arXiv subjects

Shuai Li

Publications and source records attributed to Shuai Li.

At least 379 records · Page 21Linked to original sources

arXiv2019

Online Learning to Rank with Features

We introduce a new model for online ranking in which the click probability factors into an examination and attractiveness function and the attractiveness function is a linear function of a feature vector and an unknown parameter. Only relatively mild assumptions are made on the examination function. A novel algorithm for this setup is analysed, showing that the dependence on the number of items is replaced by a dependence on the dimension, allowing the new algorithm to handle a large number of items. When reduced to the orthogonal case, the regret of the algorithm improves on the state-of-the-art.

stat.ML↗

arXiv2019

Topological flat bands in twisted trilayer graphene

Twisted trilayer graphene (TLG) may be the simplest realistic system so far, which has flat bands with nontrivial topology. Here, we give a detailed calculation about its band structures and the band topology, i.e. valley Chern number of the nearly flat bands, with the continuum model. With realistic parameters, the magic angle of twisted TLG is about 1.12$^{\circ}$, at which two nearly flat bands appears. Unlike the twisted bilayer graphene, a small twist angle can induce a tiny gap at all the Dirac points, which can be enlarged further by a perpendicular electric field. The valley Chern numbers of the two nearly flat bands in the twisted TLG depends on the twist angle $θ$ and the perpendicular electric field $E_\perp$. Considering its topological flat bands, the twisted TLG should be an ideal experimental platform to study the strongly correlated physics in topologically nontrivial flat band systems. And, due to its reduced symmetry, the correlated states in twisted TLG should be quite different from that in twisted bilayer graphene and twisted double bilayer graphene.

cond-mat.mes-hall↗

arXiv2019

Dual-Color Laser Induced Terahertz Generation in Strong Field Approximation

The mechanism of the terahertz (THz) wave generation (TWG) in dual-color fields is elucidated within the theoretical framework of single-atom based strong field approximation (SFA). Evaluating the transition dipole moment, the continuum-continuum (CC) transition, rather than the continuum-bound recombination for the high-order harmonic generation, is confirmed to be the core mechanism of the TWG. The analytic form of the SFA-based CC description is consistent with the classical photoelectric current model, establishing the quantum-classical correspondence for the TWG. The theory is supported by parametric dependence of experimental THz yields calibrated by the joint measurement of the third-order harmonics. Present studies leave open the possibility of probing the ultrafast dynamics of continuum electron.

quant-ph↗

arXiv2019

Convergence analysis of beetle antennae search algorithm and its applications

The beetle antennae search algorithm was recently proposed and investigated for solving global optimization problems. Although the performance of the algorithm and its variants were shown to be better than some existing meta-heuristic algorithms, there is still a lack of convergence analysis. In this paper, we provide theoretical analysis on the convergence of the beetle antennae search algorithm. We test the performance of the BAS algorithm via some representative benchmark functions. Meanwhile, some applications of the BAS algorithm are also presented.

cs.NE↗

arXiv2019

Disorder-induced nonlinear Hall effect with time-reversal symmetry

The nonlinear Hall effect has opened the door towards deeper understanding of topological states of matter. It can be observed as the double-frequency Hall voltage response to an ac longitudinal current in the presence of time-reversal symmetry. Disorder plays indispensable roles in various linear Hall effects, such as the localization in the quantized Hall effects and the extrinsic mechanisms of the anomalous, spin, and valley Hall effects. Unlike in the linear Hall effects, disorder enters the nonlinear Hall effect even in the leading order. However, the disorder-induced contribution to the nonlinear Hall effect has not been addressed. Here, we derive the formulas of the nonlinear Hall conductivity in the presence of disorder scattering. We apply the formulas to calculate the nonlinear Hall response of the tilted 2D Dirac model, which is the symmetry-allowed minimal model for the nonlinear Hall effect and can serve as a building block in realistic band structures. More importantly, we construct the general scaling law of the nonlinear Hall effect, which may help in experiments to distinguish disorder-induced contributions to the nonlinear Hall effect. This work will be instructive for exploring unconventional responses upon breaking discrete or crystal symmetries in emergent physical systems and materials.

