arXiv subjects

Hui Wang

Publications and source records attributed to Hui Wang.

At least 397 records · Page 22Linked to original sources

arXiv2021

Adaptive Warm-Start MCTS in AlphaZero-like Deep Reinforcement Learning

AlphaZero has achieved impressive performance in deep reinforcement learning by utilizing an architecture that combines search and training of a neural network in self-play. Many researchers are looking for ways to reproduce and improve results for other games/tasks. However, the architecture is designed to learn from scratch, tabula rasa, accepting a cold-start problem in self-play. Recently, a warm-start enhancement method for Monte Carlo Tree Search was proposed to improve the self-play starting phase. It employs a fixed parameter $I^\prime$ to control the warm-start length. Improved performance was reported in small board games. In this paper we present results with an adaptive switch method. Experiments show that our approach works better than the fixed $I^\prime$, especially for "deep," tactical, games (Othello and Connect Four). We conjecture that the adaptive value for $I^\prime$ is also influenced by the size of the game, and that on average $I^\prime$ will increase with game size. We conclude that AlphaZero-like deep reinforcement learning benefits from adaptive rollout based warm-start, as Rapid Action Value Estimate did for rollout-based reinforcement learning 15 years ago.

cs.AI↗

arXiv2021

PanGu-$α$: Large-scale Autoregressive Pretrained Chinese Language Models with Auto-parallel Computation

Large-scale Pretrained Language Models (PLMs) have become the new paradigm for Natural Language Processing (NLP). PLMs with hundreds of billions parameters such as GPT-3 have demonstrated strong performances on natural language understanding and generation with \textit{few-shot in-context} learning. In this work, we present our practice on training large-scale autoregressive language models named PanGu-$α$, with up to 200 billion parameters. PanGu-$α$ is developed under the MindSpore and trained on a cluster of 2048 Ascend 910 AI processors. The training parallelism strategy is implemented based on MindSpore Auto-parallel, which composes five parallelism dimensions to scale the training task to 2048 processors efficiently, including data parallelism, op-level model parallelism, pipeline model parallelism, optimizer model parallelism and rematerialization. To enhance the generalization ability of PanGu-$α$, we collect 1.1TB high-quality Chinese data from a wide range of domains to pretrain the model. We empirically test the generation ability of PanGu-$α$ in various scenarios including text summarization, question answering, dialogue generation, etc. Moreover, we investigate the effect of model scales on the few-shot performances across a broad range of Chinese NLP tasks. The experimental results demonstrate the superior capabilities of PanGu-$α$ in performing various tasks under few-shot or zero-shot settings.

cs.CL↗

arXiv2021

An optics-free computational spectrometer using a broadband and tunable dynamic detector

Optical spectrometers are the central instruments for exploring the interaction between light and matter. The current pursuit of the field is to design a spectrometer without the need for wavelength multiplexing optics to effectively reduce the complexity and physical size of the hardware. Based on computational spectroscopic results and combining a broadband-responsive dynamic detector, we successfully demonstrate an optics-free single-detector spectrometer that maps the tunable quantum efficiency of a superconducting nanowire into an ill-conditioned matrix to build a solvable inverse mathematical equation. Such a spectrometer can realize a broadband spectral responsivity ranging from 660 to 1900 nm. The spectral resolution at the telecom is 6 nm, exceeding the energy resolving capacity of existing infrared single-photon detectors. Meanwhile, benefiting from the optics-free setup, precise time-of-flight measurements can be simultaneously achieved. We have demonstrated a spectral LiDAR with 8 spectral channels. This work provides a concise method for building multifunctional spectrometers and paves the way for applying superconducting nanowire detectors in spectroscopy.

