arXiv subjects

Yuxuan Wan

Publications and source records attributed to Yuxuan Wan.

48 records · Page 3Linked to original sources

arXiv2023

Probabilistic Categorical Adversarial Attack & Adversarial Training

The existence of adversarial examples brings huge concern for people to apply Deep Neural Networks (DNNs) in safety-critical tasks. However, how to generate adversarial examples with categorical data is an important problem but lack of extensive exploration. Previously established methods leverage greedy search method, which can be very time-consuming to conduct successful attack. This also limits the development of adversarial training and potential defenses for categorical data. To tackle this problem, we propose Probabilistic Categorical Adversarial Attack (PCAA), which transfers the discrete optimization problem to a continuous problem that can be solved efficiently by Projected Gradient Descent. In our paper, we theoretically analyze its optimality and time complexity to demonstrate its significant advantage over current greedy based attacks. Moreover, based on our attack, we propose an efficient adversarial training framework. Through a comprehensive empirical study, we justify the effectiveness of our proposed attack and defense algorithms.

cs.LG↗

arXiv2023

BiasAsker: Measuring the Bias in Conversational AI System

Powered by advanced Artificial Intelligence (AI) techniques, conversational AI systems, such as ChatGPT and digital assistants like Siri, have been widely deployed in daily life. However, such systems may still produce content containing biases and stereotypes, causing potential social problems. Due to the data-driven, black-box nature of modern AI techniques, comprehensively identifying and measuring biases in conversational systems remains a challenging task. Particularly, it is hard to generate inputs that can comprehensively trigger potential bias due to the lack of data containing both social groups as well as biased properties. In addition, modern conversational systems can produce diverse responses (e.g., chatting and explanation), which makes existing bias detection methods simply based on the sentiment and the toxicity hardly being adopted. In this paper, we propose BiasAsker, an automated framework to identify and measure social bias in conversational AI systems. To obtain social groups and biased properties, we construct a comprehensive social bias dataset, containing a total of 841 groups and 8,110 biased properties. Given the dataset, BiasAsker automatically generates questions and adopts a novel method based on existence measurement to identify two types of biases (i.e., absolute bias and related bias) in conversational systems. Extensive experiments on 8 commercial systems and 2 famous research models, such as ChatGPT and GPT-3, show that 32.83% of the questions generated by BiasAsker can trigger biased behaviors in these widely deployed conversational systems. All the code, data, and experimental results have been released to facilitate future research.

cs.CL↗

arXiv2023

ChatGPT or Grammarly? Evaluating ChatGPT on Grammatical Error Correction Benchmark

ChatGPT is a cutting-edge artificial intelligence language model developed by OpenAI, which has attracted a lot of attention due to its surprisingly strong ability in answering follow-up questions. In this report, we aim to evaluate ChatGPT on the Grammatical Error Correction(GEC) task, and compare it with commercial GEC product (e.g., Grammarly) and state-of-the-art models (e.g., GECToR). By testing on the CoNLL2014 benchmark dataset, we find that ChatGPT performs not as well as those baselines in terms of the automatic evaluation metrics (e.g., $F_{0.5}$ score), particularly on long sentences. We inspect the outputs and find that ChatGPT goes beyond one-by-one corrections. Specifically, it prefers to change the surface expression of certain phrases or sentence structure while maintaining grammatical correctness. Human evaluation quantitatively confirms this and suggests that ChatGPT produces less under-correction or mis-correction issues but more over-corrections. These results demonstrate that ChatGPT is severely under-estimated by the automatic evaluation metrics and could be a promising tool for GEC.

cs.CL↗

arXiv2023

Origin of magnetically dead layers in spinel ferrites $M\text{Fe}_2\text{O}_4$ grown on $\text{Al}_2\text{O}_3$: Effects of post-deposition annealing studied by XMCD

We study the electronic and magnetic states of as-grown and annealed $M\text{Fe}_2\text{O}_4$(111)/$\text{Al}_2\text{O}_3$(111) ($M=\text{Co, Ni}$) thin films with various thicknesses grown on Si(111) substrates with the $γ$-$\text{Al}_2\text{O}_3$(111) buffer layers by using x-ray absorption spectroscopy (XAS) and x-ray magnetic circular dichroism (XMCD), to investigate magnetically dead layers in these films. Although the magnetically dead layers in the as-grown samples are formed near the interface with the $\text{Al}_2\text{O}_3$ buffer layer, we reveal that ferrimagnetic order is partially recovered by post-deposition annealing at 973 K for 48 hours in air. By analyzing the line shapes of the XAS and XMCD spectra, we conclude that, in the dead layers, there are a significant number of vacancies at the $T_d$ sites of the spinel structure, which may be the microscopic origin of the degraded ferrimagnetic order in the $M\text{Fe}_2\text{O}_4$(111) thin films.