cond-mat.mes-hall↗

arXiv2019

TopRank: A practical algorithm for online stochastic ranking

Online learning to rank is a sequential decision-making problem where in each round the learning agent chooses a list of items and receives feedback in the form of clicks from the user. Many sample-efficient algorithms have been proposed for this problem that assume a specific click model connecting rankings and user behavior. We propose a generalized click model that encompasses many existing models, including the position-based and cascade models. Our generalization motivates a novel online learning algorithm based on topological sort, which we call TopRank. TopRank is (a) more natural than existing algorithms, (b) has stronger regret guarantees than existing algorithms with comparable generality, (c) has a more insightful proof that leaves the door open to many generalizations, (d) outperforms existing algorithms empirically.

stat.ML↗

arXiv2019

Online Clustering of Contextual Cascading Bandits

We consider a new setting of online clustering of contextual cascading bandits, an online learning problem where the underlying cluster structure over users is unknown and needs to be learned from a random prefix feedback. More precisely, a learning agent recommends an ordered list of items to a user, who checks the list and stops at the first satisfactory item, if any. We propose an algorithm of CLUB-cascade for this setting and prove a $T$-step regret bound of order $\tilde{O}(\sqrt{T})$. Previous work corresponds to the degenerate case of only one cluster, and our general regret bound in this special case also significantly improves theirs. We conduct experiments on both synthetic and real data, and demonstrate the effectiveness of our algorithm and the advantage of incorporating online clustering method.

cs.LG↗

arXiv2018

A Survey on Blockchain Technology and Its Potential Applications in Distributed Control and Cooperative Robots

As a disruptive technology, blockchain, particularly its original form of bitcoin as a type of digital currency, has attracted great attentions. The innovative distributed decision making and security mechanism lay the technical foundation for its success, making us consider to penetrate the power of blockchain technology to distributed control and cooperative robotics, in which the distributed and secure mechanism is also highly demanded. Actually, security and distributed communication have long been unsolved problems in the field of distributed control and cooperative robotics. It has been reported on the network failure and intruder attacks of distributed control and multi-robotic systems. Blockchain technology provides promise to remedy this situation thoroughly. This work is intended to create a global picture of blockchain technology on its working principle and key elements in the language of control and robotics, to provide a shortcut for beginners to step into this research field.

cs.CR↗

arXiv2018

High-Resolution Limited-Angle Phase Tomography of Dense Layered Objects Using Deep Neural Networks

We present a Machine Learning-based method for tomographic reconstruction of dense layered objects, with range of projection angles limited to $\pm $10$^\circ$. Whereas previous approaches to phase tomography generally require two steps, first to retrieve phase projections from intensity projections and then perform tomographic reconstruction on the retrieved phase projections, in our work a physics-informed pre-processor followed by a Deep Neural Network (DNN) conduct the three-dimensional reconstruction directly from the intensity projections. We demonstrate this single-step method experimentally in the visible optical domain on a scaled up integrated circuit phantom. We show that even under conditions of highly attenuated photon fluxes a DNN trained only on synthetic data can be used to successfully reconstruct physical samples disjoint from the synthetic training set. Thus, the need of producing a large number of physical examples for training is ameliorated. The method is generally applicable to tomography with electromagnetic or other types of radiation at all bands.

