arXiv subjects

Yi Liu

Publications and source records attributed to Yi Liu.

At least 613 records · Page 34Linked to original sources

arXiv2022

A practical framework for multi-domain speech recognition and an instance sampling method to neural language modeling

Automatic speech recognition (ASR) systems used on smart phones or vehicles are usually required to process speech queries from very different domains. In such situations, a vanilla ASR system usually fails to perform well on every domain. This paper proposes a multi-domain ASR framework for Tencent Map, a navigation app used on smart phones and in-vehicle infotainment systems. The proposed framework consists of three core parts: a basic ASR module to generate n-best lists of a speech query, a text classification module to determine which domain the speech query belongs to, and a reranking module to rescore n-best lists using domain-specific language models. In addition, an instance sampling based method to training neural network language models (NNLMs) is proposed to address the data imbalance problem in multi-domain ASR. In experiments, the proposed framework was evaluated on navigation domain and music domain, since navigating and playing music are two main features of Tencent Map. Compared to a general ASR system, the proposed framework achieves a relative 13% $\sim$ 22% character error rate reduction on several test sets collected from Tencent Map and our in-car voice assistant.

eess.AS↗

arXiv2022

Mapping classes are almost determined by their finite quotient actions

Given any connected compact orientable surface, a pair of mapping classes are said to be procongruently conjugate if they induce a conjugate pair of outer automophisms on the profinite completion of the fundamental group of the surface. For example, this occurs if they induce conjugate outer automorphisms on every characteristic finite quotient of the fundamental group. In this paper, it is shown that every procongruent conjugacy class of mapping classes, as a subset of the surface mapping class group, is the disjoint union of at most finitely many conjugacy classes of mapping classes. For any pseudo-Anosov mapping class of a connected closed orientable surface, several topological features are confirmed to depend only on the procongurent conjugacy class of the mapping class, including: the stretching factor, the topological type of the prong singularities, the transverse orientability of the invariant foliations, and the isomorphism type of the symplectic Floer homology.

math.GT↗

arXiv2022

Commentary: analyzing binary data using MCPMod when zero counts are expected

Bretz et al (2005) proposed multiple Comparison Procedure and Modeling (MCPMod) method to design and analyze dose-finding study. Pinheiro (2014) then generalized it to various types of endpoint, including but not limited to binary endpoint, survival endpoint, count data, and longitudinal data. Pinheiro (2013) recommended to use the estimated covariance matrix from the observed data to recalculate the optimal contrast and the critical value of the test For many phase II studies it is common to have small sample sizes per arm with low placebo response rates jointly. Under such circumstances, it cannot be excluded to have a zero count observed. For example, when the placebo response rate is 10%, there is about 4% chance to observe zero responders in the placebo group, or other dose group(s), which has a similar response rate as placebo. In this manuscript, we would like to illustrate the potential problem of Pinheiro (2013) using a case study and simulations. An alternative method using Firth's logistic regression was evaluated to get a stable estimate of response for each dose group. In addition, we evaluated two options to address the issue with problematic contrast coefficients.

stat.ME↗

arXiv2022

The Volcspeech system for the ICASSP 2022 multi-channel multi-party meeting transcription challenge

This paper describes our submission to ICASSP 2022 Multi-channel Multi-party Meeting Transcription (M2MeT) Challenge. For Track 1, we propose several approaches to empower the clustering-based speaker diarization system to handle overlapped speech. Front-end dereverberation and the direction-of-arrival (DOA) estimation are used to improve the accuracy of speaker diarization. Multi-channel combination and overlap detection are applied to reduce the missed speaker error. A modified DOVER-Lap is also proposed to fuse the results of different systems. We achieve the final DER of 5.79% on the Eval set and 7.23% on the Test set. For Track 2, we develop our system using the Conformer model in a joint CTC-attention architecture. Serialized output training is adopted to multi-speaker overlapped speech recognition. We propose a neural front-end module to model multi-channel audio and train the model end-to-end. Various data augmentation methods are utilized to mitigate over-fitting in the multi-channel multi-speaker E2E system. Transformer language model fusion is developed to achieve better performance. The final CER is 19.2% on the Eval set and 20.8% on the Test set.