cond-mat.mtrl-sci↗

arXiv2022

Towards Fair Classification against Poisoning Attacks

Fair classification aims to stress the classification models to achieve the equality (treatment or prediction quality) among different sensitive groups. However, fair classification can be under the risk of poisoning attacks that deliberately insert malicious training samples to manipulate the trained classifiers' performance. In this work, we study the poisoning scenario where the attacker can insert a small fraction of samples into training data, with arbitrary sensitive attributes as well as other predictive features. We demonstrate that the fairly trained classifiers can be greatly vulnerable to such poisoning attacks, with much worse accuracy & fairness trade-off, even when we apply some of the most effective defenses (originally proposed to defend traditional classification tasks). As countermeasures to defend fair classification tasks, we propose a general and theoretically guaranteed framework which accommodates traditional defense methods to fair classification against poisoning attacks. Through extensive experiments, the results validate that the proposed defense framework obtains better robustness in terms of accuracy and fairness than representative baseline methods.

cs.LG↗

arXiv2022

Transferable Unlearnable Examples

With more people publishing their personal data online, unauthorized data usage has become a serious concern. The unlearnable strategies have been introduced to prevent third parties from training on the data without permission. They add perturbations to the users' data before publishing, which aims to make the models trained on the perturbed published dataset invalidated. These perturbations have been generated for a specific training setting and a target dataset. However, their unlearnable effects significantly decrease when used in other training settings and datasets. To tackle this issue, we propose a novel unlearnable strategy based on Classwise Separability Discriminant (CSD), which aims to better transfer the unlearnable effects to other training settings and datasets by enhancing the linear separability. Extensive experiments demonstrate the transferability of the proposed unlearnable examples across training settings and datasets.

cs.LG↗

arXiv2022

Defense Against Gradient Leakage Attacks via Learning to Obscure Data

Federated learning is considered as an effective privacy-preserving learning mechanism that separates the client's data and model training process. However, federated learning is still under the risk of privacy leakage because of the existence of attackers who deliberately conduct gradient leakage attacks to reconstruct the client data. Recently, popular strategies such as gradient perturbation methods and input encryption methods have been proposed to defend against gradient leakage attacks. Nevertheless, these defenses can either greatly sacrifice the model performance, or be evaded by more advanced attacks. In this paper, we propose a new defense method to protect the privacy of clients' data by learning to obscure data. Our defense method can generate synthetic samples that are totally distinct from the original samples, but they can also maximally preserve their predictive features and guarantee the model performance. Furthermore, our defense strategy makes the gradient leakage attack and its variants extremely difficult to reconstruct the client data. Through extensive experiments, we show that our proposed defense method obtains better privacy protection while preserving high accuracy compared with state-of-the-art methods.

cs.LG↗

arXiv2021

Selective observation of surface and bulk bands in polar WTe2 by laser-based spin- and angle-resolved photoemission spectroscopy

The electronic state of WTe2, a candidate of type-II Weyl semimetal, is investigated by using laser-based spin- and angle-resolved photoemission spectroscopy (SARPES). We prepare the pair of WTe2 samples, one with (001) surface and the other with (00-1) surface, by "sandwich method", and measure the band structures of each surface separately. The Fermi arcs are observed on both surfaces. We identify that the Fermi arcs on the two surfaces are both originating from surface states. We further find a surface resonance band, which connects with the Fermi-arc band, forming a Dirac-cone-like band dispersion. Our results indicate that the bulk electron and hole bands are much closer in momentum space than band calculations.

cond-mat.mtrl-sci↗

arXiv2021

Reduced magnetocrystalline anisotropy of CoFe$_2$O$_4$ thin films studied by angle-dependent x-ray magnetic circular dichroism

Spinel-type CoFe$_2$O$_4$ is a ferrimagnetic insulator with the Néel temperature exceeding 790 K, and shows a strong cubic magnetocrystalline anisotropy (MCA) in bulk materials. However, when a CoFe$_2$O$_4$ film is grown on other materials, its magnetic properties are degraded so that so-called magnetically dead layers are expected to be formed in the interfacial region. We investigate how the magnetic anisotropy of CoFe$_2$O$_4$ is modified at the interface of CoFe$_2$O$_4$/Al$_2$O$_3$ bilayers grown on Si(111) using x-ray magnetic circular dichroism (XMCD). We find that the thinner CoFe$_2$O$_4$ films have significantly smaller MCA values than bulk materials. The reduction of MCA is explained by the reduced number of Co$^{2+}$ ions at the $O_h$ site reported by a previous study [Y. K. Wakabayashi $\textit{et al.}$, Phys. Rev. B $\textbf{96}$, 104410 (2017)].