eess.IV↗

arXiv2018

Measure, Manifold, Learning, and Optimization: A Theory Of Neural Networks

We present a formal measure-theoretical theory of neural networks (NN) built on probability coupling theory. Our main contributions are summarized as follows. * Built on the formalism of probability coupling theory, we derive an algorithm framework, named Hierarchical Measure Group and Approximate System (HMGAS), nicknamed S-System, that is designed to learn the complex hierarchical, statistical dependency in the physical world. * We show that NNs are special cases of S-System when the probability kernels assume certain exponential family distributions. Activation Functions are derived formally. We further endow geometry on NNs through information geometry, show that intermediate feature spaces of NNs are stochastic manifolds, and prove that "distance" between samples is contracted as layers stack up. * S-System shows NNs are inherently stochastic, and under a set of realistic boundedness and diversity conditions, it enables us to prove that for large size nonlinear deep NNs with a class of losses, including the hinge loss, all local minima are global minima with zero loss errors, and regions around the minima are flat basins where all eigenvalues of Hessians are concentrated around zero, using tools and ideas from mean field theory, random matrix theory, and nonlinear operator equations. * S-System, the information-geometry structure and the optimization behaviors combined completes the analog between Renormalization Group (RG) and NNs. It shows that a NN is a complex adaptive system that estimates the statistic dependency of microscopic object, e.g., pixels, in multiple scales. Unlike clear-cut physical quantity produced by RG in physics, e.g., temperature, NNs renormalize/recompose manifolds emerging through learning/optimization that divide the sample space into highly semantically meaningful groups that are dictated by supervised labels (in supervised NNs).

cs.LG↗

arXiv2018

Detection and spatial mapping of conductive filaments in metal/oxide/metal cross-point devices

A simple means of detecting and spatially mapping volatile and nonvolatile conductive filaments in metal/oxide/metal cross-point devices is introduced and its application demonstrated. The technique is based on thermal discolouration of a thin photoresist layer deposited on the top electrode of the cross-point device and relies on the increase in temperature produced by local Joule heating of an underlying conductive filament. Finite element modelling of the temperature distribution and its dependencies shows that the maximum temperature at the top-electrode/photoresist interface is particularly sensitive to the top-electrode thickness. The technique is demonstrated on NbOx based metal-oxide-metal cross-point devices with a 25 nm thick top (Pt) electrode, where it is used to undertake a statistical analysis of the filament location as a function of device area. This shows that filament formation is heterogeneous. The majority of filaments form preferentially along the top-electrode edge and the fraction of these increases with decreasing device area. Transmission electron microscopy of the top and bottom electrode edges is used to explain this observation and suggests that it is due to a reduction in the effective oxide thickness in this region.

physics.app-ph↗

arXiv2018

Learning to synthesize: splitting and recombining low and high spatial frequencies for image recovery

Deep Neural Network (DNN)-based image reconstruction, despite many successes, often exhibits uneven fidelity between high and low spatial frequency bands. In this paper we propose the Learning Synthesis by DNN (LS-DNN) approach where two DNNs process the low and high spatial frequencies, respectively, and, improving over [30], the two DNNs are trained separately and a third DNN combines them into an image with high fidelity at all bands. We demonstrate LS-DNN in two canonical inverse problems: super-resolution (SR) in diffraction-limited imaging (DLI), and quantitative phase retrieval (QPR). Our results also show comparable or improved performance over perceptual-loss based SR [21], and can be generalized to a wider range of image recovery problems.

cs.CV↗

arXiv2018

Spectral pre-modulation of training examples enhances the spatial resolution of the Phase Extraction Neural Network (PhENN)

The Phase Extraction Neural Network (PhENN) is a computational architecture, based on deep machine learning, for lens-less quantitative phase retrieval from raw intensity data. PhENN is a deep convolutional neural network trained through examples consisting of pairs of true phase objects and their corresponding intensity diffraction patterns; thereafter, given a test raw intensity pattern PhENN is capable of reconstructing the original phase object robustly, in many cases even for objects outside the database where the training examples were drawn from. Here, we show that the spatial frequency content of the training examples is an important factor limiting PhENN's spatial frequency response. For example, if the training database is relatively sparse in high spatial frequencies, as most natural scenes are, PhENN's ability to resolve fine spatial features in test patterns will be correspondingly limited. To combat this issue, we propose "flattening" the power spectral density of the training examples before presenting them to PhENN. For phase objects following the statistics of natural scenes, we demonstrate experimentally that the spectral pre-modulation method enhances the spatial resolution of PhENN by a factor of 2.

eess.IV↗

arXiv2018

Mass Concentration and Local Uniqueness of Ground States for $L^2$-subcritical Nonlinear Schrödinger Equations