physics.optics↗

arXiv2021

Large-Scale Approximate k-NN Graph Construction on GPU

k-nearest neighbor graph is a key data structure in many disciplines such as manifold learning, machine learning and information retrieval, etc. NN-Descent was proposed as an effective solution for the graph construction problem. However, it cannot be directly transplanted to GPU due to the intensive memory accesses required in the approach. In this paper, NN-Descent has been redesigned to adapt to the GPU architecture. In particular, the number of memory accesses has been reduced significantly. The redesign fully exploits the parallelism of the GPU hardware. In the meantime, the genericness as well as the simplicity of NN-Descent are well-preserved. In addition, a simple but effective k-NN graph merge approach is presented. It allows two graphs to be merged efficiently on GPUs. More importantly, it makes the construction of high-quality k-NN graphs for out-of-GPU-memory datasets tractable. The results show that our approach is 100-250x faster than single-thread NN-Descent and is 2.5-5x faster than existing GPU-based approaches.

cs.DC↗

arXiv2021

Unsupervised Domain Adaptation for Image Classification via Structure-Conditioned Adversarial Learning

Unsupervised domain adaptation (UDA) typically carries out knowledge transfer from a label-rich source domain to an unlabeled target domain by adversarial learning. In principle, existing UDA approaches mainly focus on the global distribution alignment between domains while ignoring the intrinsic local distribution properties. Motivated by this observation, we propose an end-to-end structure-conditioned adversarial learning scheme (SCAL) that is able to preserve the intra-class compactness during domain distribution alignment. By using local structures as structure-aware conditions, the proposed scheme is implemented in a structure-conditioned adversarial learning pipeline. The above learning procedure is iteratively performed by alternating between local structures establishment and structure-conditioned adversarial learning. Experimental results demonstrate the effectiveness of the proposed scheme in UDA scenarios.

cs.CV↗

arXiv2021

Stable Online Computation Offloading via Lyapunov-guided Deep Reinforcement Learning

In this paper, we consider a multi-user mobile-edge computing (MEC) network with time-varying wireless channels and stochastic user task data arrivals in sequential time frames. In particular, we aim to design an online computation offloading algorithm to maximize the network data processing capability subject to the long-term data queue stability and average power constraints. The online algorithm is practical in the sense that the decisions for each time frame are made without the assumption of knowing future channel conditions and data arrivals. We formulate the problem as a multi-stage stochastic mixed integer non-linear programming (MINLP) problem that jointly determines the binary offloading (each user computes the task either locally or at the edge server) and system resource allocation decisions in sequential time frames. To address the coupling in the decisions of different time frames, we propose a novel framework, named LyDROO, that combines the advantages of Lyapunov optimization and deep reinforcement learning (DRL). Specifically, LyDROO first applies Lyapunov optimization to decouple the multi-stage stochastic MINLP into deterministic per-frame MINLP subproblems of much smaller size. Then, it integrates model-based optimization and model-free DRL to solve the per-frame MINLP problems with very low computational complexity. Simulation results show that the proposed LyDROO achieves optimal computation performance while satisfying all the long-term constraints. Besides, it induces very low execution latency that is particularly suitable for real-time implementation in fast fading environments.

cs.NI↗

arXiv2021

Gain without Pain: Offsetting DP-injected Nosies Stealthily in Cross-device Federated Learning

Federated Learning (FL) is an emerging paradigm through which decentralized devices can collaboratively train a common model. However, a serious concern is the leakage of privacy from exchanged gradient information between clients and the parameter server (PS) in FL. To protect gradient information, clients can adopt differential privacy (DP) to add additional noises and distort original gradients before they are uploaded to the PS. Nevertheless, the model accuracy will be significantly impaired by DP noises, making DP impracticable in real systems. In this work, we propose a novel Noise Information Secretly Sharing (NISS) algorithm to alleviate the disturbance of DP noises by sharing negated noises among clients. We theoretically prove that: 1) If clients are trustworthy, DP noises can be perfectly offset on the PS; 2) Clients can easily distort negated DP noises to protect themselves in case that other clients are not totally trustworthy, though the cost lowers model accuracy. NISS is particularly applicable for FL across multiple IoT (Internet of Things) systems, in which all IoT devices need to collaboratively train a model. To verify the effectiveness and the superiority of the NISS algorithm, we conduct experiments with the MNIST and CIFAR-10 datasets. The experiment results verify our analysis and demonstrate that NISS can improve model accuracy by 21% on average and obtain better privacy protection if clients are trustworthy.

