arXiv subjects

Weidong Zheng

Publications and source records attributed to Weidong Zheng.

9 recordsLinked to original sources

arXiv2026

Quantum Imaging via Kurtosis-Difference Weighted Covariance on 2D Camera

Camera-based quantum imaging detects spatially correlated photon pairs from spontaneous parametric down-conversion (SPDC). Conventional covariance methods typically require tens of thousands of frames to extract weak correlations from noise. While thick crystals can increase photon flux, they generate photon pairs from multiple emission positions within the crystal, producing multiple correlation centers with complex pairing geometries. In addition, conventional covariance methods assume a single pre-selected correlation center and cannot fully exploit these distributed correlations. We demonstrate that kurtosis difference, a fourth-order statistic measuring tail similarity, effectively discriminates correlated pixel pairs even when correlation coefficients remain low. Weighting covariance by an exponential function of absolute kurtosis difference can select symmetric pixels while preserving true coincidences. This kurtosis weighting automatically identifies correlated pairs within a broad search region and accommodates multiple pairing geometries without requiring precise correlation center calibration. At 5000 frames, our method yields a contrast-to-noise ratio (CNR) exceeding 7, whereas standard covariance remains below 2. Compared with standard covariance, the method reduces the acquisition time by 40-fold and could enable practical quantum imaging in sparse correlated-photon regimes.

quant-ph↗

arXiv2026

Classification-Head Bias in Class-Level Machine Unlearning: Diagnosis, Mitigation, and Evaluation

Class-level machine unlearning aims to remove the influence of specified classes while preserving model utility on retained classes. Existing methods are commonly evaluated by retain-set accuracy, forget-set accuracy, and unlearning time, but these metrics provide limited insight into how forgetting is achieved internally. In this paper, we reveal a bias-dominated shortcut in class-level unlearning: the prediction of forgotten classes can be suppressed by decreasing the corresponding bias terms in the final classification head. We first analyze the gradient dynamics of classification-head biases under softmax cross-entropy training, explaining why retain-set-only optimization tends to reduce the biases of absent classes. Based on this observation, we introduce BiasShift as a diagnostic baseline, showing that simple bias manipulation can satisfy conventional unlearning metrics while leaving abnormal bias patterns that reveal forgotten labels. To mitigate excessive forgotten-class bias suppression, we propose two bias-aware mechanisms, namely Two-Stage Bias Gradient Reversal Mechanism (TS-BGRM) and Lower-Bound Hinge Regularization (LB-HR). We further introduce three bias-oriented metrics, including Bias Stability Coefficient (BSC), Median Bias Gap (MBG), and Minimal Bias Score (MBS), to quantify bias dependence and potential leakage. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet demonstrate that the proposed methods maintain competitive unlearning performance while producing more stable bias distributions. We have released our code at {https://github.com/zwd2024/Beyond-the-Shadow-of-Bias-From-Classification-Head-Bias-to-Parameter-Redistribution}.

cs.LG↗

arXiv2026

Label Leakage Attacks in Machine Unlearning: A Parameter and Inversion-Based Approach

With the widespread application of artificial intelligence technologies in face recognition and other fields, data privacy security issues have received extensive attention, especially the \textit{right to be forgotten} emphasized by numerous privacy protection laws. Existing technologies have proposed various unlearning methods, but they may inadvertently leak the categories of unlearned data. This paper focuses on the category unlearning scenario, analyzes the potential problems of category leakage of unlearned data in multiple scenarios, and proposes four attack methods from the perspectives of model parameters and model inversion based on attackers with different knowledge backgrounds. At the level of model parameters, we construct discriminative features by computing either dot products or vector differences between the parameters of the target model and those of auxiliary models trained on subsets of retained data and unrelated data, respectively. These features are then processed via k-means clustering, Youden's Index, and decision tree algorithms to achieve accurate identification of the forgotten class. In the model inversion domain, we design a gradient optimization-based white-box attack and a genetic algorithm-based black-box attack to reconstruct class-prototypical samples. The prediction profiles of these synthesized samples are subsequently analyzed using a threshold criterion and an information entropy criterion to infer the forgotten class. We evaluate the proposed attacks on four standard datasets against five state-of-the-art unlearning algorithms, providing a detailed analysis of the strengths and limitations of each method. Experimental results demonstrate that our approach can effectively infer the classes forgotten by the target model.

