Search arXiv⌕ Search

arXiv subjects

Chao Zhang

Publications and source records attributed to Chao Zhang.

At least 793 records · Page 44Linked to original sources
arXiv2022

Turn-Taking Prediction for Natural Conversational Speech

While a streaming voice assistant system has been used in many applications, this system typically focuses on unnatural, one-shot interactions assuming input from a single voice query without hesitation or disfluency. However, a common conversational utterance often involves multiple queries with turn-taking, in addition to disfluencies. These disfluencies include pausing to think, hesitations, word lengthening, filled pauses and repeated phrases. This makes doing speech recognition with conversational speech, including one with multiple queries, a challenging task. To better model the conversational interaction, it is critical to discriminate disfluencies and end of query in order to allow the user to hold the floor for disfluencies while having the system respond as quickly as possible when the user has finished speaking. In this paper, we present a turntaking predictor built on top of the end-to-end (E2E) speech recognizer. Our best system is obtained by jointly optimizing for ASR task and detecting when the user is paused to think or finished speaking. The proposed approach demonstrates over 97% recall rate and 85% precision rate on predicting true turn-taking with only 100 ms latency on a test set designed with 4 types of disfluencies inserted in conversational utterances.

cs.CL↗
arXiv2022

SciAnnotate: A Tool for Integrating Weak Labeling Sources for Sequence Labeling

Weak labeling is a popular weak supervision strategy for Named Entity Recognition (NER) tasks, with the goal of reducing the necessity for hand-crafted annotations. Although there are numerous remarkable annotation tools for NER labeling, the subject of integrating weak labeling sources is still unexplored. We introduce a web-based tool for text annotation called SciAnnotate, which stands for scientific annotation tool. Compared to frequently used text annotation tools, our annotation tool allows for the development of weak labels in addition to providing a manual annotation experience. Our tool provides users with multiple user-friendly interfaces for creating weak labels. SciAnnotate additionally allows users to incorporate their own language models and visualize the output of their model for evaluation. In this study, we take multi-source weak label denoising as an example, we utilized a Bertifying Conditional Hidden Markov Model to denoise the weak label generated by our tool. We also evaluate our annotation tool against the dataset provided by Mysore which contains 230 annotated materials synthesis procedures. The results shows that a 53.7% reduction in annotation time obtained AND a 1.6\% increase in recall using weak label denoising. Online demo is available at https://sciannotate.azurewebsites.net/(demo account can be found in README), but we don't host a model server with it, please check the README in supplementary material for model server usage.

cs.CL↗
arXiv2022

Experimental beating the standard quantum limit under non-markovian dephasing environment

Entanglement enhanced quantum metrology has been well investigated for beating the standard quantum limit (SQL). However, the metrological advantage of entangled states becomes much more elusive in the presence of noise. Under strictly Markovian dephasing noise, the uncorrelated and maximally entangled states achieve exactly the same measurement precision. However, it was predicted that in a non-Markovian dephasing channel, the entangled probes can recover their metrological advantage. Here, by using a highly controlled photonic system, we simulate a non-Markovian dephasing channel fulfill the quadratic decay behaviour. Under such a channel, we demonstrate the GHZ states can surpass the SQL in a scaling manner, up to six photons. Since the quadratic decay behavior is quite general for short time expansion in open quantum systems (also known as the quantum Zeno effect), our results suggest a universal and scalable method to beat the SQL in the real-word metrology tasks.

quant-ph↗
arXiv2022

An EUV jet driven by a series of transition region micro-jets

Jets are one of the most common eruptive events in the solar atmosphere, and they are believed to be important in the context of coronal heating and solar wind acceleration. We present an observational study on a sequence of jets with the data acquired with the Solar Dynamics Observatory (SDO) and the Interface Region Imaging Spectrograph (IRIS). This sequence is peculiar in that an EUV jet, $\sim29\arcsec$ long and with a dome-like base, appears to be a consequence of a series of transition region (TR) micro-jets that are a few arcsecs in length.We find that the occurrence of any TR micro-jets is always associated with the change of geometry of micro-loops at the footpoints of the microjets. A bundle of TR flux ropes is seen to link a TR micro-jet to the dome-like structure at the base of the EUV jet. This bundle rises as a response to the TR micro-jets, with the rising motion eventually triggering the EUV jet. We propose a scenario involving a set of magnetic reconnections, in which the series of TR micro-jets are associated with the processes to remove the constraints to the TR flux ropes and thus allow them to rise and trigger the EUV jet. Our study demonstrates that small-scale dynamics in the lower solar atmosphere are crucial in understanding the energy and mass connection between the corona and the solar lower atmosphere, even though many of them might not pump mass and energy to the corona directly.