cs.SD↗

arXiv2022

Type-II Dirac Nodal Lines in double-kagome-layered CsV$_8$Sb$_{12}$

Lorentz-violating type-II Dirac nodal line semimetals (DNLSs), hosting curves of band degeneracy formed by two dispersion branches with the same sign of slope, represent a novel states of matter. While being studied extensively in theory, convincing experimental evidences of type-II DNLSs remain elusive. Recently, Vanadium-based kagome materials have emerged as a fertile ground to study the interplay between lattice symmetry and band topology. In this work, we study the low-energy band structure of double-kagome-layered CsV$_8$Sb$_{12}$ and identify it as a scarce type-II DNLS protected by mirror symmetry. We have observed multiple DNLs consisting of type-II Dirac cones close to or almost at the Fermi level via angle-resolved photoemission spectroscopy (ARPES). First-principle analyses show that spin-orbit coupling only opens a small gap, resulting effectively gapless ARPES spectra, yet generating large spin Berry curvature. These type-II DNLs, together with the interaction between a low-energy van Hove singularity and quasi-1D band as we observed in the same material, suggest CsV$_8$Sb$_{12}$ as an ideal platform for exploring novel transport properties such as chiral anomaly, the Klein tunneling and fractional quantum Hall effect.

cond-mat.str-el↗

arXiv2022

Aggregation Service for Federated Learning: An Efficient, Secure, and More Resilient Realization

Federated learning has recently emerged as a paradigm promising the benefits of harnessing rich data from diverse sources to train high quality models, with the salient features that training datasets never leave local devices. Only model updates are locally computed and shared for aggregation to produce a global model. While federated learning greatly alleviates the privacy concerns as opposed to learning with centralized data, sharing model updates still poses privacy risks. In this paper, we present a system design which offers efficient protection of individual model updates throughout the learning procedure, allowing clients to only provide obscured model updates while a cloud server can still perform the aggregation. Our federated learning system first departs from prior works by supporting lightweight encryption and aggregation, and resilience against drop-out clients with no impact on their participation in future rounds. Meanwhile, prior work largely overlooks bandwidth efficiency optimization in the ciphertext domain and the support of security against an actively adversarial cloud server, which we also fully explore in this paper and provide effective and efficient mechanisms. Extensive experiments over several benchmark datasets (MNIST, CIFAR-10, and CelebA) show our system achieves accuracy comparable to the plaintext baseline, with practical performance.

cs.CR↗

arXiv2022

Evolution from helical to collinear ferromagnetic order of the Eu$^{2+}$ spins in RbEu(Fe$_{1-x}$Ni$_{x}$)$_{4}$As$_{4}$

The ground-state magnetic structures of the Eu$^{2+}$ spins in recently discovered RbEu(Fe$_{1-x}$Ni$_{x}$)$_{4}$As$_{4}$ superconductors have been investigated by neutron powder diffraction measurements. It is found that as the superconductivity gets suppressed with the increase of Ni doping, the magnetic propagation vector of the Eu sublattice diminishes, corresponding to the decrease of the rotation angle between the moments in neighboring Eu layers. The ferromagnetic Eu layers are helically modulated along the $\mathit{c}$ axis with an incommensurate magnetic propagation vector in both the ferromagnetic superconductor RbEu(Fe$_{0.95}$Ni$_{0.05}$)$_{4}$As$_{4}$ and the superconducting ferromagnet RbEu(Fe$_{0.93}$Ni$_{0.07}$)$_{4}$As$_{4}$. Such a helical structure transforms into a purely collinear ferromagnetic structure for non-superconducting RbEu(Fe$_{0.91}$Ni$_{0.09}$)$_{4}$As$_{4}$, with all the Eu$^{2+}$ spins lying along the tetragonal (1 1 0) direction. The evolution from helical to collinear ferromagnetic order of the Eu$^{2+}$ spins with increasing Ni doping is supported by first-principles calculations. The variation of the rotation angle between adjacent Eu$^{2+}$ layers can be well explained by considering the change of magnetic exchange couplings mediated by the indirect Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction.

