arXiv subjects

Yuan Cheng

Publications and source records attributed to Yuan Cheng.

126 records · Page 7Linked to original sources

arXiv2021

CRSLab: An Open-Source Toolkit for Building Conversational Recommender System

In recent years, conversational recommender system (CRS) has received much attention in the research community. However, existing studies on CRS vary in scenarios, goals and techniques, lacking unified, standardized implementation or comparison. To tackle this challenge, we propose an open-source CRS toolkit CRSLab, which provides a unified and extensible framework with highly-decoupled modules to develop CRSs. Based on this framework, we collect 6 commonly-used human-annotated CRS datasets and implement 18 models that include recent techniques such as graph neural network and pre-training models. Besides, our toolkit provides a series of automatic evaluation protocols and a human-machine interaction interface to test and compare different CRS methods. The project and documents are released at https://github.com/RUCAIBox/CRSLab.

cs.CL↗

arXiv2020

S3-Net: A Fast and Lightweight Video Scene Understanding Network by Single-shot Segmentation

Real-time understanding in video is crucial in various AI applications such as autonomous driving. This work presents a fast single-shot segmentation strategy for video scene understanding. The proposed net, called S3-Net, quickly locates and segments target sub-scenes, meanwhile extracts structured time-series semantic features as inputs to an LSTM-based spatio-temporal model. Utilizing tensorization and quantization techniques, S3-Net is intended to be lightweight for edge computing. Experiments using CityScapes, UCF11, HMDB51 and MOMENTS datasets demonstrate that the proposed S3-Net achieves an accuracy improvement of 8.1% versus the 3D-CNN based approach on UCF11, a storage reduction of 6.9x and an inference speed of 22.8 FPS on CityScapes with a GTX1080Ti GPU.

cs.CV↗

arXiv2020

Magnon-magnon interaction and magnon relaxation time in ferromagnetic Cr2Ge2Te6 monolayer

Despite the intense amount of attention and huge potential of two-dimensional (2D) magnets for applications in novel magnetic, magneto-optical, magneto-thermal and magneto-electronic devices, there has yet to be a robust strategy developed to systematically understand magnon-magnon (MMI) interactions at finite temperature. In this paper, we present a first-principles theoretical method to introduce the finite temperature magnon-magnon interaction into Heisenberg Hamiltonian through a nonlinear correction energy. The Wick theorem is used to decouple the four-magnon operators to two-magnon order. We demonstrate the capabilities of this method by studying the strength of MMI in Cr2Ge2Te6 (CGT) monolayer. The spin wave spectrum at finite temperature and the time-dependent spin autocorrelation function are explored. It is found that the magnon relaxation time due to magnon-magnon scattering increases with temperature because of the reduction in magnon energy, while decreases with wavevector and external magnetic field. Our results provide a new insight to understand the magnon damping and energy dissipation in two-dimensional ferromagnetic materials.

cond-mat.mes-hall↗

arXiv2020

Material Platforms for Defect Qubits and Single Photon Emitters

Quantum technology has grown out of quantum information theory and now provides a valuable tool that researchers from numerous fields can add to their toolbox of research methods. To date, various systems have been exploited to promote the application of quantum information processing. The systems that can be used for quantum technology include superconducting circuits, ultra-cold atoms, trapped ions, semiconductor quantum dots, and solid-state spins and emitters. In this review, we will discuss the state of the art on material platforms for spin-based quantum technology, with a focus on the progress in solid-state spins and emitters in several leading host materials, including diamond, silicon carbide, boron nitride, silicon, two-dimensional semiconductors, and other materials. We will highlight how first-principles calculations can serve as an exceptionally robust tool for finding the novel defect qubits and single photon emitters in solids, through detailed predictions of the electronic, magnetic and optical properties.

cond-mat.mtrl-sci↗

arXiv2020

HOTCAKE: Higher Order Tucker Articulated Kernels for Deeper CNN Compression

The emerging edge computing has promoted immense interests in compacting a neural network without sacrificing much accuracy. In this regard, low-rank tensor decomposition constitutes a powerful tool to compress convolutional neural networks (CNNs) by decomposing the 4-way kernel tensor into multi-stage smaller ones. Building on top of Tucker-2 decomposition, we propose a generalized Higher Order Tucker Articulated Kernels (HOTCAKE) scheme comprising four steps: input channel decomposition, guided Tucker rank selection, higher order Tucker decomposition and fine-tuning. By subjecting each CONV layer to HOTCAKE, a highly compressed CNN model with graceful accuracy trade-off is obtained. Experiments show HOTCAKE can compress even pre-compressed models and produce state-of-the-art lightweight networks.

