansformer-based Attribution using Contrastive Embeddings in LLMs

The rapid evolution of large language models (LLMs) represents a substantial leap forward in natural language understanding and generation. However, alongside these advancements come significant challenges related to the accountability and transparency of LLM responses. Reliable source attribution is essential to adhering to stringent legal and regulatory standards, including those set forth by the General Data Protection Regulation. Despite the well-established methods in source attribution within the computer vision domain, the application of robust attribution frameworks to natural language processing remains underexplored. To bridge this gap, we propose a novel and versatile TRansformer-based Attribution framework using Contrastive Embeddings called TRACE that, in particular, exploits contrastive learning for source attribution. We perform an extensive empirical evaluation to demonstrate the performance and efficiency of TRACE in various settings and show that TRACE significantly improves the ability to attribute sources accurately, making it a valuable tool for enhancing the reliability and trustworthiness of LLMs.

cs.CL↗

arXiv2024

On Newton's Method to Unlearn Neural Networks

With the widespread applications of neural networks (NNs) trained on personal data, machine unlearning has become increasingly important for enabling individuals to exercise their personal data ownership, particularly the "right to be forgotten" from trained NNs. Since retraining is computationally expensive, we seek approximate unlearning algorithms for NNs that return identical models to the retrained oracle. While Newton's method has been successfully used to approximately unlearn linear models, we observe that adapting it for NN is challenging due to degenerate Hessians that make computing Newton's update impossible. Additionally, we show that when coupled with popular techniques to resolve the degeneracy, Newton's method often incurs offensively large norm updates and empirically degrades model performance post-unlearning. To address these challenges, we propose CureNewton's method, a principle approach that leverages cubic regularization to handle the Hessian degeneracy effectively. The added regularizer eliminates the need for manual finetuning and affords a natural interpretation within the unlearning context. Experiments across different models and datasets show that our method can achieve competitive unlearning performance to the state-of-the-art algorithm in practical unlearning settings, while being theoretically justified and efficient in running time.

cs.LG↗

arXiv2023

Source Attribution for Large Language Model-Generated Data

The impressive performances of Large Language Models (LLMs) and their immense potential for commercialization have given rise to serious concerns over the Intellectual Property (IP) of their training data. In particular, the synthetic texts generated by LLMs may infringe the IP of the data being used to train the LLMs. To this end, it is imperative to be able to perform source attribution by identifying the data provider who contributed to the generation of a synthetic text by an LLM. In this paper, we show that this problem can be tackled by watermarking, i.e., by enabling an LLM to generate synthetic texts with embedded watermarks that contain information about their source(s). We identify the key properties of such watermarking frameworks (e.g., source attribution accuracy, robustness against adversaries), and propose a source attribution framework that satisfies these key properties due to our algorithmic designs. Our framework enables an LLM to learn an accurate mapping from the generated texts to data providers, which sets the foundation for effective source attribution. Extensive empirical evaluations show that our framework achieves effective source attribution.

cs.LG↗

arXiv2023

On a Quaternary Non-Local Isoperimetric Problem

We study a two-dimensional quaternary inhibitory system. This free energy functional combines an interface energy favoring micro-domain growth with a Coulomb-type long range interaction energy which prevents micro-domains from unlimited spreading. Here we consider a limit in which three species are vanishingly small, but interactions are correspondingly large to maintain a nontrivial limit. In this limit two energy levels are distinguished: the highest order limit encodes information on the geometry of local structures as a three-component isoperimetric problem, while the second level describes the spatial distribution of components in global minimizers. Geometrical descriptions of limit configurations are derived.

math.AP↗

arXiv2023

Action and Trajectory Planning for Urban Autonomous Driving with Hierarchical Reinforcement Learning

