Search

Deep learning has become an indispensable part of life, such as face recognition, NLP, etc., but the training of deep model has always been a challenge, and in recent years, the complexity of training data and models has shown explosive growth, so the training method is gradually transformed into distributed training. Classical synchronization strategy can guarantee accuracy but frequent communication can lead to a slow training speed, although asynchronous strategy training speed but can not guarantee the accuracy, and in the face of the training of the heterogeneous cluster, the above work is not efficient to work, on the one hand, can cause serious waste of resources, on the other hand, frequent communication also made slow training speed, so this paper proposes a semi-synchronous training strategy based on local-SDG, effectively improve the utilization efficiency of heterogeneous resources cluster and reduce communication overhead, to accelerate the training and ensure the accuracy of the model., Comment: arXiv admin note: text overlap with arXiv:1910.13598 by other authors

Delusive attacks aim to substantially deteriorate the test accuracy of the learning model by slightly perturbing the features of correctly labeled training examples. By formalizing this malicious attack as finding the worst-case training data within a specific $\infty$-Wasserstein ball, we show that minimizing adversarial risk on the perturbed data is equivalent to optimizing an upper bound of natural risk on the original data. This implies that adversarial training can serve as a principled defense against delusive attacks. Thus, the test accuracy decreased by delusive attacks can be largely recovered by adversarial training. To further understand the internal mechanism of the defense, we disclose that adversarial training can resist the delusive perturbations by preventing the learner from overly relying on non-robust features in a natural setting. Finally, we complement our theoretical findings with a set of experiments on popular benchmark datasets, which show that the defense withstands six different practical attacks. Both theoretical and empirical results vote for adversarial training when confronted with delusive adversaries., Comment: NeurIPS 2021

Computational fluid dynamics (CFD) is a branch of fluid mechanics that uses numerical methods to solve fluid flows. The finite volume method (FVM) is an important one. In FVM, space is discretized to many grid cells. When the number of grid cells grows, massive computing resources are needed correspondingly. Recently, quantum computing has been proven to outperform a classical computer on specific computational tasks. However, the quantum CFD (QCFD) solver remains a challenge because the conversion between the classical and quantum data would become the bottleneck for the time complexity. Here we propose a QCFD solver with exponential speedup over classical counterparts and focus on how a quantum computer handles classical input and output. By utilizing quantum random access memory, the algorithm realizes sublinear time at every iteration step. The QCFD solver could allow new frontiers in the CFD area by allowing a finer mesh and faster calculation., Comment: 11 pages, 3 figures

Compared with single-label image classification, multi-label image classification is more practical and challenging. Some recent studies attempted to leverage the semantic information of categories for improving multi-label image classification performance. However, these semantic-based methods only take semantic information as type of complements for visual representation without further exploitation. In this paper, we present an innovative path towards the solution of the multi-label image classification which considers it as a dictionary learning task. A novel end-to-end model named Deep Semantic Dictionary Learning (DSDL) is designed. In DSDL, an auto-encoder is applied to generate the semantic dictionary from class-level semantics and then such dictionary is utilized for representing the visual features extracted by Convolutional Neural Network (CNN) with label embeddings. The DSDL provides a simple but elegant way to exploit and reconcile the label, semantic and visual spaces simultaneously via conducting the dictionary learning among them. Moreover, inspired by iterative optimization of traditional dictionary learning, we further devise a novel training strategy named Alternately Parameters Update Strategy (APUS) for optimizing DSDL, which alternately optimizes the representation coefficients and the semantic dictionary in forward and backward propagation. Extensive experimental results on three popular benchmarks demonstrate that our method achieves promising performances in comparison with the state-of-the-arts. Our codes and models have been released at {https://github.com/ZFT-CQU/DSDL}., Comment: Accepted by AAAI2021

Topological crystalline superconductors have attracted rapidly rising attention due to the possibility of higher-order phases, which support Majorana modes on boundaries in $d-2$ or lower dimensions. However, although the classification and bulk topological invariants in such systems have been well studied, it is generally difficult to faithfully predict the boundary Majoranas from the band-structure information due to the lack of well-established bulk-boundary correspondence. Here we propose a protocol for deriving symmetry indicators that depend on a minimal set of necessary symmetry data of the bulk bands and can diagnose boundary features. Specifically, to obtain indicators manifesting clear bulk-boundary correspondence, we combine the topological crystal classification scheme in the real space and a twisted equivariant K group analysis in the momentum space. The key step is to disentangle the generally mixed strong and weak indicators through a systematic basis-matching procedure between our real-space and momentum-space approaches. We demonstrate our protocol using an example of two-dimensional time-reversal odd-parity superconductors, where the inversion symmetry is known to protect a higher-order phase with corner Majoranas. Symmetry indicators derived from our protocol can be readily applied to ab initio database and could fuel material predictions for strong and weak topological crystalline superconductors with various boundary features., Comment: 27 pages, 9 figures, 2 tables

Researchers have been overwhelmed by the explosion of research articles published by various research communities. Many research scholarly websites, search engines, and digital libraries have been created to help researchers identify potential research topics and keep up with recent progress on research of interests. However, it is still difficult for researchers to keep track of the research topic diffusion and evolution without spending a large amount of time reviewing numerous relevant and irrelevant articles. In this paper, we consider a novel topic diffusion discovery technique. Specifically, we propose using a Deep Non-negative Autoencoder with information divergence measurement that monitors evolutionary distance of the topic diffusion to understand how research topics change with time. The experimental results show that the proposed approach is able to identify the evolution of research topics as well as to discover topic diffusions in online fashions.