cs.DC↗

arXiv2021

Heralded non-destructive quantum entangling gate with single-photon sources

Heralded entangling quantum gates are an essential element for the implementation of large-scale optical quantum computation. Yet, the experimental demonstration of genuine heralded entangling gates with free-flying output photons in linear optical system, was hindered by the intrinsically probabilistic source and double-pair emission in parametric down-conversion. Here, by using an on-demand single-photon source based on a semiconductor quantum dot embedded in a micro-pillar cavity, we demonstrate a heralded controlled-NOT (CNOT) operation between two single photons for the first time. To characterize the performance of the CNOT gate, we estimate its average quantum gate fidelity of ($87.8\pm1.2$)%. As an application, we generated event-ready Bell states with a fidelity of ($83.4\pm2.4$)%. Our results are an important step towards the development of photon-photon quantum logic gates.

quant-ph↗

arXiv2020

Joint Beamforming and Power Control for Throughput Maximization in IRS-assisted MISO WPCNs

Intelligent reflecting surface (IRS) is an emerging technology to enhance the energy- and spectrum-efficiency of wireless powered communication networks (WPCNs). In this paper, we investigate an IRS-assisted multiuser multiple-input single-output (MISO) WPCN, where the single-antenna wireless devices (WDs) harvest wireless energy in the downlink (DL) and transmit their information simultaneously in the uplink (UL) to a common hybrid access point (HAP) equipped with multiple antennas. Our goal is to maximize the weighted sum rate (WSR) of all the energy-harvesting users. To make full use of the beamforming gain provided by both the HAP and the IRS, we jointly optimize the active beamforming of the HAP and the reflecting coefficients (passive beamforming) of the IRS in both DL and UL transmissions, as well as the transmit power of the WDs to mitigate the inter-user interference at the HAP. To tackle the challenging optimization problem, we first consider fixing the passive beamforming, and converting the remaining joint active beamforming and user transmit power control problem into an equivalent weighted minimum mean square error (WMMSE) problem, where we solve it using an efficient block-coordinate descent (BCD) method. Then, we fix the active beamforming and user transmit power, and optimize the passive beamforming coefficients of the IRS in both the DL and UL using a semidefinite relaxation (SDR) method. Accordingly, we apply a block-structured optimization (BSO) method to update the two sets of variables alternately. Numerical results show that the proposed joint optimization achieves significant performance gain over other representative benchmark methods and effectively improves the throughput performance in multiuser MISO WPCNs.

cs.NI↗

arXiv2020

Quantum computational advantage using photons

Gaussian boson sampling exploits squeezed states to provide a highly efficient way to demonstrate quantum computational advantage. We perform experiments with 50 input single-mode squeezed states with high indistinguishability and squeezing parameters, which are fed into a 100-mode ultralow-loss interferometer with full connectivity and random transformation, and sampled using 100 high-efficiency single-photon detectors. The whole optical set-up is phase-locked to maintain a high coherence between the superposition of all photon number states. We observe up to 76 output photon-clicks, which yield an output state space dimension of $10^{30}$ and a sampling rate that is $10^{14}$ faster than using the state-of-the-art simulation strategy and supercomputers. The obtained samples are validated against various hypotheses including using thermal states, distinguishable photons, and uniform distribution.

quant-ph↗

arXiv2020

The dissipation of toroidal magnetic fields and spin-down evolution of young and strongly magnetized pulsars

