Search

We revisit the ground state of the Hubbard model on 2-legged ladders in this work. We perform DMRG calculation on large system sizes with large kept states and perform extrapolation of DMRG results with truncation errors in the converged region. We find the superconducting correlation exponent $K_{sc}$ extracted from the pair-pair correlation is very sensitive to the position of the reference bond, reflecting a huge boundary effect on it. By systematically removing the effects from boundary conditions, finite sizes, and truncation errors in DMRG, we obtain the most accurate value of $K_{sc}$ and $K_\rho$ so far with DMRG. With these exponents, we confirm that the 2-legged Hubbard model is in the Luther-Emery liquid phase with $K_{sc} \cdot K_\rho = 1$ from tiny doping near half-filling to $1/8$ hole doping. When the doping is increased to $\delta \gtrapprox 1/6$, the behaviors of charge, pairing, and spin correlations don't change qualitatively, but the relationship $K_{sc} \cdot K_\rho = 1$ is likely to be violated. With the further increase of the doping to $\delta = 1/3$, the quasi long-ranged charge correlation turns to a true long-ranged charge order and the spin gap is closed, while the pair-pair correlation still decays algebraically. Our work provides a standard way to analyze the correlation functions when studying systems with open boundary conditions., Comment: 11 pages,11 figures,4 tables

We propose Equiangular Basis Vectors (EBVs) for classification tasks. In deep neural networks, models usually end with a k-way fully connected layer with softmax to handle different classification tasks. The learning objective of these methods can be summarized as mapping the learned feature representations to the samples' label space. While in metric learning approaches, the main objective is to learn a transformation function that maps training data points from the original space to a new space where similar points are closer while dissimilar points become farther apart. Different from previous methods, our EBVs generate normalized vector embeddings as "predefined classifiers" which are required to not only be with the equal status between each other, but also be as orthogonal as possible. By minimizing the spherical distance of the embedding of an input between its categorical EBV in training, the predictions can be obtained by identifying the categorical EBV with the smallest distance during inference. Various experiments on the ImageNet-1K dataset and other downstream tasks demonstrate that our method outperforms the general fully connected classifier while it does not introduce huge additional computation compared with classical metric learning methods. Our EBVs won the first place in the 2022 DIGIX Global AI Challenge, and our code is open-source and available at https://github.com/NJUST-VIPGroup/Equiangular-Basis-Vectors., Comment: CVPR 2023

GPT series models, such as GPT-3, CodeX, InstructGPT, ChatGPT, and so on, have gained considerable attention due to their exceptional natural language processing capabilities. However, despite the abundance of research on the difference in capabilities between GPT series models and fine-tuned models, there has been limited attention given to the evolution of GPT series models' capabilities over time. To conduct a comprehensive analysis of the capabilities of GPT series models, we select six representative models, comprising two GPT-3 series models (i.e., davinci and text-davinci-001) and four GPT-3.5 series models (i.e., code-davinci-002, text-davinci-002, text-davinci-003, and gpt-3.5-turbo). We evaluate their performance on nine natural language understanding (NLU) tasks using 21 datasets. In particular, we compare the performance and robustness of different models for each task under zero-shot and few-shot scenarios. Our extensive experiments reveal that the overall ability of GPT series models on NLU tasks does not increase gradually as the models evolve, especially with the introduction of the RLHF training strategy. While this strategy enhances the models' ability to generate human-like responses, it also compromises their ability to solve some tasks. Furthermore, our findings indicate that there is still room for improvement in areas such as model robustness.

Extracting useful signals is key to both classical and quantum technologies. Conventional noise filtering methods rely on different patterns of signal and noise in frequency or time domains, thus limiting their scope of application, especially in quantum sensing. Here, we propose a signal-nature-based (not signal-pattern-based) approach which singles out a quantum signal from its classical noise background by employing the intrinsic quantum nature of the system. We design a novel protocol to extract the Quantum Correlation signal and use it to single out the signal of a remote nuclear spin from its overwhelming classical noise backgrounds, which is impossible to be accomplished by conventional filter methods. Our work demonstrates the quantum/classical nature as a new degree of freedom in quantum sensing. The further generalization of this quantum nature-based method opens a new direction in quantum research.