Patients with Acute Kidney Injury (AKI) increase mortality, morbidity, and long-term adverse events. Therefore, early identification of AKI may improve renal function recovery, decrease comorbidities, and further improve patients' survival. To control certain risk factors and develop targeted prevention strategies are important to reduce the risk of AKI. Drug-drug interactions and drug-disease interactions are critical issues for AKI. Typical statistical approaches cannot handle the complexity of drug-drug and drug-disease interactions. In this paper, we propose a novel learning algorithm, Deep Rule Forests (DRF), which discovers rules from multilayer tree models as the combinations of drug usages and disease indications to help identify such interactions. We found that several disease and drug usages are considered having significant impact on the occurrence of AKI. Our experimental results also show that the DRF model performs comparatively better than typical tree-based and other state-of-the-art algorithms in terms of prediction accuracy and model interpretability.

The discovery of the quantization of particle transport in adiabatic pumping cycles of periodic structures by Thouless [Phys. Rev. B 27, 6083 (1983)] linked the Chern number, a topological invariant characterizing the quantum Hall effect in two-dimensional electron gases, with the topology of dynamical periodic systems in one dimension. Here, we demonstrate its counterpart for higher-order topology. Specifically, we show that adiabatic cycles in two-dimensional crystals with vanishing dipole moments (and therefore zero overall particle transport) can nevertheless be topologically nontrivial. These cycles are associated with higher-order topology and can be diagnosed by their ability to produce corner-to-corner transport in certain metamaterial platforms. We experimentally verify the corner to corner transport associated with this topological pump by using an array of photonic waveguides adiabatically modulated in their separations and refractive indices. By mapping the dynamical phenomenon demonstrated here from two spatial and one temporal dimensions to three spatial dimensions, our observations are equivalent to the observation of chiral hinge states in a three-dimensional second-order topological insulator., Comment: 12 pages, 8 figures

Reinforcement learning has achieved great success in various applications. To learn an effective policy for the agent, it usually requires a huge amount of data by interacting with the environment, which could be computational costly and time consuming. To overcome this challenge, the framework called Reinforcement Learning with Expert Demonstrations (RLED) was proposed to exploit the supervision from expert demonstrations. Although the RLED methods can reduce the number of learning iterations, they usually assume the demonstrations are perfect, and thus may be seriously misled by the noisy demonstrations in real applications. In this paper, we propose a novel framework to adaptively learn the policy by jointly interacting with the environment and exploiting the expert demonstrations. Specifically, for each step of the demonstration trajectory, we form an instance, and define a joint loss function to simultaneously maximize the expected reward and minimize the difference between agent behaviors and demonstrations. Most importantly, by calculating the expected gain of the value function, we assign each instance with a weight to estimate its potential utility, and thus can emphasize the more helpful demonstrations while filter out noisy ones. Experimental results in various environments with multiple popular reinforcement learning algorithms show that the proposed approach can learn robustly with noisy demonstrations, and achieve higher performance in fewer iterations.

We present a novel deep learning framework named the Iteratively Optimized Patch Label Inference Network (IOPLIN) for automatically detecting various pavement distresses that are not solely limited to specific ones, such as cracks and potholes. IOPLIN can be iteratively trained with only the image label via the Expectation-Maximization Inspired Patch Label Distillation (EMIPLD) strategy, and accomplish this task well by inferring the labels of patches from the pavement images. IOPLIN enjoys many desirable properties over the state-of-the-art single branch CNN models such as GoogLeNet and EfficientNet. It is able to handle images in different resolutions, and sufficiently utilize image information particularly for the high-resolution ones, since IOPLIN extracts the visual features from unrevised image patches instead of the resized entire image. Moreover, it can roughly localize the pavement distress without using any prior localization information in the training phase. In order to better evaluate the effectiveness of our method in practice, we construct a large-scale Bituminous Pavement Disease Detection dataset named CQU-BPDD consisting of 60,059 high-resolution pavement images, which are acquired from different areas at different times. Extensive results on this dataset demonstrate the superiority of IOPLIN over the state-of-the-art image classification approaches in automatic pavement distress detection. The source codes of IOPLIN are released on \url{https://github.com/DearCaat/ioplin}, and the CQU-BPDD dataset is able to be accessed on \url{https://dearcaat.github.io/CQU-BPDD/}., Comment: Published on IEEE Trans on ITS

We study bosonic symmetry protected topological (SPT) phases with $C_{2}$ rotational symmetry in four spatial dimensions which is not captured by the group cohomology classification. By using the topological crystal approach, we show that the topological crystalline state of this SPT phase is given by placing an $E_{8}$ state on the two dimensional rotational invariant plane, which provides a simple physical picture of this phase. Based on this understanding, we show that a variant of QED4 with charge-1 and charge-3 Dirac fermions is a field theoretical description of the three dimensional boundary. We also discuss the connection to symmetric gapped boundary with topological order and its anomaly signature., Comment: 6 pages, 1 figure