Reinforcement Learning (RL) has made promising progress in planning and decision-making for Autonomous Vehicles (AVs) in simple driving scenarios. However, existing RL algorithms for AVs fail to learn critical driving skills in complex urban scenarios. First, urban driving scenarios require AVs to handle multiple driving tasks of which conventional RL algorithms are incapable. Second, the presence of other vehicles in urban scenarios results in a dynamically changing environment, which challenges RL algorithms to plan the action and trajectory of the AV. In this work, we propose an action and trajectory planner using Hierarchical Reinforcement Learning (atHRL) method, which models the agent behavior in a hierarchical model by using the perception of the lidar and birdeye view. The proposed atHRL method learns to make decisions about the agent's future trajectory and computes target waypoints under continuous settings based on a hierarchical DDPG algorithm. The waypoints planned by the atHRL model are then sent to a low-level controller to generate the steering and throttle commands required for the vehicle maneuver. We empirically verify the efficacy of atHRL through extensive experiments in complex urban driving scenarios that compose multiple tasks with the presence of other vehicles in the CARLA simulator. The experimental results suggest a significant performance improvement compared to the state-of-the-art RL methods.

cs.RO↗

arXiv2023

Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs

Empowering large language models to accurately express confidence in their answers is essential for trustworthy decision-making. Previous confidence elicitation methods, which primarily rely on white-box access to internal model information or model fine-tuning, have become less suitable for LLMs, especially closed-source commercial APIs. This leads to a growing need to explore the untapped area of black-box approaches for LLM uncertainty estimation. To better break down the problem, we define a systematic framework with three components: prompting strategies for eliciting verbalized confidence, sampling methods for generating multiple responses, and aggregation techniques for computing consistency. We then benchmark these methods on two key tasks-confidence calibration and failure prediction-across five types of datasets (e.g., commonsense and arithmetic reasoning) and five widely-used LLMs including GPT-4 and LLaMA 2 Chat. Our analysis uncovers several key insights: 1) LLMs, when verbalizing their confidence, tend to be overconfident, potentially imitating human patterns of expressing confidence. 2) As model capability scales up, both calibration and failure prediction performance improve. 3) Employing our proposed strategies, such as human-inspired prompts, consistency among multiple responses, and better aggregation strategies can help mitigate this overconfidence from various perspectives. 4) Comparisons with white-box methods indicate that while white-box methods perform better, the gap is narrow, e.g., 0.522 to 0.605 in AUROC. Despite these advancements, none of these techniques consistently outperform others, and all investigated methods struggle in challenging tasks, such as those requiring professional knowledge, indicating significant scope for improvement. We believe this study can serve as a strong baseline and provide insights for eliciting confidence in black-box LLMs.

cs.CL↗

arXiv2022

Core shells and double bubbles in a weighted nonlocal isoperimetric problem

We consider a sharp-interface model of $ABC$ triblock copolymers, for which the surface tension $\sigma_{ij}$ across the interface separating phase $i$ from phase $j$ may depend on the components. We study global minimizers of the associated ternary local isoperimetric problem in $\mathbb{R}^2$, and show how the geometry of minimizers changes with the surface tensions $\sigma_{ij}$, varying from symmetric double-bubbles for equal surface tensions, through asymmetric double bubbles, to core shells as the values of $\sigma_{ij}$ become more disparate. Then we consider the effect of nonlocal interactions in a droplet scaling regime, in which vanishingly small particles of two phases are distributed in a sea of the third phase. We are particularly interested in a degenerate case of $\sigma_{ij}$ in which minimizers exhibit core shell geometry, as this phase configuration is expected on physical grounds in nonlocal ternary systems.

math.AP↗

arXiv2021

Regularity of a gradient flow generated by the anisotropic Landau-de Gennes energy with a singular potential

In this paper we study a gradient flow generated by the Landau-de Gennes free energy that describes nematic liquid crystal configurations in the space of $Q$-tensors. This free energy density functional is composed of three quadratic terms as the elastic energy density part, and a singular potential in the bulk part that is considered as a natural enforcement of a physical constraint on the eigenvalues of $Q$. The system is a non-diagonal parabolic system with a singular potential which trends to infinity logarithmically when the eigenvalues of $Q$ approaches the physical boundary. We give a rigorous proof that for rather general initial data with possibly infinite free energy, the system has a unique strong solution after any positive time $t_0$. Furthermore, this unique strong solution detaches from the physical boundary after a sufficiently large time $T_0$. We also give estimate of the Hausdorff measure of the set where the solution touches the physical boundary and thus prove a partial regularity result of the solution in the intermediate stage $(0,T_0)$.