cs.LG↗

arXiv2019

Early Predictions for Medical Crowdfunding: A Deep Learning Approach Using Diverse Inputs

Medical crowdfunding is a popular channel for people needing financial help paying medical bills to collect donations from large numbers of people. However, large heterogeneity exists in donations across cases, and fundraisers face significant uncertainty in whether their crowdfunding campaigns can meet fundraising goals. Therefore, it is important to provide early warnings for fundraisers if such a channel will eventually fail. In this study, we aim to develop novel algorithms to provide accurate and timely predictions of fundraising performance, to better inform fundraisers. In particular, we propose a new approach to combine time-series features and time-invariant features in the deep learning model, to process diverse sources of input data. Compared with baseline models, our model achieves better accuracy and requires a shorter observation window of the time-varying features from the campaign launch to provide robust predictions with high confidence. To extract interpretable insights, we further conduct a multivariate time-series clustering analysis and identify four typical temporal donation patterns. This demonstrates the heterogeneity in the features and how they relate to the fundraising outcome. The prediction model and the interpretable insights can be applied to assist fundraisers with better promoting their fundraising campaigns and can potentially help crowdfunding platforms to provide more timely feedback to all fundraisers. Our proposed framework is also generalizable to other fields where diverse structured and unstructured data are valuable for predictions.

cs.LG↗

arXiv2019

Career Choice as an Extended Spatial Evolutionary Public Goods Game

We propose an extended spatial evolutionary public goods game (SEPGG) model to study the dynamics of individual career choice and the corresponding social output. Based on the social value orientation theory, we categorized two classes of work, namely the public work if it serves public interests, and the private work if it serves personal interests. Under the context of SEPGG, choosing public work is to cooperate and choosing private work is to defect. We then investigate the effects of employee productivity, human capital and external subsidies on individual career choices of the two work types, as well as the overall social welfare. From simulation results, we found that when employee productivity of public work is low, people are more willing to enter the private sector. Although this will make both the effort level and human capital of individuals doing private work higher than those engaging in public work, the total outcome of the private sector is still lower than that of the public sector provided a low level of public subsidies. When the employee productivity is higher for public work, a certain amount of subsidy can greatly improve system output. On the contrary, when the employee productivity of public work is low, provisions of subsidy to the public sector can result in a decline in social output.

physics.soc-ph↗

arXiv2019

Mathematical aspect of the combinatorial game "Mahjong"

We illustrate how one can use basic combinatorial theory and computer programming technique (Python) to analyze the combinatorial game: Mahjong. The results confirm some folklore concerning the game, and expose some unexpected results. Related results and possible future research in connection to artificial intelligence are mentioned. Readers interested in the subject may further develop the techniques to deepen the study of the game, or study other combinatorial games.

math.HO↗

arXiv2018

DEEPEYE: A Compact and Accurate Video Comprehension at Terminal Devices Compressed with Quantization and Tensorization

As it requires a huge number of parameters when exposed to high dimensional inputs in video detection and classification, there is a grand challenge to develop a compact yet accurate video comprehension at terminal devices. Current works focus on optimizations of video detection and classification in a separated fashion. In this paper, we introduce a video comprehension (object detection and action recognition) system for terminal devices, namely DEEPEYE. Based on You Only Look Once (YOLO), we have developed an 8-bit quantization method when training YOLO; and also developed a tensorized-compression method of Recurrent Neural Network (RNN) composed of features extracted from YOLO. The developed quantization and tensorization can significantly compress the original network model yet with maintained accuracy. Using the challenging video datasets: MOMENTS and UCF11 as benchmarks, the results show that the proposed DEEPEYE achieves 3.994x model compression rate with only 0.47% mAP decreased; and 15,047x parameter reduction and 2.87x speed-up with 16.58% accuracy improvement.

cs.CV↗

arXiv2018

Testing the Universality of Free Fall at ${10^{ - 10}}$ level by Comparing the Atoms in Different Hyperfine States with Bragg Diffraction