Magnetars are a kind of pulsars powered mainly by superhigh magnetic fields. They are popular sources with many unsolved issues in themselves, but also linked to various high energy phenomena, such as QPOs, giant flares, fast radio bursts and super-luminous supernovae. In this work, we first review our recent works on the dissipation of toroidal magnetic fields in magnetars and rotationally powered pulsars, then review the spin-down evolution of young and strongly magnetized pulsars, especially of magnetars. We present an interesting and important relation between the magnetization parameter, and magnetic field in the magnetar crust. Finally, we introduce our two works in progress: to explain the magnetar "anti-gltich" events in the thermal plastic flow model and to revisit the expression of braking index $n$, which is independent of the second derivative of spin frequency of a pulsar and give some proposals for our future work.

astro-ph.HE↗

arXiv2020

Micro-cracking, microstructure and mechanical properties of Hastelloy-X alloy printed by laser powder bed fusion: as-built, annealed and hot-isostatic pressed

This study analyses literature data to identify optimised print parameters and assesses the consolidation, microstructure, and mechanical properties of Hastelloy-X printed by laser powder bed fusion. Effects of post annealing and hot-isostatic pressing (HIP) on the microstructure and mechanical properties are also revealed. The susceptibility to the solidification cracking and the as-built microstructure such as precipitation and chemical segregation were predicted by the calculation of thermodynamics phase diagrams. The distribution of solidification cracks throughout the builds was quantified for the as-built, annealed and HIP conditions. The assessment reveals the variation of crack density towards the bottom, top and free surface of solid builds. This distribution of cracks is found to be associate with the thermal gradient and thermal conductivity which were estimated by analytical thermal calculations. While the annealing and HIP both can alter the as-printed microstructure thanks to recovery and recrystallisation, the micro-cracks and pores were only successfully removed by the HIP. In addition to the removal, recrystallisation and precipitation in the HIP (stronger than in annealing), resulting in optimal mechanical properties including a substantial increase in elongation from 13% to 20%, significant improvement of ultimate tensile stress from 965 MPa to 1045 MPa with moderately high yield stress thanks to precipitation.

cond-mat.mtrl-sci↗

arXiv2020

Approximate k-NN Graph Construction: a Generic Online Approach

Nearest neighbor search and k-nearest neighbor graph construction are two fundamental issues arise from many disciplines such as multimedia information retrieval, data-mining and machine learning. They become more and more imminent given the big data emerge in various fields in recent years. In this paper, a simple but effective solution both for approximate k-nearest neighbor search and approximate k-nearest neighbor graph construction is presented. These two issues are addressed jointly in our solution. On the one hand, the approximate k-nearest neighbor graph construction is treated as a search task. Each sample along with its k-nearest neighbors are joined into the k-nearest neighbor graph by performing the nearest neighbor search sequentially on the graph under construction. On the other hand, the built k-nearest neighbor graph is used to support k-nearest neighbor search. Since the graph is built online, the dynamic update on the graph, which is not possible from most of the existing solutions, is supported. This solution is feasible for various distance measures. Its effectiveness both as k-nearest neighbor construction and k-nearest neighbor search approaches is verified across different types of data in different scales, various dimensions and under different metrics.

cs.IR↗

arXiv2020

Observation of intensity squeezing in resonance fluorescence from a solid-state device

Intensity squeezing, i.e., photon number fluctuations below the shot noise limit, is a fundamental aspect of quantum optics and has wide applications in quantum metrology. It was predicted in 1979 that the intensity squeezing could be observed in resonance fluorescence from a two-level quantum system. Yet, its experimental observation in solid states was hindered by inefficiencies in generating, collecting and detecting resonance fluorescence. Here, we report the intensity squeezing in a single-mode fibre-coupled resonance fluorescence single-photon source based on a quantum dot-micropillar system. We detect pulsed single-photon streams with 22.6% system efficiency, which show subshot-noise intensity fluctuation with an intensity squeezing of 0.59 dB. We estimate a corrected squeezing of 3.29 dB at the first lens. The observed intensity squeezing provides the last piece of the fundamental picture of resonance fluorescence; which can be used as a new standard for optical radiation and in scalable quantum metrology with indistinguishable single photons.