To develop effective thermal management strategies for GaN transistors, it is essential to accurately predict the device junction temperature. Since the width of the heat generation in the devices is comparable to phonon mean free paths of GaN, phonon ballistic transport exists and can significantly affect the heat transport process, which necessitates a thorough understanding of the influence of the phonon ballistic effects in GaN transistors. In this paper, the ballistic-diffusive phonon transport in GaN-on-SiC devices is examined by measuring the hotspot temperature using the thermoreflectance thermal imaging TTI combined with the hybrid phonon Monte Carlo-diffusion simulations. A series of Au heaters are fabricated on the top of the GaN layer to quantitatively mimic the different heat source distributions during device operation. The experimental and simulation results show a good consistency and both indicate that the phonon ballistic effects can significantly increase the hotspot temperature. With the size of the heat source decreasing, the errors of Fourier's law-based predictions increase, which emphasizes the necessity to carefully consider the phonon ballistic transport in device thermal simulations.

Simplicity Bias (SB) is a phenomenon that deep neural networks tend to rely favorably on simpler predictive patterns but ignore some complex features when applied to supervised discriminative tasks. In this work, we investigate SB in long-tailed image recognition and find the tail classes suffer more severely from SB, which harms the generalization performance of such underrepresented classes. We empirically report that self-supervised learning (SSL) can mitigate SB and perform in complementary to the supervised counterpart by enriching the features extracted from tail samples and consequently taking better advantage of such rare samples. However, standard SSL methods are designed without explicitly considering the inherent data distribution in terms of classes and may not be optimal for long-tailed distributed data. To address this limitation, we propose a novel SSL method tailored to imbalanced data. It leverages SSL by triple diverse levels, i.e., holistic-, partial-, and augmented-level, to enhance the learning of predictive complex patterns, which provides the potential to overcome the severe SB on tail data. Both quantitative and qualitative experimental results on five long-tailed benchmark datasets show our method can effectively mitigate SB and significantly outperform the competing state-of-the-arts.

This paper targets at improving the generalizability of hypergraph neural networks in the low-label regime, through applying the contrastive learning approach from images/graphs (we refer to it as HyperGCL). We focus on the following question: How to construct contrastive views for hypergraphs via augmentations? We provide the solutions in two folds. First, guided by domain knowledge, we fabricate two schemes to augment hyperedges with higher-order relations encoded, and adopt three vertex augmentation strategies from graph-structured data. Second, in search of more effective views in a data-driven manner, we for the first time propose a hypergraph generative model to generate augmented views, and then an end-to-end differentiable pipeline to jointly learn hypergraph augmentations and model parameters. Our technical innovations are reflected in designing both fabricated and generative augmentations of hypergraphs. The experimental findings include: (i) Among fabricated augmentations in HyperGCL, augmenting hyperedges provides the most numerical gains, implying that higher-order information in structures is usually more downstream-relevant; (ii) Generative augmentations do better in preserving higher-order information to further benefit generalizability; (iii) HyperGCL also boosts robustness and fairness in hypergraph representation learning. Codes are released at https://github.com/weitianxin/HyperGCL., Comment: NeurIPS 2022. Supplementary materials are available at https://weitianxin.github.io/files/neurips22_hypergcl_appendix.pdf

In this paper, we propose Suppression-Enhancing Mask based attention and Interactive Channel transformatiON (SEMICON) to learn binary hash codes for dealing with large-scale fine-grained image retrieval tasks. In SEMICON, we first develop a suppression-enhancing mask (SEM) based attention to dynamically localize discriminative image regions. More importantly, different from existing attention mechanism simply erasing previous discriminative regions, our SEM is developed to restrain such regions and then discover other complementary regions by considering the relation between activated regions in a stage-by-stage fashion. In each stage, the interactive channel transformation (ICON) module is afterwards designed to exploit correlations across channels of attended activation tensors. Since channels could generally correspond to the parts of fine-grained objects, the part correlation can be also modeled accordingly, which further improves fine-grained retrieval accuracy. Moreover, to be computational economy, ICON is realized by an efficient two-step process. Finally, the hash learning of our SEMICON consists of both global- and local-level branches for better representing fine-grained objects and then generating binary hash codes explicitly corresponding to multiple levels. Experiments on five benchmark fine-grained datasets show our superiority over competing methods., Comment: ECCV 2022

