arXiv subjects

Li Zhang

Publications and source records attributed to Li Zhang.

At least 883 records · Page 49Linked to original sources

arXiv2017

Spatially Adaptive Computation Time for Residual Networks

This paper proposes a deep learning architecture based on Residual Network that dynamically adjusts the number of executed layers for the regions of the image. This architecture is end-to-end trainable, deterministic and problem-agnostic. It is therefore applicable without any modifications to a wide range of computer vision problems such as image classification, object detection and image segmentation. We present experimental results showing that this model improves the computational efficiency of Residual Networks on the challenging ImageNet classification and COCO object detection datasets. Additionally, we evaluate the computation time maps on the visual saliency dataset cat2000 and find that they correlate surprisingly well with human eye fixation positions.

cs.CV↗

arXiv2017

Learning to Learn: Meta-Critic Networks for Sample Efficient Learning

We propose a novel and flexible approach to meta-learning for learning-to-learn from only a few examples. Our framework is motivated by actor-critic reinforcement learning, but can be applied to both reinforcement and supervised learning. The key idea is to learn a meta-critic: an action-value function neural network that learns to criticise any actor trying to solve any specified task. For supervised learning, this corresponds to the novel idea of a trainable task-parametrised loss generator. This meta-critic approach provides a route to knowledge transfer that can flexibly deal with few-shot and semi-supervised conditions for both reinforcement and supervised learning. Promising results are shown on both reinforcement and supervised learning problems.

cs.LG↗

arXiv2017

Design-Phase Buffer Allocation for Post-Silicon Clock Binning by Iterative Learning

At submicron manufacturing technology nodes, pro- cess variations affect circuit performance significantly. To counter these variations, engineers are reserving more timing margin to maintain yield, leading to an unaffordable overdesign. Most of these margins, however, are wasted after manufacturing, because process variations cause only some chips to be really slow, while other chips can easily meet given timing specifications. To reduce this pessimism, we can reserve less timing margin and tune failed chips after manufacturing with clock buffers to make them meet timing specifications. With this post-silicon clock tuning, critical paths can be balanced with neighboring paths in each chip specifically to counter the effect of process variations. Consequently, chips with timing failures can be rescued and the yield can thus be improved. This is specially useful in high- performance designs, e.g., high-end CPUs, where clock binning makes chips with higher performance much more profitable. In this paper, we propose a method to determine where to insert post-silicon tuning buffers during the design phase to improve the overall profit with clock binning. This method learns the buffer locations with a Sobol sequence iteratively and reduces the buffer ranges afterwards with tuning concentration and buffer grouping. Experimental results demonstrate that the proposed method can achieve a profit improvement of about 14% on average and up to 26%, with only a small number of tuning buffers inserted into the circuit.

cs.AR↗

arXiv2017

Nonlinear parametric excitation effect induces stability transitions in swimming direction of flexible superparamagnetic microswimmers

Microscopic artificial swimmers have recently become highly attractive due to their promising potential for biomedical applications. The pioneering work of Dreyfus et al (2005) has demonstrated the motion of a microswimmer with an undulating chain of superparamagnetic beads, which is actuated by an oscillating external magnetic field. Interestingly, it has also been theoretically predicted that the swimming direction of this swimmer will undergo a $90^\circ$-transition when the magnetic field's oscillations amplitude is increased above a critical value of $\sqrt{2}$. In this work, we further investigate this transition both theoretically and experimentally by using numerical simulations and presenting a novel flexible microswimmer with a superparamagnetic head. We realize the $90^\circ$-transition in swimming direction, prove that this effect depends on both frequency and amplitude of the oscillating magnetic field, and demonstrate the existence of an optimal amplitude, under which, maximal swimming speed can be achieved. By asymptotically analyzing the dynamic motion of microswimmer with a minimal two-link model, we reveal that the stability transitions representing the changes in the swimming direction are induced by the effect of nonlinear parametric excitation.

physics.flu-dyn↗

arXiv2017

Superconductor to weak-insulator transitions in disordered Tantalum Nitride films