cs.CR↗

arXiv2025

Departures from Standard Disk Predictions in Intensive Ground-Based Monitoring of Three AGN

We present ground-based, multi-band light curves of the AGN Mrk~509, NGC\,4151, and NGC\,4593 obtained contemporaneously with \sw\, monitoring. We measure cross-correlation lags relative to \sw\, UVW2 (1928~Å) and test the standard prediction for disk reprocessing, which assumes a geometrically thin, optically thick accretion disk where continuum interband delays follow the relation $ τ(λ) \propto λ^{4/3} $. For Mrk~509 the 273-d \sw\, campaign gives well-defined lags that increase with wavelength as $τ(λ)\proptoλ^{2.17\pm0.2}$, steeper than the thin-disk prediction, and the optical lags are a factor of $\sim5$ longer than expected for a simple disk-reprocessing model. This ``disk-size discrepancy'' as well as excess lags in the $u$ and $r$ bands (which include the Balmer continuum and H$α$, respectively) suggest a mix of short lags from the disk and longer lags from nebular continuum originating in the broad-line region. The shorter \sw\, campaigns, 69~d on NGC\,4151 and 22~d on NGC\,4593, yield less well-defined, shorter lags $<2$~d. The NGC\,4593 lags are consistent with $τ(λ) \propto λ^{4/3}$ but with uncertainties too large for a strong test. For NGC\,4151 the \sw\, lags match $τ(λ) \propto λ^{4/3}$, with a small $U$-band excess, but the ground-based lags in the $r$, $i$, and $z$ bands are significantly shorter than the $B$ and $g$ lags, and also shorter than expected from the thin-disk prediction. The interpretation of this unusual lag spectrum is unclear. Overall these results indicate significant diversity in the $τ-λ$ relation across the optical/UV/NIR, which differs from the more homogeneous behavior seen in the \sw\, bands.

astro-ph.GA↗

arXiv2025

Extending the Low-Frequency Limit of Time-Domain Thermoreflectance via Periodic Waveform Analysis

Time-domain thermoreflectance (TDTR) is a powerful technique for characterizing the thermal properties of layered materials. However, its effectiveness at modulation frequencies below 0.1 MHz is hindered by pulse accumulation effects, limiting its ability to accurately measure in-plane thermal conductivities below 6 W/(m K). Here, we present a periodic waveform analysis-based TDTR (PWA-TDTR) method that extends the measurable frequency range down to 50 Hz with minimal modifications to the conventional setup. This advancement greatly enhances measurement sensitivity, enabling accurate measurements of in-plane thermal conductivities as low as 0.2 W/(m K). We validate the technique by measuring polymethyl methacrylate (PMMA) and fused silica, using PWA-TDTR to obtain in-plane thermal diffusivity and conventional TDTR to measure cross-plane thermal effusivity. Together, these allow the extraction of both thermal conductivity and volumetric heat capacity, with results in excellent agreement with literature values. We further demonstrate the versatility of PWA-TDTR through (1) thermal conductivity and heat capacity measurements of thin liquid films and (2) depth-resolved thermal conductivity profiling in lithium niobate crystals, revealing point defect-induced inhomogeneities at depths up to 100 um. By overcoming frequency and sensitivity constraints, PWA-TDTR significantly expands the applicability of TDTR, enabling detailed investigations of thermal transport in materials and conditions that were previously challenging to study.

physics.app-ph↗

arXiv2022

Non-equilibrium Phonon Thermal Resistance at MoS2/Oxide and Graphene/Oxide Interfaces

Accurate measurements and physical understanding of thermal boundary resistance (R) of two-dimensional (2D) materials are imperative for effective thermal management of 2D electronics and photonics. In previous studies, heat dissipation from 2D material devices was presumed to be dominated by phonon transport across the interfaces. In this study, we find that in addition to phonon transport, thermal resistance between non-equilibrium phonons in the 2D materials could play a critical role too when the 2D material devices are internally self-heated, either optically or electrically. We accurately measure R of oxide/MoS2/oxide and oxide/graphene/oxide interfaces for three oxides (SiO2, HfO2, Al2O3) by differential time-domain thermoreflectance (TDTR). Our measurements of R across these interfaces with external heating are 2-to-4 times lower than previously reported R of the similar interfaces measured by Raman thermometry with internal self-heating. Using a simple model, we show that the observed discrepancy can be explained by an additional internal thermal resistance (Rint) between non-equilibrium phonons present during Raman measurements. We subsequently estimate that for MoS2 and graphene, Rint is about 31 and 22 m2 K/GW, respectively. The values are comparable to the thermal resistance due to finite phonon transmission across interfaces of 2D materials and thus cannot be ignored in the design of 2D material devices. Moreover, the non-equilibrium phonons also lead to a different temperature dependence than that by phonon transport. As such, our work provides important insights into physical understanding of heat dissipation in 2D material devices.