Single color centers in solid have emerged as promising physical platforms for quantum information science. Creating these centers with excellent quantum properties is a key foundation for further technological developments. In particular, the microscopic understanding of the spin bath environments is the key to engineer color centers for quantum control. In this work, we propose and demonstrate a distinct high-temperature annealing (HTA) approach for creating high-quality nitrogen vacancy (NV) centers in implantation-free diamonds. Simultaneously using the created NV centers as probes for their local environment we verify that no damage was microscopically induced by the HTA. Nearly all single NV centers created in ultra-low-nitrogen-concentration membranes possess stable and Fourier-transform-limited optical spectra. Furthermore, HTA strongly reduces noise sources naturally grown in ensemble samples, and leads to more than three-fold improvements of decoherence time and sensitivity. We also verify that the vacancy activation and defect reformation, especially H3 and P1 centers, can explain the reconfiguration between spin baths and color centers. This novel approach will become a powerful tool in vacancy-based quantum technology.

To understand the microscopic mechanism of the charge order observed in the parent compound of the infinite-layer nickelate superconductors, we consider a minimal three-legged model consisting of a two-legged Hubbard ladder for the Ni $3d_{x^2-y^2}$ electrons and a free conduction electron chain from the rare-earths. With highly accurate density matrix renormalization group calculations, when the chemical potential difference is adjusted to make the Hubbard ladder with $1/3$ hole doping, we find a long-range charge order with period $3$ in the ground state of the model, while the spin excitation has a small energy gap. Moreover, the electron pair-pair correlation has a quasi-long-range behavior, indicating an instability of superconductivity even at half-filling. As a comparison, the same method is applied to a pure two-legged Hubbard model with $1/3$ hole doping in which the period-3 charge order is a quasi-long range one. The difference between them demonstrates that the free electron chain of the three-legged ladder plays the role of a charge reservoir and enhances the charge order in the undoped infinite-layer nickelates., Comment: 9 pages, 10 figures

We consider monotone mean-variance (MMV) portfolio selection problems with a conic convex constraint under diffusion models, and their counterpart problems under mean-variance (MV) preferences. We obtain the precommitted optimal strategies to both problems in closed form and find that they coincide, without and with the presence of the conic constraint. This result generalizes the equivalence between MMV and MV preferences from non-constrained cases to a specific constrained case. A comparison analysis reveals that the orthogonality property under the conic convex set is a key to ensuring the equivalence result.

Brooks and Makover developed a combinatorial model of random hyperbolic surfaces by gluing certain hyperbolic ideal triangles. In this paper we show that for any $\epsilon>0$, as the number of ideal triangles goes to infinity, a generic hyperbolic surface in Brooks-Makover's model has Cheeger constant less than $\frac{3}{2\pi}+\epsilon$., Comment: International Mathematics Research Notices (IMRN), to appear

Let $\mathcal{M}_{g,n(g)}$ be the moduli space of hyperbolic surfaces of genus $g$ with $n(g)$ punctures endowed with the Weil-Petersson metric. In this paper we study the asymptotic behavior of the Cheeger constants and spectral gaps of random hyperbolic surfaces in $\mathcal{M}_{g,n(g)}$, when $n(g)$ grows slower than $g$ as $g\to \infty$., Comment: 38 pages, 2 figures. This paper is rewritten

Exact assessment of thermal spreading resistance is of great importance to the thermal management of electronic devices, especially when completely considering the heat conduction process from the nanoscale heat source to the macroscopic scale heat sink. The existing simulation methods are either based on convectional Fourier's law or limited to small system sizes, making it difficult to accurately and efficiently study the cross-scale heat transfer. In this paper, a hybrid phonon Monte Carlo-diffusion method that couples phonon Monte Carlo (MC) method with Fourier's law by dividing the computational domain is adopted to analyze thermal spreading resistance in ballistic-diffusive regime. Compared with phonon MC simulation, the junction temperature of the hybrid method has the same precision, while the time costs could be reduced up to 2 orders of magnitude at most. Furthermore, the simulation results indicate that the heating scheme has a remarkable impact on phonon transport. The thermal resistance of the heat source (HS) scheme can be larger than that of the heat flux (HF) scheme, which is opposite from the prediction of Fourier's law. In the HS scheme, the enhanced phonon-boundary scattering counteracts the broadening of the heat source, leading to a stronger ballistic effect as the heat source thickness decreases. The conclusion is verified by a one-dimensional thermal resistance model. This work has opened up an opportunity for the fast and extensive thermal modeling of cross-scale heat transfer in electronic devices and highlighted the influence of heating schemes.