We have performed a precision atomic interferometry experiment on testing the universality of free fall (UFF) considering atoms' spin degree of freedom. Our experiment employs the Bragg atom interferometer with $^{87}$Rb atoms either in hyperfine state $\left| {F = 1,{m_F} = 0} \right\rangle $ or $\left| {F = 2,{m_F} = 0} \right\rangle $, and the wave packets in these two states are diffracted in one pair of Bragg beams alternatively, which can help suppress the common-mode systematic errors. We have obtained an E$\rm{\ddot{o}}$tv$\rm{\ddot{o}}$s ratio $η= \left( { 0.9 \pm 2.7} \right) \times {10^{ - 10}}$, and set a new record on the precision with a nearly 5 times improvement. Our experiment gives stronger restrictions on the possible UFF breaking mechanism.

quant-ph↗

arXiv2018

Unravelling the origin of piezo/ferro-electric properties of metal-organic frameworks (MOFs) nanocrystals

Metal-organic framework (MOF) UiO-66 nanocrystals were previously believed to be piezo/ferro-electrically inactive because of their centrosymmetric lattice symmetries (Fm-3m (225)) revealed by Powder X-ray diffraction. However, via delicate dual AC resonance tracking piezoresponse force microscopy and piezoresponse force spectroscopy characterizations, our nanoscale probing for the first time demonstrate that UiO-66 nanocrystals show piezo/ferro-electric response. Our compelling experimental and theoretically analyses disclose that the structure of UiO-66 should not be the highly centrosymmetric Fm-3m (225) but a reduced symmetry form instead. UiO-66(Hf)-type MOFs possess stronger piezoresponse and better ferroelectric switching behaviours than their counterparts UiO-66 (Zr)-type MOFs. Our study not only enriches the structural understanding of UiO-66 MOF, but also suggests possible modification of electronic property of the MOFs by judicious selection of metal ions and functional ligands.

cond-mat.mtrl-sci↗

arXiv2017

Event-Radar: Real-time Local Event Detection System for Geo-Tagged Tweet Streams

The local event detection is to use posting messages with geotags on social networks to reveal the related ongoing events and their locations. Recent studies have demonstrated that the geo-tagged tweet stream serves as an unprecedentedly valuable source for local event detection. Nevertheless, how to effectively extract local events from large geo-tagged tweet streams in real time remains challenging. A robust and efficient cloud-based real-time local event detection software system would benefit various aspects in the real-life society, from shopping recommendation for customer service providers to disaster alarming for emergency departments. We use the preliminary research GeoBurst as a starting point, which proposed a novel method to detect local events. GeoBurst+ leverages a novel cross-modal authority measure to identify several pivots in the query window. Such pivots reveal different geo-topical activities and naturally attract related tweets to form candidate events. It further summarises the continuous stream and compares the candidates against the historical summaries to pinpoint truly interesting local events. We mainly implement a website demonstration system Event-Radar with an improved algorithm to show the real-time local events online for public interests. Better still, as the query window shifts, our method can update the event list with little time cost, thus achieving continuous monitoring of the stream.

cs.IR↗

arXiv2016

Equity Market Impact Modeling: an Empirical Analysis for Chinese Market

Market impact has become a subject of increasing concern among academics and industry experts. We put forward a price impact model which considers the heteroscedasticity of price in the time dimension and dependency between permanent impact and temporary impact. We discuss and derive the extremum of the expectation of permanent impact and realized impact by constructing several special trading trajectories. Given our use of a large trade and quote tick records of 17,213,238,343 compiled from the Chinese stock market, the model assessment ultimately suggest that our model is better than Almgren's model. Interestingly, the result of random effect analysis indicates the parameter $α$, which is the exponent of the impact function, is a constant with a value of around 0.7 across all stocks. Our model and empirical result would give academia some insight of mechanism of Chinese market, and can be applied to algorithm trading.

q-fin.TR↗

arXiv2016

Theoretical Simulation of 87Rb Absorption Spectrum in a Thermal Cell

In this paper, we present a theoretical simulation of 87Rb absorption spectrum in a thermal cm-cell which is adaptive to the experimental observation. In experiment, the coupling and probe beams are configured to copropagate but perpendicular polarized, making up to five velocity selective optical pumping (VSOP) absorption dips able to be identified. A $Λ$-type electromagnetically induced transparency (EIT) is also observed for each group of velocity-selected atoms. The spectrum by only sweeping the probe beam can be decomposed into a combination of Doppler-broadened background and three VSOP dips for each group of velocity-selected atoms, companied by an EIT peak. This proposed theoretical model can be used to simulate the spectrum adaptive to the experimental observation by non-linear least-square fit method. The fit for high quality of experimental observation can determine valuable transition parameters such as decaying rates and coupling beam power accurately.