cond-mat.str-el↗

arXiv2022

Networks with pixels embedding: a method to improve noise resistance in images classification

In the task of image classification, usually, the network is sensitive to noises. For example, an image of cat with noises might be misclassified as an ostrich. Conventionally, to overcome the problem of noises, one uses the technique of data augmentation, that is, to teach the network to distinguish noises by adding more images with noises in the training dataset. In this work, we provide a noise-resistance network in images classification by introducing a technique of pixel embedding. We test the network with pixel embedding, which is abbreviated as the network with PE, on the mnist database of handwritten digits. It shows that the network with PE outperforms the conventional network on images with noises. The technique of pixel embedding can be used in many tasks of image classification to improve noise resistance.

cs.CV↗

arXiv2022

Measuring neutrino dynamics in NMSSM with a right-handed sneutrino LSP at the ILC

We study the possibility of measuring neutrino Yukawa couplings in the Next-to-Minimal Supersymmetric Standard Model with right-handed neutrinos (NMSSMr) when the lightest right-handed sneutrino is the Dark Matter (DM) candidate, by exploiting a `dijet + dilepton + Missing Transverse Energy' signature. We show that, contrary to the miminal realisation of Supersymmetry (SUSY), the MSSM, wherein the DM candidate is typically a much heavier (fermionic) neutralino state, this extended model of SUSY offers one with a much lighter (bosonic) state as DM that can then be produced at the next generation of $e^+e^-$ colliders with energies up to 500 GeV or so. The ensuing signal, energing from chargino pair production and subsequent decay, is extremely pure so it also affords one with the possibility of extracting the Yukawa parameters of the (s)neutrino sector. Altogether, our results serve the purpose of motivating searches for light DM signals at such machines, where the DM candidate can have a mass around the electroweak scale.

hep-ph↗

arXiv2022

FamilySeer: Towards Optimized Tensor Codes by Exploiting Computation Subgraph Similarity

Deploying various deep learning (DL) models efficiently has boosted the research on DL compilers. The difficulty of generating optimized tensor codes drives DL compiler to ask for the auto-tuning approaches, and the increasing demands require increasing auto-tuning efficiency and quality. Currently, the DL compilers partition the input DL models into several subgraphs and leverage the auto-tuning to find the optimal tensor codes of these subgraphs. However, existing auto-tuning approaches usually regard subgraphs as individual ones and overlook the similarities across them, and thus fail to exploit better tensor codes under limited time budgets. We propose FamilySeer, an auto-tuning framework for DL compilers that can generate better tensor codes even with limited time budgets. FamilySeer exploits the similarities and differences among subgraphs can organize them into subgraph families, where the tuning of one subgraph can also improve other subgraphs within the same family. The cost model of each family gets more purified training samples generated by the family and becomes more accurate so that the costly measurements on real hardware can be replaced with the lightweight estimation through cost model. Our experiments show that FamilySeer can generate model codes with the same code performance more efficiently than state-of-the-art auto-tuning frameworks.

cs.LG↗

arXiv2021

An invertible crystallographic representation for general inverse design of inorganic crystals with targeted properties

Realizing general inverse design could greatly accelerate the discovery of new materials with user-defined properties. However, state-of-the-art generative models tend to be limited to a specific composition or crystal structure. Herein, we present a framework capable of general inverse design (not limited to a given set of elements or crystal structures), featuring a generalized invertible representation that encodes crystals in both real and reciprocal space, and a property-structured latent space from a variational autoencoder (VAE). In three design cases, the framework generates 142 new crystals with user-defined formation energies, bandgap, thermoelectric (TE) power factor, and combinations thereof. These generated crystals, absent in the training database, are validated by first-principles calculations. The success rates (number of first-principles-validated target-satisfying crystals/number of designed crystals) ranges between 7.1% and 38.9%. These results represent a significant step toward property-driven general inverse design using generative models, although practical challenges remain when coupled with experimental synthesis.

physics.comp-ph↗

arXiv2021

PP-HumanSeg: Connectivity-Aware Portrait Segmentation with a Large-Scale Teleconferencing Video Dataset