cond-mat.mtrl-sci↗

arXiv2021

Large Orbital Magnetic Moment and Strong Perpendicular Magnetic Anisotropy in Heavily Intercalated Fe$_{x}$TiS$_2$

Titanium disulfide TiS$_2$, which is a member of the layered transition-metal dichalcogenides with the 1T-CdI$_2$-type crystal structure, is known to exhibit a wide variety of magnetism through intercalating various kinds of transition-metal atoms of different concentrations. Among them, Fe-intercalated titanium disulfide Fe$_x$TiS$_2$ is known to be ferromagnetic with strong perpendicular magnetic anisotropy (PMA) and large coercive fields ($H_\text{c}$). In order to study the microscopic origin of the magnetism of this compound, we have performed X-ray absorption spectroscopy (XAS) and X-ray magnetic circular dichroism (XMCD) measurements on single crystals of heavily intercalated Fe$_x$TiS$_2$ ($x\sim0.5$). The grown single crystals showed a strong PMA with a large $H_\text{c}$ of $μ_0H_\text{c} \simeq 1.0\ \text{T}$. XAS and XMCD spectra showed that Fe is fully in the valence states of 2+ and that Ti is in an itinerant electronic state, indicating electron transfer from the intercalated Fe atoms to the host TiS$_2$ bands. The Fe$^{2+}$ ions were shown to have a large orbital magnetic moment of $\simeq 0.59\ μ_\text{B}\text{/Fe}$, to which, combined with the spin-orbit interaction and the trigonal crystal field, we attribute the strong magnetic anisotropy of Fe$_x$TiS$_2$.

cond-mat.mtrl-sci↗

arXiv2020

Anisotropic spin distribution and perpendicular magnetic anisotropy in the layered ferromagnetic semiconductor (Ba,K)(Zn,Mn)$_{2}$As$_{2}$

Perpendicular magnetic anisotropy of the new ferromagnetic semiconductor (Ba,K)(Zn,Mn)$_{2}$As$_{2}$ is studied by angle-dependent x-ray magnetic circular dichroism measurements. The large magnetic anisotropy with the anisotropy field of 0.85 T is deduced by fitting the Stoner-Wohlfarth model to the magnetic-field-angle dependence of the projected magnetic moment. Transverse XMCD spectra highlights the anisotropic distribution of Mn 3$d$ electrons, where the $d_{xz}$ and $d_{yz}$ orbitals are less populated than the $d_{xy}$ state because of the $D_{2d}$ splitting arising from the elongated MnAs$_{4}$ tetrahedra. It is suggested that the magnetic anisotropy originates from the degeneracy lifting of $p$-$d_{xz}$, $d_{yz}$ hybridized states at the Fermi level and resulting energy gain due to spin-orbit coupling when spins are aligned along the $z$ direction.

cond-mat.str-el↗

arXiv2019

Magnetization process of the insulating ferromagnetic semiconductor (Al,Fe)Sb

We have studied the magnetization process of the new insulating ferromagnetic semiconductor (Al,Fe)Sb by means of x-ray magnetic circular dichroism. For an optimally doped sample with 10% Fe, a magnetization was found to rapidly increase at low magnetic fields and to saturate at high magnetic fields at room temperature, well above the Curie temperature of 40 K. We attribute this behavior to the existence of nanoscale Fe-rich ferromagnetic domains acting as superparamagnets. By fitting the magnetization curves using the Langevin function representing superparamagnetism plus the paramagnetic linear function, we estimated the average magnetic moment of the nanoscale ferromagnetic domain to be 300-400 $μ_{B}$, and the fraction of Fe atoms participating in the nano-scale ferromagnetism to be $\sim$50%. Such behavior was also reported for (In,Fe)As:Be and Ge:Fe, and seems to be a universal characteristic of the Fe-doped ferromagnetic semiconductors. Further Fe doping up to 14% led to the weakening of the ferromagnetism probably because antiferromagnetic superexchange interaction between nearest-neighbor Fe-Fe pairs becomes dominant.

cond-mat.str-el↗