arXiv subjects

Aobo Chen

Publications and source records attributed to Aobo Chen.

13 recordsLinked to original sources

arXiv2026

Walk dimension and vanishing curve modulus in metric measure spaces

On a regular local $p$-Dirichlet space supporting a Poincar\'{e} inequality and a capacity upper bound, we characterize the existence of curve families of positive $p$-modulus in terms of the small-scale behaviour of the scaling function. As an application, we show that, on an Ahlfors-regular metric measure space carrying a strongly local regular Dirichlet form with sub-Gaussian heat kernel estimates, the comparison between the walk dimension and $2$ determines whether the given metric attains its Ahlfors-regular conformal dimension.

math.MG↗

arXiv2026

Towards Unveiling Vulnerabilities of Large Reasoning Models in Machine Unlearning

Large language models (LLMs) possess strong semantic understanding, driving significant progress in data mining applications. This is further enhanced by large reasoning models (LRMs), which provide explicit multi-step reasoning traces. On the other hand, the growing need for the right to be forgotten has driven the development of machine unlearning techniques, which aim to eliminate the influence of specific data from trained models without full retraining. However, unlearning may also introduce new security vulnerabilities by exposing additional interaction surfaces. Although many studies have investigated unlearning attacks, there is no prior work on LRMs. To bridge the gap, we first in this paper propose LRM unlearning attack that forces incorrect final answers while generating convincing but misleading reasoning traces. This objective is challenging due to non-differentiable logical constraints, weak optimization effect over long rationales, and discrete forget set selection. To overcome these challenges, we introduce a bi-level exact unlearning attack that incorporates a differentiable objective function, influential token alignment, and a relaxed indicator strategy. To demonstrate the effectiveness and generalizability of our attack, we also design novel optimization frameworks and conduct comprehensive experiments in both white-box and black-box settings, aiming to raise awareness of the emerging threats to LRM unlearning pipelines.

cs.LG↗

arXiv2025

Elliptic Harnack inequalities for mixed local and nonlocal $p$-energy form on metric measure spaces

In the context of metric measure spaces, we introduce an axiomatic formulation of mixed local and nonlocal $p$-energy forms. Within this framework, we use the Poincar\'{e} inequality, the cutoff Sobolev inequality, and mild assumptions on the jump measure to establish the weak and strong elliptic Harnack inequalities for such mixed forms. Our approach is based on the De Giorgi--Nash--Moser method and extends the corresponding results for Dirichlet forms without the killing part, as well as for mixed energy forms on Euclidean spaces. Some consequences of elliptic Harnack inequalities are discussed.

math.AP↗

arXiv2025

Spectral bounds and heat kernel upper estimates for Dirichlet forms

We use a Harnack-type inequality on exit times and spectral bounds to characterize upper bounds of the heat kernel associated with any regular Dirichlet form without killing part, where the scale function may vary with position. We further show that this Harnack-type inequality is preserved under quasi-symmetric changes of metric on uniformly perfect metric spaces. This generalized the work of Mariano and Wang [Stochastic Process. Appl. 189 (2025) 104707)].

math.PR↗

arXiv2025

Membership Inference Attacks with False Discovery Rate Control

Recent studies have shown that deep learning models are vulnerable to membership inference attacks (MIAs), which aim to infer whether a data record was used to train a target model or not. To analyze and study these vulnerabilities, various MIA methods have been proposed. Despite the significance and popularity of MIAs, existing works on MIAs are limited in providing guarantees on the false discovery rate (FDR), which refers to the expected proportion of false discoveries among the identified positive discoveries. However, it is very challenging to ensure the false discovery rate guarantees, because the underlying distribution is usually unknown, and the estimated non-member probabilities often exhibit interdependence. To tackle the above challenges, in this paper, we design a novel membership inference attack method, which can provide the guarantees on the false discovery rate. Additionally, we show that our method can also provide the marginal probability guarantee on labeling true non-member data as member data. Notably, our method can work as a wrapper that can be seamlessly integrated with existing MIA methods in a post-hoc manner, while also providing the FDR control. We perform the theoretical analysis for our method. Extensive experiments in various settings (e.g., the black-box setting and the lifelong learning setting) are also conducted to verify the desirable performance of our method.