astro-ph.SR↗
arXiv2022

Detection of scalar fields by Extreme Mass Ratio Inspirals with a Kerr black hole

We study extreme mass ratio inspirals occurring in modified gravity, for which the system is modeled by a small compact object with scalar charge spiraling into a supermassive Kerr black hole. Besides the tensorial gravitational waves arising from the metric perturbations, radiations are also induced by the scalar field. The relevant metric and scalar perturbations are triggered by the orbital motion of the small object, which give rise to a system of inhomogeneous differential equations under the adiabatic approximation. Such a system of equations is then solved numerically using Green's function furnished by the solutions of the corresponding homogeneous equations. To explore the present scenario from an observational perspective, we investigate how the pertinent observables are dependent on specific spacetime configurations. In this regard, the energy fluxes and the gravitational wave dephasing accumulated during the process are evaluated, as functions of the scalar charge, mass ratio, and spin of the central supermassive black hole. In particular, the presence of additional scalar emission leads to a more significant rate of overall energy loss, which, in turn, decreases the total number of orbital cycles before the small object plunges into the central black hole. Moreover, for a central black hole with a higher spin, the imprints of the scalar charge on the resultant gravitational radiations are found to be more significant, which indicates the possibility of detecting the scalar charge.

gr-qc↗
arXiv2022

Adaptive Marginalized Semantic Hashing for Unpaired Cross-Modal Retrieval

In recent years, Cross-Modal Hashing (CMH) has aroused much attention due to its fast query speed and efficient storage. Previous literatures have achieved promising results for Cross-Modal Retrieval (CMR) by discovering discriminative hash codes and modality-specific hash functions. Nonetheless, most existing CMR works are subjected to some restrictions: 1) It is assumed that data of different modalities are fully paired, which is impractical in real applications due to sample missing and false data alignment, and 2) binary regression targets including the label matrix and binary codes are too rigid to effectively learn semantic-preserving hash codes and hash functions. To address these problems, this paper proposes an Adaptive Marginalized Semantic Hashing (AMSH) method which not only enhances the discrimination of latent representations and hash codes by adaptive margins, but also can be used for both paired and unpaired CMR. As a two-step method, in the first step, AMSH generates semantic-aware modality-specific latent representations with adaptively marginalized labels, which enlarges the distances between different classes, and exploits the labels to preserve the inter-modal and intra-modal semantic similarities into latent representations and hash codes. In the second step, adaptive margin matrices are embedded into the hash codes, and enlarge the gaps between positive and negative bits, which improves the discrimination and robustness of hash functions. On this basis, AMSH generates similarity-preserving hash codes and robust hash functions without strict one-to-one data correspondence requirement. Experiments are conducted on several benchmark datasets to demonstrate the superiority and flexibility of AMSH over some state-of-the-art CMR methods.

cs.MM↗
arXiv2022

Learning to Predict Diverse Human Motions from a Single Image via Mixture Density Networks

Human motion prediction, which plays a key role in computer vision, generally requires a past motion sequence as input. However, in real applications, a complete and correct past motion sequence can be too expensive to achieve. In this paper, we propose a novel approach to predicting future human motions from a much weaker condition, i.e., a single image, with mixture density networks (MDN) modeling. Contrary to most existing deep human motion prediction approaches, the multimodal nature of MDN enables the generation of diverse future motion hypotheses, which well compensates for the strong stochastic ambiguity aggregated by the single input and human motion uncertainty. In designing the loss function, we further introduce the energy-based formulation to flexibly impose prior losses over the learnable parameters of MDN to maintain motion coherence as well as improve the prediction accuracy by customizing the energy functions. Our trained model directly takes an image as input and generates multiple plausible motions that satisfy the given condition. Extensive experiments on two standard benchmark datasets demonstrate the effectiveness of our method in terms of prediction diversity and accuracy.

cs.CV↗
arXiv2022

Ergodic Rate Analysis of STAR-RIS Aided NOMA Systems

This letter analyzes the ergodic rates of a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) aided non-orthogonal multiple access (NOMA) system, where the direct links from the base station to cell-edge users are non-line-of-sight due to obstacles, and STAR-RIS is used to provide line-of-sight links to these cell-edge users. By fitting the distribution of the composite channel power gain to a gamma distribution, we derive the closed-form expressions of ergodic rates and high signal-to-noise ratio (SNR) slopes for cell-edge users. Numerical results reveal that 1) the ergodic rates increase with the number of STAR-RIS elements, and the high SNR slopes are fixed as constants; 2) STAR-RIS aided NOMA systems achieve higher ergodic rates than conventional RIS aided NOMA systems.