As the COVID-19 pandemic rampages across the world, the demands of video conferencing surge. To this end, real-time portrait segmentation becomes a popular feature to replace backgrounds of conferencing participants. While feature-rich datasets, models and algorithms have been offered for segmentation that extract body postures from life scenes, portrait segmentation has yet not been well covered in a video conferencing context. To facilitate the progress in this field, we introduce an open-source solution named PP-HumanSeg. This work is the first to construct a large-scale video portrait dataset that contains 291 videos from 23 conference scenes with 14K fine-labeled frames and extensions to multi-camera teleconferencing. Furthermore, we propose a novel Semantic Connectivity-aware Learning (SCL) for semantic segmentation, which introduces a semantic connectivity-aware loss to improve the quality of segmentation results from the perspective of connectivity. And we propose an ultra-lightweight model with SCL for practical portrait segmentation, which achieves the best trade-off between IoU and the speed of inference. Extensive evaluations on our dataset demonstrate the superiority of SCL and our model. The source code is available at https://github.com/PaddlePaddle/PaddleSeg.

cs.CV↗

arXiv2021

Activation functions are not needed: the ratio net

A deep neural network for classification tasks is essentially consist of two components: feature extractors and function approximators. They usually work as an integrated whole, however, improvements on any components can promote the performance of the whole algorithm. This paper focus on designing a new function approximator. Conventionally, to build a function approximator, one usually uses the method based on the nonlinear activation function or the nonlinear kernel function and yields classical networks such as the feed-forward neural network (MLP) and the radial basis function network (RBF). In this paper, a new function approximator that is effective and efficient is proposed. Instead of designing new activation functions or kernel functions, the new proposed network uses the fractional form. For the sake of convenience, we name the network the ratio net. We compare the effectiveness and efficiency of the ratio net and that of the RBF and the MLP with various kinds of activation functions in the classification task on the mnist database of handwritten digits and the Internet Movie Database (IMDb) which is a binary sentiment analysis dataset. It shows that, in most cases, the ratio net converges faster and outperforms both the MLP and the RBF.

cs.LG↗

arXiv2021

Direct Training via Backpropagation for Ultra-low Latency Spiking Neural Networks with Multi-threshold

Spiking neural networks (SNNs) can utilize spatio-temporal information and have a nature of energy efficiency which is a good alternative to deep neural networks(DNNs). The event-driven information processing makes SNNs can reduce the expensive computation of DNNs and save a lot of energy consumption. However, high training and inference latency is a limitation of the development of deeper SNNs. SNNs usually need tens or even hundreds of time steps during the training and inference process which causes not only the increase of latency but also the waste of energy consumption. To overcome this problem, we proposed a novel training method based on backpropagation (BP) for ultra-low latency(1-2 time steps) SNN with multi-threshold. In order to increase the information capacity of each spike, we introduce the multi-threshold Leaky Integrate and Fired (LIF) model. In our proposed training method, we proposed three approximated derivative for spike activity to solve the problem of the non-differentiable issue which cause difficulties for direct training SNNs based on BP. The experimental results show that our proposed method achieves an average accuracy of 99.56%, 93.08%, and 87.90% on MNIST, FashionMNIST, and CIFAR10, respectively with only 2 time steps. For the CIFAR10 dataset, our proposed method achieve 1.12% accuracy improvement over the previously reported direct trained SNNs with fewer time steps.

cs.NE↗

arXiv2021

Dirac Nodal Lines and Nodal Loops in a Topological Kagome Superconductor CsV$_3$Sb$_5$

The intertwining of charge order, superconductivity and band topology has promoted the AV$_3$Sb$_5$ (A=K, Rb, Cs) family of materials to the center of attention in condensed matter physics. Underlying those mysterious macroscopic properties such as giant anomalous Hall conductivity (AHC) and chiral charge density wave is their nontrivial band topology. While there have been numerous experimental and theoretical works investigating the nontrivial band structure and especially the van Hove singularities, the exact topological phase of this family remains to be clarified. In this work, we identify CsV$_3$Sb$_5$ as a Dirac nodal line semimetal based on the observation of multiple Dirac nodal lines and loops close to the Fermi level. Combining photoemission spectroscopy and density functional theory, we identify two groups of Dirac nodal lines along $k_z$ direction and one group of Dirac nodal loops in the A-H-L plane. These nodal loops are located at the Fermi level within the instrumental resolution limit. Importantly, our first-principle analyses indicate that these nodal loops may be a crucial source of the mysterious giant AHC observed. Our results not only provide a clear picture to categorize the band structure topology of this family of materials, but also suggest the dominant role of topological nodal loops in shaping their transport behavior.