physics.atom-ph↗

arXiv2016

From Brittle to Ductile: A Structure Dependent Ductility of Diamond Nanothread

As a potential building block for the next generation of devices or multifunctional materials that are spreading almost every technology sector, one-dimensional (1D) carbon nanomaterial has received intensive research interests. Recently, a new ultra-thin diamond nanothread (DNT) has joined this palette, which is a 1D structure with poly-benzene sections connected by Stone-Wales (SW) transformation defects. Using large-scale molecular dynamics simulations, we found that this sp3 bonded DNT can transit from a brittle to a ductile characteristic by varying the length of the poly-benzene sections, suggesting that DNT possesses entirely different mechanical responses than other 1D carbon allotropies. Analogously, the SW defects behave like a grain boundary that interrupts the consistency of the poly-benzene sections. For a DNT with a fixed length, the yield strength fluctuates in the vicinity of a certain value and is independent of the "grain size". On the other hand, both yield strength and yield strain show a clear dependence on the total length of DNT, which is due to the fact that the failure of the DNT is dominated by the SW defects. Its highly tunable ductility together with its ultra-light density and high Young's modulus makes diamond nanothread ideal for creation of extremely strong three-dimensional nano-architectures.

cond-mat.mtrl-sci↗

arXiv2016

Diamond nanothread as a new reinforcement for nanocomposites

This work explores the application of a new one-dimensional carbon nanomaterial, the diamond nanothread (DNT), as a reinforcement for nanocomposites. Owing to the existence of Stone-Wales transformation defects, the DNT intrinsically possesses irregular surfaces, which is expected to enhance the non-covalent interfacial load transfer. Through a series of in silico pull-out studies of the DNT in polyethylene (PE) matrix, we found that the load transfer between DNT and PE matrix is dominated by the non-covalent interactions, in particular the van der Waals interactions. Although the hydrogenated surface of the DNT reduces the strength of the van der Waals interactions at the interface, the irregular surface of the DNT can compensate for the weak bonds. These factors lead to an interfacial shear strength of the DNT/PE interface comparable with that of the carbon nanotube (CNT)/PE interface. Our results show that the DNT/PE interfacial shear strength remains high even as the number of Stone-Wales transformation defects decreases. It can be enhanced further by increasing the PE density or introduction of functional groups to the DNT, both of which greatly increase the non-covalent interactions.

cond-mat.mtrl-sci↗

arXiv2014

A McKean optimal transportation perspective on Feynman-Kac formulae with application to data assimilation

Data assimilation is the task of combining mathematical models with observational data. From a mathematical perspective data assimilation leads to Bayesian inference problems which can be formulated in terms of Feynman-Kac formulae. In this paper we focus on the sequential nature of many data assimilation problems and their numerical implementation in form of Monte Carlo methods. We demonstrate how sequential data assimilation can be interpreted as time-dependent Markov processes, which is often referred to as the McKean approach to Feynman-Kac formulae. It is shown that the McKean approach has very natural links to coupling of random variables and optimal transportation. This link allows one to propose novel sequential Monte Carlo methods/particle filters. In combination with localization these novel algorithms have the potential of beating the curse of dimensionality, which has prevented particle filters from being applied to spatially extended systems.

math.DS↗

arXiv2012

A Multivariate Graphical Stochastic Volatility Model

The Gaussian Graphical Model (GGM) is a popular tool for incorporating sparsity into joint multivariate distributions. The G-Wishart distribution, a conjugate prior for precision matrices satisfying general GGM constraints, has now been in existence for over a decade. However, due to the lack of a direct sampler, its use has been limited in hierarchical Bayesian contexts, relegating mixing over the class of GGMs mostly to situations involving standard Gaussian likelihoods. Recent work, however, has developed methods that couple model and parameter moves, first through reversible jump methods and later by direct evaluation of conditional Bayes factors and subsequent resampling. Further, methods for avoiding prior normalizing constant calculations--a serious bottleneck and source of numerical instability--have been proposed. We review and clarify these developments and then propose a new methodology for GGM comparison that blends many recent themes. Theoretical developments and computational timing experiments reveal an algorithm that has limited computational demands and dramatically improves on computing times of existing methods. We conclude by developing a parsimonious multivariate stochastic volatility model that embeds GGM uncertainty in a larger hierarchical framework. The method is shown to be capable of adapting to the extreme swings in market volatility experienced in 2008 after the collapse of Lehman Brothers, offering considerable improvement in posterior predictive distribution calibration.

stat.CO↗