cs.IT↗
arXiv2022

A Reachability Index for Recursive Label-Concatenated Graph Queries

Reachability queries checking the existence of a path from a source node to a target node are fundamental operators for querying and processing graph data. Current approaches for index-based evaluation of reachability queries either focus on plain reachability or constraint-based reachability with alternation only. In this paper, for the first time we study the problem of index-based processing for recursive label-concatenated reachability queries, referred to as RLC queries. These queries check the existence of a path that can satisfy the constraint defined by a concatenation of at most k edge labels under the Kleene plus. Many practical graph database and network analysis applications exhibit RLC queries. However, their evaluation remains prohibitive in current graph database engines. We introduce the RLC index, the first reachability index to efficiently process RLC queries. The RLC index checks whether the source vertex can reach an intermediate vertex that can also reach the target vertex under a recursive label-concatenated constraint. We propose an indexing algorithm to build the RLC index, which guarantees the soundness and the completeness of query execution and avoids recording redundant index entries. Comprehensive experiments on real-world graphs show that the RLC index can significantly reduce both the offline processing cost and the memory overhead of transitive closure while improving query processing up to six orders of magnitude over online traversals. Finally, our open-source implementation of the RLC index significantly outperforms current mainstream graph engines for evaluating RLC queries.

cs.DB↗
arXiv2022

Tandem Multitask Training of Speaker Diarisation and Speech Recognition for Meeting Transcription

Self-supervised-learning-based pre-trained models for speech data, such as Wav2Vec 2.0 (W2V2), have become the backbone of many speech tasks. In this paper, to achieve speaker diarisation and speech recognition using a single model, a tandem multitask training (TMT) method is proposed to fine-tune W2V2. For speaker diarisation, the tasks of voice activity detection (VAD) and speaker classification (SC) are required, and connectionist temporal classification (CTC) is used for ASR. The multitask framework implements VAD, SC, and ASR using an early layer, middle layer, and late layer of W2V2, which coincides with the order of segmenting the audio with VAD, clustering the segments based on speaker embeddings, and transcribing each segment with ASR. Experimental results on the augmented multi-party (AMI) dataset showed that using different W2V2 layers for VAD, SC, and ASR from the earlier to later layers for TMT not only saves computational cost, but also reduces diarisation error rates (DERs). Joint fine-tuning of VAD, SC, and ASR yielded 16%/17% relative reductions of DER with manual/automatic segmentation respectively, and consistent reductions in speaker attributed word error rate, compared to the baseline with separately fine-tuned models.

eess.AS↗
arXiv2022

Experimental implementation of Hardy-like quantum pigeonhole paradoxes

We present the general Hardy-like quantum pigeonhole paradoxes for \textit{n}-particle states, and find that each of such paradoxes can be simply associated to an un-colorable solution of a specific vertex-coloring problem induced from the projected-coloring graph (a kind of unconventional graph). Besides, as a special kind of Hardy's paradox, several kinds of Hardy-like quantum pigeonhole paradoxes can even give rise to higher success probability in demonstrating the conflict between quantum mechanics and local or noncontextual realism than the previous Hardy's paradoxes. Moreover, not only multi-qubit states, but high-dimensional states can exhibit the paradoxes. In contrast to only one type of contradiction presented in the original quantum pigeonhole paradox, two kinds of three-qubit projected-coloring graph states as the minimal illustration are discussed in our work, and an optical experiment to verify such stronger paradox is performed. This quantum paradox provides innovative thoughts and methods in exploring new types of stronger multi-party quantum nonlocality and may have potential applications in multi-party untrusted communications and device-independent random number generation.

quant-ph↗
arXiv2022

Learning Disentangled Representations for Controllable Human Motion Prediction

Generative model-based motion prediction techniques have recently realized predicting controlled human motions, such as predicting multiple upper human body motions with similar lower-body motions. However, to achieve this, the state-of-the-art methods require either subsequently learning mapping functions to seek similar motions or training the model repetitively to enable control over the desired portion of body. In this paper, we propose a novel framework to learn disentangled representations for controllable human motion prediction. Our network involves a conditional variational auto-encoder (CVAE) architecture to model full-body human motion, and an extra CVAE path to learn only the corresponding partial-body (e.g., lower-body) motion. Specifically, the inductive bias imposed by the extra CVAE path encourages two latent variables in two paths to respectively govern separate representations for each partial-body motion. With a single training, our model is able to provide two types of controls for the generated human motions: (i) strictly controlling one portion of human body and (ii) adaptively controlling the other portion, by sampling from a pair of latent spaces. Additionally, we extend and adapt a sampling strategy to our trained model to diversify the controllable predictions. Our framework also potentially allows new forms of control by flexibly customizing the input for the extra CVAE path. Extensive experimental results and ablation studies demonstrate that our approach is capable of predicting state-of-the-art controllable human motions both qualitatively and quantitatively.