stat.ML↗

arXiv2024

Stability of heat kernel bounds under pointed Gromov--Hausdorff convergence

We construct a conservative and strongly local regular symmetric Dirichlet form on the pointed Gromov--Hausdorff limit space and demonstrate the stability of heat kernel estimates under this convergence. Furthermore, we establish the Mosco convergence of the associated energy forms along a subsequence.

math.MG↗

arXiv2024

Besov-Lipschitz norm and $p$-energy measure on scale-irregular Vicsek sets

In this paper, we establish the existence of $p$-energy norms and the corresponding $p$-energy measures for scale-irregular Vicsek sets, which may lack self-similarity. We also investigate the characterizations of $p$-energy norms in terms of Besov-Lipschitz norms, with their weak monotonicity and the corresponding Bourgain-Brezis-Mironescu convergence.

math.FA↗

arXiv2024

Exploring Fairness in Educational Data Mining in the Context of the Right to be Forgotten

In education data mining (EDM) communities, machine learning has achieved remarkable success in discovering patterns and structures to tackle educational challenges. Notably, fairness and algorithmic bias have gained attention in learning analytics of EDM. With the increasing demand for the right to be forgotten, there is a growing need for machine learning models to forget sensitive data and its impact, particularly within the realm of EDM. The paradigm of selective forgetting, also known as machine unlearning, has been extensively studied to address this need by eliminating the influence of specific data from a pre-trained model without complete retraining. However, existing research assumes that interactive data removal operations are conducted in secure and reliable environments, neglecting potential malicious unlearning requests to undermine the fairness of machine learning systems. In this paper, we introduce a novel class of selective forgetting attacks designed to compromise the fairness of learning models while maintaining their predictive accuracy, thereby preventing the model owner from detecting the degradation in model performance. Additionally, we propose an innovative optimization framework for selective forgetting attacks, capable of generating malicious unlearning requests across various attack scenarios. We validate the effectiveness of our proposed selective forgetting attacks on fairness through extensive experiments using diverse EDM datasets.

cs.LG↗

arXiv2024

Boundary Harnack principle on uniform domains

We present a proof of scale-invariant boundary Harnack principle for uniform domains when the underlying space satisfies a scale-invariant elliptic Harnack inequality. Our approach does not assume the underlying space to be geodesic. Additionally, the existence of Green functions is also not assumed beforehand and is ensured by a recent result from M. T. Barlow, Z.-Q. Chen and M. Murugan.

math.PR↗

arXiv2022

Cloaked near-field probe for non-invasive near-field optical microscopy

Near-field scanning optical microscopy is a powerful technique for imaging below the diffraction limit, which has been extensively used in bio-medical imaging and nanophotonics. However, when the electromagnetic fields under measurement are strongly confined, they can be heavily perturbed by the presence of the near-field probe itself. Here, taking inspiration from scattering-cancellation invisibility cloaks, Huygens-Kerker scatterers, and cloaked sensors, we design and fabricate a cloaked near-field probe. We show that, by suitably nanostructuring the probe, its electric and magnetic polarizabilities can be controlled and balanced. As a result, probe-induced perturbations can be largely suppressed, effectively cloaking the near-field probe without preventing its ability to measure. We experimentally demonstrate the cloaking effect by comparing the interaction of conventional and nanostructured probes with a representative nanophotonic structure, namely, a 1D photonic-crystal cavity. Our results show that, by engineering the structure of the probe, one can systematically control its back-action on the resonant fields of the sample and decrease the perturbation by >70% with most of our modified probes, and by up to one order of magnitude for the best probe, at probe-sample distances of 100 nm. Our work paves the way for non-invasive near-field optical microscopy of classical and quantum nano-systems.