We study the two-dimensional superconductor-insulator transition (SIT) in thin films of tantalum nitride. At zero magnetic field, films can be disorder-tuned across the SIT by adjusting thickness and film stoichiometry; insulating films exhibit classical hopping transport. Superconducting films exhibit a magnetic field-tuned SIT, whose insulating ground state at high field appears to be a quantum-corrected metal. Scaling behavior at the field-tuned SIT shows classical percolation critical exponents $z ν\approx$ 1.3, with a corresponding critical field $H_c \ll H_{c2}$. The Hall effect shows a crossing point near $H_c$, but with a non-universal critical value $ρ_{xy}^c$ comparable to the normal state Hall resistivity. We propose that high-carrier density metals will always exhibit this pattern of behavior at the boundary between superconducting and (trivially) insulating ground states.

cond-mat.supr-con↗

arXiv2017

Topological invariant and cotranslational symmetry in strongly interacting multi-magnon systems

It is still an outstanding challenge to characterize and understand the topological features of strongly interacting states such as bound-states in interacting quantum systems. Here, by introducing a cotranslational symmetry in an interacting multi-particle quantum system, we systematically develop a method to define a Chern invariant, which is a generalization of the well-known Thouless-Kohmoto-Nightingale-den Nijs invariant, for identifying strongly interacting topological states. As an example, we study the topological multi-magnon states in a generalized Heisenberg XXZ model, which can be realized by the currently available experiment techniques of cold atoms [Phys. Rev. Lett. \textbf{111}, 185301 (2013); Phys. Rev. Lett. \textbf{111}, 185302 (2013)]. Through calculating the two-magnon excitation spectrum and the defined Chern number, we explore the emergence of topological edge bound-states and give their topological phase diagram. We also analytically derive an effective single-particle Hofstadter superlattice model for a better understanding of the topological bound-states. Our results not only provide a new approach to defining a topological invariant for interacting multi-particle systems, but also give insights into the characterization and understanding of strongly interacting topological states.

cond-mat.quant-gas↗

arXiv2017

Localizing High-Lying Rydberg Wave Packets with Two-Color Laser Fields

We demonstrate control over the localization of high-lying Rydberg wave packets in argon atoms with phase-locked orthogonally polarized two-color (OTC) laser fields. With a reaction microscope, we measured ionization signals of high-lying Rydberg states induced by a weak dc field and black-body radiation as a function of the relative phase between the two-color fields. We find that the dc-field ionization yields of high-lying Rydberg argon atoms oscillate with the relative two-color phase with a period of $2π$ while the photoionization signal by black-body radiation shows a period of $π$. These observations are a clear signature of the asymmetric localization of electrons recaptured into high-lying Rydberg states after conclusion of the laser pulse and are supported by a semiclassical simulation of argon-OTC laser interaction. Our findings thus open an effective pathway to control the localization of high-lying Rydberg wave packets.

physics.atom-ph↗

arXiv2017

Minimum Enclosing Circle of a Set of Static Points with Dynamic Weight from One Free Point

Given a set $S$ of $n$ static points and a free point $p$ in the Euclidean plane, we study a new variation of the minimum enclosing circle problem, in which a dynamic weight that equals to the reciprocal of the distance from the free point $p$ to the undetermined circle center is included. In this work, we prove the optimal solution of the new problem is unique and lies on the boundary of the farthest-point Voronoi diagram of $S$, once $p$ does not coincide with any vertex of the convex hull of $S$. We propose a tree structure constructed from the boundary of the farthest-point Voronoi diagram and use the hierarchical relationship between edges to locate the optimal solution. The plane could be divide into at most $3n-4$ non-overlapping regions. When $p$ lies in one of the regions, the optimal solution locates at one node or lies on the interior of one edge in the boundary of the farthest-point Voronoi diagram. Moreover, we apply the new variation to calculate the maximum displacement of one point $p$ under the condition that the displacements of points in $S$ are restricted in 2D rigid motion.