physics.app-ph↗

arXiv2018

Anisotropic model with truncated linear dispersion for lattice and interfacial thermal transport in layered materials

Recently, an anisotropic Debye model [Dames et al., Physical Review B 87, 12 (2013)] was proposed for calculations of the interfacial thermal conductance and the minimum thermal conductivity of graphite-like layered materials. Despite successes of the model in explaining heat transport mechanisms in layered materials (e.g., phonon focusing in highly anisotropic materials), the anisotropic Debye model assumes a phonon dispersion with unrealistic speeds of sounds especially for the flexural (ZA) phonons and overestimated cutoffs for all phonon branches. The deficiencies lead to substantially underestimated phonon irradiation for low-frequency phonons. Here, we develop an anisotropic model with truncated linear dispersion that resembles the real phonon dispersion, using speeds of sounds derived from elastic constants and cutoff frequencies derived from Brillouin zone boundaries. We also employ a piecewise linear function for the ZA phonons. Our model correctly calculates the phonon irradiation over a wide temperature range, verifying the accuracy of our model.We compare calculations of our and the Dames models to measurements of thermal conductivity of graphite and thermal conductance of metal/graphite interfaces, and find that the two models differ significantly for heat transport across the basal planes in graphite even at high temperatures. Our work thus provides a convenient analytical tool to study the phonon transport properties in layered materials.

physics.app-ph↗

arXiv2018

Tuning electronic heat transport in graphene/metal heterostructures with ultralow thermal conductivity

Prior ultralow thermal conductivity materials are not suitable for thermoelectric applications due to the limited electronic transport in the materials. Here, we present a new class of ultralow thermal conductivity materials with substantial electronic heat transport. Our samples are graphene/metal heterostructures of transferred graphene and ultrathin metal films (Pd, Au and Ni) deposited by either thermal evaporation or rf magnetron sputtering. For the evaporated samples, we achieve an ultralow thermal conductivity of 0.06 W m-1 K-1, with phonons as the dominant heat carriers. The ultralow thermal conductivity is due to a huge disparity in phonon energy in graphene and metals. Interestingly, for the sputtered samples, we find that about 50 % of heat is carried by electrons, even when thermal conductivity is about 0.1 W m-1 K-1. We attribute the electronic contribution to transmission of electrons across atomic-scale pinholes in graphene. With the ultralow thermal conductivity and substantial electronic transport, the new materials could be explored for thermoelectric applications.

cond-mat.mtrl-sci↗

arXiv2018

Achieving huge thermal conductance of metallic nitride on graphene through enhanced elastic and inelastic phonon transmission

Low thermal conductance of metal contacts is one of the main challenges in thermal management of nanoscale devices of graphene and other 2D materials. Previous attempts to search for metal contacts with high thermal conductance yielded limited success due to incomplete understanding of the origins of the low thermal conductance. In this paper, we carefully study the intrinsic thermal conductance across metal/graphene/metal interfaces to identify the heat transport mechanisms across graphene interfaces. We find that unlike metal contacts on diamond, the intrinsic thermal conductance of most graphene interfaces (except Ti and TiNx) is only about 50 % of the phonon radiation limit, suggesting that heat is carried across graphene interfaces mainly through elastic transmission of phonons. We thus propose a convenient approach to substantially enhance the phononic heat transport across metal contacts on graphene, by better matching the energy of phonons in metals and graphene, e.g., using metallic nitrides. We test the idea with TiNx, with phonon frequencies of up to 1.18*10^14 rad/s, 47 % of the highest phonon frequencies in graphene of 2.51*10^14 rad/s . Interestingly, we obtain a huge thermal conductance of 270 MW m-2 K-1 for TiNx/graphene interfaces, which is about 140 % of the phonon radiation limit. The huge thermal conductance could be partially attributed to inelastic phonon transport across the TiNx/graphene interface. Our work provides guidance for the search for good metal contacts on 2D materials and devices.

physics.app-ph↗