We consider ground states of $L^2$-subcritical nonlinear Schrödinger equation (1.1), which can be described equivalently by minimizers of the following constraint minimization problem $$ e(ρ):=\inf\{E_ρ(u):u\in \mathcal{H}(\mathbb{R}^d),\|u\|_2^2=1\}.$$ The energy functional $E_ρ(u)$ is defined by $$ E_ρ(u):=\frac{1}{2}\int_{\mathbb{R}^d}|\nabla u|^2dx +\frac{1}{2}\int_{\mathbb{R}^d}V(x)|u|^2dx-\frac{ρ^{p-1}}{p+1}\int_{\mathbb{R}^d}|u|^{p+1}dx,$$ where $d\geq1$, $ρ>0$, $p\in\big(1, 1+\frac{4}{d}\big)$ and $0\leq V(x)\to\infty$ as $|x| \to\infty$. We present a detailed analysis on the concentration behavior of ground states as $ρ\to\infty$, which extends the concentration results shown in [22]. Moreover, the uniqueness of nonnegative ground states is also proved when $ρ$ is large enough.

math.AP↗

arXiv2018

Low Photon Count Phase Retrieval Using Deep Learning

Imaging systems' performance at low light intensity is affected by shot noise, which becomes increasingly strong as the power of the light source decreases. In this paper we experimentally demonstrate the use of deep neural networks to recover objects illuminated with weak light and demonstrate better performance than with the classical Gerchberg-Saxton phase retrieval algorithm for equivalent signal over noise ratio. Prior knowledge about the object is implicitly contained in the training data set and feature detection is possible for a signal over noise ratio close to one. We apply this principle to a phase retrieval problem and show successful recovery of the object's most salient features with as little as one photon per detector pixel on average in the illumination beam. We also show that the phase reconstruction is significantly improved by training the neural network with an initial estimate of the object, as opposed as training it with the raw intensity measurement.

eess.IV↗

arXiv2018

Bloch bound state of spin-orbit-coupled fermions in an optical lattice

Understanding fundamentals of few-body physics provides an interesting bottom-up approach for the clarification of many-body properties. The remarkable experimental progress in realizing spin-orbit coupling (SOC) in optical Raman lattices offers a renewed thrust towards discovering novel few-body features induced by the interplay between SOC and optical lattices. Using the Wilson renormalization method to account for high-band effects, we study the low-energy two-body scattering processes of spin-$1/2$ fermions in spin-orbit coupled optical lattices. We demonstrate that, under weak SOC, adding a small lattice potential would destabilize shallow two-body bound states, contrary to conventional wisdom. On the other hand, when lattice is sufficiently deep, two-body bound states are always stabilized by increasing the lattice depth. This intriguing non-monotonic behavior of the bound-state stability derives from the competition between SOC and optical lattices, and can be explained by analyzing the low-energy density of states. We also discuss the impact of high-band effects on such a behavior, as well as potential experimental detections.

cond-mat.quant-gas↗

arXiv2018

Offline Evaluation of Ranking Policies with Click Models

Many web systems rank and present a list of items to users, from recommender systems to search and advertising. An important problem in practice is to evaluate new ranking policies offline and optimize them before they are deployed. We address this problem by proposing evaluation algorithms for estimating the expected number of clicks on ranked lists from historical logged data. The existing algorithms are not guaranteed to be statistically efficient in our problem because the number of recommended lists can grow exponentially with their length. To overcome this challenge, we use models of user interaction with the list of items, the so-called click models, to construct estimators that learn statistically efficiently. We analyze our estimators and prove that they are more efficient than the estimators that do not use the structure of the click model, under the assumption that the click model holds. We evaluate our estimators in a series of experiments on a real-world dataset and show that they consistently outperform prior estimators.

cs.LG↗

arXiv2018

Gear Training: A new way to implement high-performance model-parallel training

The training of Deep Neural Networks usually needs tremendous computing resources. Therefore many deep models are trained in large cluster instead of single machine or GPU. Though major researchs at present try to run whole model on all machines by using asynchronous asynchronous stochastic gradient descent (ASGD), we present a new approach to train deep model parallely -- split the model and then seperately train different parts of it in different speed.

cs.LG↗