cs.CV↗
arXiv2022

Tree-constrained Pointer Generator with Graph Neural Network Encodings for Contextual Speech Recognition

Incorporating biasing words obtained as contextual knowledge is critical for many automatic speech recognition (ASR) applications. This paper proposes the use of graph neural network (GNN) encodings in a tree-constrained pointer generator (TCPGen) component for end-to-end contextual ASR. By encoding the biasing words in the prefix-tree with a tree-based GNN, lookahead for future wordpieces in end-to-end ASR decoding is achieved at each tree node by incorporating information about all wordpieces on the tree branches rooted from it, which allows a more accurate prediction of the generation probability of the biasing words. Systems were evaluated on the Librispeech corpus using simulated biasing tasks, and on the AMI corpus by proposing a novel visual-grounded contextual ASR pipeline that extracts biasing words from slides alongside each meeting. Results showed that TCPGen with GNN encodings achieved about a further 15% relative WER reduction on the biasing words compared to the original TCPGen, with a negligible increase in the computation cost for decoding.

cs.SD↗
arXiv2022

Imprints of dark matter on gravitational ringing of supermassive black holes

Gravitational waves emitted from the gravitational ringing of supermassive black holes are important targets to test general relativity and probe the matter environment surrounding such black holes. The main components of the ringing waveform are black hole quasi-normal modes. In this paper, we study the effects of the dark matter halos with three different density profiles on the gravitational polar (even-parity) perturbations of a supermassive black hole. For this purpose, we first consider modified Schwarzschild spacetime with three different dark matter profiles and derive the equation of motion of the polar perturbations of the supermassive black hole. It is shown that by ignoring the dark matter perturbations, a Zerilli-like master equation with a modified potential for the polar perturbation can be obtained explicitly. Then we calculate the complex frequencies of the quasi-normal modes of the supermassive black hole in the dark matter halos. The corresponding gravitational wave spectra with the effects of the dark matter halos and their detectability have also been discussed.

gr-qc↗
arXiv2022

Adaptive Multi-view Rule Discovery for Weakly-Supervised Compatible Products Prediction

On e-commerce platforms, predicting if two products are compatible with each other is an important functionality to achieve trustworthy product recommendation and search experience for consumers. However, accurately predicting product compatibility is difficult due to the heterogeneous product data and the lack of manually curated training data. We study the problem of discovering effective labeling rules that can enable weakly-supervised product compatibility prediction. We develop AMRule, a multi-view rule discovery framework that can (1) adaptively and iteratively discover novel rulers that can complement the current weakly-supervised model to improve compatibility prediction; (2) discover interpretable rules from both structured attribute tables and unstructured product descriptions. AMRule adaptively discovers labeling rules from large-error instances via a boosting-style strategy, the high-quality rules can remedy the current model's weak spots and refine the model iteratively. For rule discovery from structured product attributes, we generate composable high-order rules from decision trees; and for rule discovery from unstructured product descriptions, we generate prompt-based rules from a pre-trained language model. Experiments on 4 real-world datasets show that AMRule outperforms the baselines by 5.98% on average and improves rule quality and rule proposal efficiency.

cs.LG↗
arXiv2022

Source localizations with the network of space-based gravitational wave detectors

The sky localization of the gravitational wave (GW) source is an important scientific objective for GW observations. A network of space-based GW detectors dramatically improves the sky localization accuracy compared with an individual detector not only in the inspiral stage but also in the ringdown stage. It is interesting to explore what plays an important role in the improvement. We find that the angle between the detector planes dominates the improvement, and the time delay is the next important factor. A detector network can dramatically improve the source localization for short signals and long signals with most contributions to the signal-to-noise ratio (SNR) coming from a small part of the signal in a short time, and the more SNR contributed by smaller parts, the better improvement by the network. We also find the effects of the arm length in the transfer function and higher harmonics are negligible for source localization with the detector network.

gr-qc↗
arXiv2022

Robust mixed finite element methods for a quad-curl singular perturbation problem

Robust mixed finite element methods are developed for a quad-curl singular perturbation problem. Lower order H(grad curl)-nonconforming but H(curl)-conforming finite elements are constructed, which are extended to nonconforming finite element Stokes complexes and the associated commutative diagrams. Then H(grad curl)-nonconforming finite elements are employed to discretize the quad-curl singular perturbation problem, which possess the sharp and uniform error estimates with respect to the perturbation parameter. The Nitsche's technique is exploited to achieve the optimal convergence rate in the case of the boundary layers. Numerical results are provided to verify the theoretical convergence rates. In addition, the regularity of the quad-curl singular perturbation problem is established.

math.NA↗