quant-ph↗

arXiv2020

Sample-efficient benchmarking of multi-photon interference on a boson sampler in the sparse regime

Verification of a quantum advantage in the presence of noise is a key open problem in the study of near-term quantum devices. In this work, we show how to assess the quality of photonic interference in a linear optical quantum device (boson sampler) by using a maximum likelihood method to measure the strength at which various noise sources are present in the experiment. This allows us to use a sparse set of samples to test whether a given boson sampling experiment meets known upper bounds on the level of noise permissible to demonstrate a quantum advantage. Furthermore, this method allows us monitor the evolution of noise in real time, creating a valuable diagnostic tool. Finally, we observe that sources of noise in the experiment compound, meaning that the observed value of the mutual photon indistinguishability, which is the main imperfection in our study, is an effective value taking into account all sources of error in the experiment.

quant-ph↗

arXiv2020

Leveraging Historical Interaction Data for Improving Conversational Recommender System

Recently, conversational recommender system (CRS) has become an emerging and practical research topic. Most of the existing CRS methods focus on learning effective preference representations for users from conversation data alone. While, we take a new perspective to leverage historical interaction data for improving CRS. For this purpose, we propose a novel pre-training approach to integrating both item-based preference sequence (from historical interaction data) and attribute-based preference sequence (from conversation data) via pre-training methods. We carefully design two pre-training tasks to enhance information fusion between item- and attribute-based preference. To improve the learning performance, we further develop an effective negative sample generator which can produce high-quality negative samples. Experiment results on two real-world datasets have demonstrated the effectiveness of our approach for improving CRS.

cs.IR↗

arXiv2020

S^3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization

Recently, significant progress has been made in sequential recommendation with deep learning. Existing neural sequential recommendation models usually rely on the item prediction loss to learn model parameters or data representations. However, the model trained with this loss is prone to suffer from data sparsity problem. Since it overemphasizes the final performance, the association or fusion between context data and sequence data has not been well captured and utilized for sequential recommendation. To tackle this problem, we propose the model S^3-Rec, which stands for Self-Supervised learning for Sequential Recommendation, based on the self-attentive neural architecture. The main idea of our approach is to utilize the intrinsic data correlation to derive self-supervision signals and enhance the data representations via pre-training methods for improving sequential recommendation. For our task, we devise four auxiliary self-supervised objectives to learn the correlations among attribute, item, subsequence, and sequence by utilizing the mutual information maximization (MIM) principle. MIM provides a unified way to characterize the correlation between different types of data, which is particularly suitable in our scenario. Extensive experiments conducted on six real-world datasets demonstrate the superiority of our proposed method over existing state-of-the-art methods, especially when only limited training data is available. Besides, we extend our self-supervised learning method to other recommendation models, which also improve their performance.

cs.IR↗

arXiv2020

Throughput Optimization of Intelligent Reflecting Surface Assisted User Cooperation in WPCNs

Intelligent reflecting surface (IRS) can effectively enhance the energy and spectral efficiency of wireless communication system through the use of a large number of lowcost passive reflecting elements. In this paper, we investigate throughput optimization of IRS-assisted user cooperation in a wireless powered communication network (WPCN), where the two WDs harvest wireless energy and transmit information to a common hybrid access point (HAP). In particular, the two WDs first exchange their independent information with each other and then form a virtual antenna array to transmit jointly to the HAP. We aim to maximize the common (minimum) throughput performance by jointly optimizing the transmit time and power allocations of the two WDs on wireless energy and information transmissions and the passive array coefficients on reflecting the wireless energy and information signals. By comparing with some existing benchmark schemes, our results show that the proposed IRS-assisted user cooperation method can effectively improve the throughput performance of cooperative transmission in WPCNs.

cs.NI↗