cond-mat.str-el↗

arXiv2021

EdgeFlow: Achieving Practical Interactive Segmentation with Edge-Guided Flow

High-quality training data play a key role in image segmentation tasks. Usually, pixel-level annotations are expensive, laborious and time-consuming for the large volume of training data. To reduce labelling cost and improve segmentation quality, interactive segmentation methods have been proposed, which provide the result with just a few clicks. However, their performance does not meet the requirements of practical segmentation tasks in terms of speed and accuracy. In this work, we propose EdgeFlow, a novel architecture that fully utilizes interactive information of user clicks with edge-guided flow. Our method achieves state-of-the-art performance without any post-processing or iterative optimization scheme. Comprehensive experiments on benchmarks also demonstrate the superiority of our method. In addition, with the proposed method, we develop an efficient interactive segmentation tool for practical data annotation tasks. The source code and tool is avaliable at https://github.com/PaddlePaddle/PaddleSeg.

cs.CV↗

arXiv2021

Doping evolution of superconductivity, charge order and band topology in hole-doped topological kagome superconductors Cs(V$_{1-x}$Ti$_x$)$_3$Sb$_5$

The newly discovered Kagome superconductors $A$V$_3$Sb$_5$ ($A$ = K, Rb, Cs) exhibit superconductivity, charge order, and band topology simultaneously. To explore the intricate interplay between the superconducting and charge orders, we investigate the doping evolution of superconductivity, charge-density-wave (CDW) order, and band topology in doped topological kagome superconductors Cs(V$_{1-x}$Ti$_x$)$_3$Sb$_5$ where the Ti-dopant introduces hole-like charge carriers. Despite the absence of the CDW phase transition in doped compounds even for the lowest doping level of $x=0.047$, the superconductivity survives in all doped samples with enhanced critical temperatures. The high-resolution angle-resolved photoemission spectroscopy (ARPES) measurements reveal that the Ti-dopant in the kagome plane lowers the chemical potential, pushing the van Hove singularity (VHS) at $M$ point above the Fermi level. First-principle simulations corroborate the doping evolution of the band structure observed in ARPES, and affirm that the CDW instability does not occur once the VHS is pushed above the Fermi level, explaining the absence of the CDW ordering in our doped samples Cs(V$_{1-x}$Ti$_x$)$_3$Sb$_5$. Our results demonstrate a competition between the CDW and superconducting orders in the kagome-metal superconductor CsV$_3$Sb$_5$, although the superconductivity is likely inconsequential of the CDW order.

cond-mat.supr-con↗

arXiv2021

Damped Anderson Mixing for Deep Reinforcement Learning: Acceleration, Convergence, and Stabilization

Anderson mixing has been heuristically applied to reinforcement learning (RL) algorithms for accelerating convergence and improving the sampling efficiency of deep RL. Despite its heuristic improvement of convergence, a rigorous mathematical justification for the benefits of Anderson mixing in RL has not yet been put forward. In this paper, we provide deeper insights into a class of acceleration schemes built on Anderson mixing that improve the convergence of deep RL algorithms. Our main results establish a connection between Anderson mixing and quasi-Newton methods and prove that Anderson mixing increases the convergence radius of policy iteration schemes by an extra contraction factor. The key focus of the analysis roots in the fixed-point iteration nature of RL. We further propose a stabilization strategy by introducing a stable regularization term in Anderson mixing and a differentiable, non-expansive MellowMax operator that can allow both faster convergence and more stable behavior. Extensive experiments demonstrate that our proposed method enhances the convergence, stability, and performance of RL algorithms.

cs.LG↗