physics.optics↗

arXiv2021

Spectral Causality and the Scattering of Waves

Causality - the principle stating that the output of a system cannot temporally precede the input - is a universal property of nature. Here, we show that analogous input-output relations can also be realized in the spectral domain by leveraging the peculiar properties of time-modulated non-Hermitian photonic systems. Specifically, we uncover the existence of a broad class of complex time-modulated metamaterials which obey the time-domain equivalent of the well-established frequency-domain Kramers-Kronig relations (a direct consequence of causality). We find that, in the scattering response of such time-modulated systems, the output frequencies are inherently prohibited from spectrally preceding the input frequencies, hence we refer to these systems as 'spectrally causal'. We explore the consequences of this newly introduced concept for several relevant applications, including broadband perfect absorption, temporal cloaking of an 'event', and truly unidirectional propagation along a synthetic dimension. By emulating the concept of causality in the spectral domain and providing new tools to extend the field of temporally modulated metamaterials ("chrono-metamaterials") into the complex realm, our findings may open unexplored opportunities and enable relevant technological advances in various areas of photonics and, more broadly, of wave physics and engineering.

physics.optics↗

arXiv2021

Physical limitations on broadband invisibility based on fast-light media

This note is a comment on a recent article [Tsakmakidis, et al., Nat Commun 10 (2019)] that presents a thought-provoking proposal to overcome the bandwidth restrictions of invisibility cloaks based on using media that support superluminal (faster than light in free space) group and phase velocities. As illustrated in Fig. 1 of the original article, a wave packet propagating through such a fast-light cloak is alleged to be able to reach the side behind the cloaked object simultaneously with a corresponding wave packet propagating through the shorter, direct route in free space without the object, so that "no shadow or waveform distortion arises." As the authors claim, the "extra pathlength is balanced out by the correspondingly larger group velocity of the pulse in the cloak", which allows to "restore the incident field distribution all around the object in, both, amplitude and phase". This fast-light effect may be achieved in a broadband fashion using active (gain) materials. The authors claim that such a fast-light cloak can hide an object, even from time-of-flight detection techniques, and achieve invisibility "over any desired frequency band." We disagree with these claims and believe that a thorough clarification of the ideas put forward in the original article is important and necessary for the broad wave-physics community. Specifically, in this comment we clarify that invisibility cloaks based on fast-light media suffer from fundamental bandwidth restrictions that arise due to causality, the nature of superluminal wave propagation, and the stability issues of active systems. These limitations and issues were not addressed in [Tsakmakidis, et al., Nat Commun 10 (2019)]. Most importantly, we show that the material model considered in the original article is unphysical.

physics.optics↗

arXiv2021

Dielectric nonlocal metasurfaces for fully solid-state ultra-thin optical systems

Most optical systems involve a combination of lenses separated by free-space regions where light acquires the required angle-dependent phase delay for a certain functionality. Very recently, flat-optics structures have been proposed to compress these large free-space volumes and miniaturize the overall optical system. However, these early designs can only replace free-space volumes of limited length, or operate in a very narrow angular range, or require a high-index background. These issues raise questions about the applicability of these devices in practical scenarios. Here, we first derive a fundamental trade-off between the length of compressed free space and the operating angular range, which explains some of the limitations of earlier designs, and we then propose a solution to relax this trade-off using nonlocal metasurface structures composed of suitably coupled resonant layers. This strategy, inspired by coupled-resonator-based band-pass filters, allows replacing free-space volumes of arbitrary length over wide angular ranges, and with very high transmittance. Finally, we theoretically demonstrate, for the first time, the potential of combining local and nonlocal metasurfaces to realize compact, fully solid-state, planar structures for focusing, imaging, and magnification, in which the focal length of the lens (and hence its magnifying power) does not dictate the actual distance at which focusing is achieved. Our findings are expected to extend the reach of the field of metasurfaces and open new unexplored opportunities.

physics.optics↗