math.AP↗

arXiv2021

Blowup rate estimates of the Ball-Majumdar potential and its gradient in the Landau-de Gennes theory

In this paper we revisit a singular bulk potential in the Landau-de Gennes free energy that describes nematic liquid crystal configurations in the framework of the Q-tensor order parameter. This singular potential, called Ball-Majumdar potential, is introduced in [3], and is considered as a natural enforcement of a physical constraint on the eigenvalues of symmetric, traceless Q-tensors. Specifically, we establish blowup rates of both this singular potential and its gradient as Q approaches its physical boundary.

math.AP↗

arXiv2020

Average-distance problem with curvature penalization for data parameterization: regularity of minimizers

We propose a model for finding one-dimensional structure in a given measure. Our approach is based on minimizing an objective functional which combines the average-distance functional to measure the quality of the approximation and penalizes the curvature, similarly to the elastica functional. Introducing the curvature penalization overcomes some of the shortcomings of the average-distance functional, in particular the lack of regularity of minimizers. We establish existence, uniqueness and regularity of minimizers of the proposed functional. In particular we establish $C^{1,1}$ estimates on the minimizers.

math.AP↗

arXiv2019

Periodic Minimizers of a Ternary Non-Local Isoperimetric Problem

We study a two-dimensional ternary inhibitory system derived as a sharp-interface limit of the Nakazawa-Ohta density functional theory of triblock copolymers. This free energy functional combines an interface energy favoring micro-domain growth with a Coulomb-type long range interaction energy which prevents micro-domains from unlimited spreading. Here we consider a limit in which two species are vanishingly small, but interactions are correspondingly large to maintain a nontrivial limit. In this limit two energy levels are distinguished: the highest order limit encodes information on the geometry of local structures as a two-component isoperimetric problem, while the second level describes the spatial distribution of components in global minimizers. We provide a sharp rigorous derivation of the asymptotic limit, both for minimizers and in the context of Gamma-convergence. Geometrical descriptions of limit configurations are derived; among other results, we will show that, quite unexpectedly, coexistence of single and double bubbles can arise. The main difficulties are hidden in the optimal solution of two-component isoperimetric problem: compared to binary systems, not only it lacks an explicit formula, but, more crucially, it can be neither concave nor convex on parts of its domain.

math.AP↗

arXiv2019

DBSCAN: Optimal Rates For Density Based Clustering

We study the problem of optimal estimation of the density cluster tree under various assumptions on the underlying density. Building up from the seminal work of Chaudhuri et al. [2014], we formulate a new notion of clustering consistency which is better suited to smooth densities, and derive minimax rates of consistency for cluster tree estimation for Holder smooth densities of arbitrary degree α. We present a computationally efficient, rate optimal cluster tree estimator based on a straightforward extension of the popular density-based clustering algorithm DBSCAN by Ester et al. [1996]. The procedure relies on a kernel density estimator with an appropriate choice of the kernel and bandwidth to produce a sequence of nested random geometric graphs whose connected components form a hierarchy of clusters. The resulting optimal rates for cluster tree estimation depend on the degree of smoothness of the underlying density and, interestingly, match minimax rates for density estimation under the supremum norm. Our results complement and extend the analysis of the DBSCAN algorithm in Sriperumbudur and Steinwart [2012]. Finally, we consider level set estimation and cluster consistency for densities with jump discontinuities, where the sizes of the jumps and the distance among clusters are allowed to vanish as the sample size increases. We demonstrate that our DBSCAN-based algorithm remains minimax rate optimal in this setting as well.

math.ST↗