SearcharxivSearch

arXiv subjects

Jiquan Pei

Publications and source records attributed to Jiquan Pei.

5 recordsLinked to original sources

Trustworthy AI: From Principles to Practices

The rapid development of Artificial Intelligence (AI) technology has enabled the deployment of various systems based on it. However, many current AI systems are found vulnerable to imperceptible attacks, biased against underrepresented groups, lacking in user privacy protection. These shortcomings degrade user experience and erode people's trust in all AI systems. In this review, we provide AI practitioners with a comprehensive guide for building trustworthy AI systems. We first introduce the theoretical framework of important aspects of AI trustworthiness, including robustness, generalization, explainability, transparency, reproducibility, fairness, privacy preservation, and accountability. To unify currently available but fragmented approaches toward trustworthy AI, we organize them in a systematic approach that considers the entire lifecycle of AI systems, ranging from data acquisition to model development, to system development and deployment, finally to continuous monitoring and governance. In this framework, we offer concrete action items for practitioners and societal stakeholders (e.g., researchers, engineers, and regulators) to improve AI trustworthiness. Finally, we identify key opportunities and challenges for the future development of trustworthy AI systems, where we identify the need for a paradigm shift toward comprehensively trustworthy AI systems.

cs.AI

Which Style Makes Me Attractive? Interpretable Control Discovery and Counterfactual Explanation on StyleGAN

The semantically disentangled latent subspace in GAN provides rich interpretable controls in image generation. This paper includes two contributions on semantic latent subspace analysis in the scenario of face generation using StyleGAN2. First, we propose a novel approach to disentangle latent subspace semantics by exploiting existing face analysis models, e.g., face parsers and face landmark detectors. These models provide the flexibility to construct various criterions with very concrete and interpretable semantic meanings (e.g., change face shape or change skin color) to restrict latent subspace disentanglement. Rich latent space controls unknown previously can be discovered using the constructed criterions. Second, we propose a new perspective to explain the behavior of a CNN classifier by generating counterfactuals in the interpretable latent subspaces we discovered. This explanation helps reveal whether the classifier learns semantics as intended. Experiments on various disentanglement criterions demonstrate the effectiveness of our approach. We believe this approach contributes to both areas of image manipulation and counterfactual explainability of CNNs. The code is available at \url{https://github.com/prclibo/ice}.

cs.CV

On the high energy spin excitations in the high temperature superconductors: an RVB theory

The high energy spin excitation in the high T$_{c}$ cuprates is studied in the single mode approximation for the $t-t'-J$ model. An exact form for the mode dispersion is derived. When the Gutzwiller projected BCS state is used as the variational ground state, a spin-wave-like dispersion of about 2.2$J$ is uncovered along the $Γ=(0,0)$ to $\mathrm{M}=(π,0)$ line. Both the mode energy and the integrated intensity of the spin fluctuation spectrum are found to be almost doping independent in large doping range, which agrees very well with the observations of recent RIXS measurements. Together with previous studies on the quasiparticle properties of the Gutzwiiler projected BCS state, our results indicate that such a Fermionic RVB theory can provide a consistent description of both the itinerant and the local aspect of electronic excitations in the high T$_{c}$ cuprates.

cond-mat.str-el

Topological entanglement entropy in Gutzwiller projected spin liquids

The topological entanglement entropy(TEE) of Gutzwiller projected RVB state is studied with Monte Carlo simulation. New tricks are proposed to improve the convergence of TEE, which enable us to show that the spin liquid state studied in Ref.\cite{Vishwanath} actually does not support $Z_{2}$ topological order, a conclusion that is consistent with the information drawn from the inspection of the topological degeneracy on the same state. We find both a long ranged RVB amplitude and an approximate Marshall sign structure are at the origin of the suppression of vison gap in this spin liquid state. On the other hand, robust signature of $Z_{2}$ topological order, i.e., a TEE of $\ln2$, is clearly demonstrated for a Gutzwiiler projected RVB state on triangular lattice which is evolved from the RVB state proposed originally by Anderson\cite{Sorella}. We also find that it is the sign, rather than the amplitude of the RVB wave function, that dominates the TEE and is responsible for a positive value of TEE, which implies that the nonlocal entanglement in the RVB state is mainly encoded in the sign of the RVB wave function. Our results indicate that some information that is important for the topological property of a RVB state is missed in the effective field theory description and that a $Z_{2}$ gauge structure in the saddle point action is not enough for the RVB state to exhibit topological order, even if its spin correlation is extremely short ranged.

cond-mat.str-el

An efficient Monte Carlo algorithm for the evaluation of Renyi entanglement entropy of a general quantum dimer model at the R-K point

A highly efficient and simple to implement Monte Carlo algorithm is proposed for the evaluation of the Renyi entanglement entropy(REE) of quantum dimer model(QDM) at the Rokhsar-Kivelson(R-K) point. It makes possible the evaluation of REE at the R-K point to the thermodynamic limit for a general QDM. We apply the algorithm to QDM on both triangular and square lattice as demonstrations and find the REE on both lattices follow perfect linear scaling in the thermodynamic limit, apart from an even-odd oscillation in the latter case. We also evaluate the topological entanglement entropy(TEE) on both lattices with a subtraction procedure. While the expected TEE of $\ln2$ is clearly demonstrated for QDM on triangular lattice, a strong oscillation of the result is found for QDM on square lattice, which implies the relevance of boundary perturbation in such a critical system.

cond-mat.str-el