cs.CG↗

arXiv2017

Classification of Neurological Gait Disorders Using Multi-task Feature Learning

As our population ages, neurological impairments and degeneration of the musculoskeletal system yield gait abnormalities, which can significantly reduce quality of life. Gait rehabilitative therapy has been widely adopted to help patients maximize community participation and living independence. To further improve the precision and efficiency of rehabilitative therapy, more objective methods need to be developed based on sensory data. In this paper, an algorithmic framework is proposed to provide classification of gait disorders caused by two common neurological diseases, stroke and Parkinson's Disease (PD), from ground contact force (GCF) data. An advanced machine learning method, multi-task feature learning (MTFL), is used to jointly train classification models of a subject's gait in three classes, post-stroke, PD and healthy gait. Gait parameters related to mobility, balance, strength and rhythm are used as features for the classification. Out of all the features used, the MTFL models capture the more important ones per disease, which will help provide better objective assessment and therapy progress tracking. To evaluate the proposed methodology we use data from a human participant study, which includes five PD patients, three post-stroke patients, and three healthy subjects. Despite the diversity of abnormalities, the evaluation shows that the proposed approach can successfully distinguish post-stroke and PD gait from healthy gait, as well as post-stroke from PD gait, with Area Under the Curve (AUC) score of at least 0.96. Moreover, the methodology helps select important gait features to better understand the key characteristics that distinguish abnormal gaits and design personalized treatment.

cs.CV↗

arXiv2017

Entire Solution in an Ignition Nonlocal Dispersal Equation: Asymmetric Kernel

This paper mainly focus on the front-like entire solution of a classical nonlocal dispersal equation with ignition nonlinearity. Especially, the dispersal kernel function $J$ may not be symmetric here. The asymmetry of $J$ has a great influence on the profile of the traveling waves and the sign of the wave speeds, which further makes the properties of the entire solution more diverse. We first investigate the asymptotic behavior of the traveling wave solutions since it plays an essential role in obtaining the front-like entire solution. Due to the impact of $f'(0)=0$, we can no longer use the common method which mainly depending on Ikehara theorem and bilateral Laplace transform to study the asymptotic rates of the nondecreasing traveling wave and the nonincreasing one tending to 0, respectively, thus we adopt another method to investigate them. Afterwards, we establish a new entire solution and obtain its qualitative properties by constructing proper supersolution and subsolution and by classifying the sign and size of the wave speeds.

math.AP↗

arXiv2016

Study of linearity of LYSO crystal for HERD calorimeter

The High Energy cosmic Radiation Detection (HERD) facility is a space mission designed for detecting cosmic ray (CR) electrons, $γ$-rays up to tens of TeV and CR nuclei from proton to iron up to several PeV. The main instrument of HERD is a 3-D imaging calorimeter (CALO) composed of nearly ten thousand cubic LYSO crystals. A large dynamic range of single HERD CALO Cell (HCC) is necessary to achieve HERD's PeV observation objectives, which means that the response of HCC should maintain a good linearity from minimum ionizing particle (MIP) calibration to PeV shower maximum. In order to study the linearity of HCC over such a large energy range, a beam test has been implemented at the E2 and E3 beam lines of BEPC. High intensity pulsed electron beam provided by E2 line are used for producing high energy density within HCC; $π^{+}$/proton provided by E3 line are used for HCC calibration. The results show that no saturation effect occurs and the linearity of HCC is better than 10% from 30 MeV (1 MIP) to 1.1$\times$10$^{3}$ TeV (energy density is 93 TeV/cm$^{3}$), which can meet the requirement mentioned above.

physics.ins-det↗

arXiv2016

Evolution of High-Energy Particle Distribution in Mature Shell-Type Supernova Remnants

Multi-wavelength observations of mature supernova remnants (SNRs), especially with recent advances in gamma-ray astronomy, make it possible to constrain energy distribution of energetic particles within these remnants. In consideration of the SNR origin of Galactic cosmic rays and physics related to particle acceleration and radiative processes, we use a simple one-zone model to fit the nonthermal emission spectra of three shell-type SNRs located within 2 degrees on the sky: RX J1713.7-3946, CTB 37B, and CTB 37A. Although radio images of these three sources all show a shell (or half-shell) structure, their radio, X-ray, and gamma-ray spectra are quite different, offering an ideal case to explore evolution of energetic particle distribution in SNRs. Our spectral fitting shows that 1) the particle distribution becomes harder with aging of these SNRs, implying a continuous acceleration process, and the particle distributions of CTB 37A and CTB 37B in the GeV range are harder than the hardest distribution that can be produced at a shock via the linear diffusive shock particle acceleration process, so spatial transport may play a role; 2) the energy loss timescale of electrons at the high-energy cutoff due to synchrotron radiation appears to be always a bit (within a factor of a few) shorter than the age of the corresponding remnant, which also requires continuous particle acceleration; 3) double power-law distributions are needed to fit the spectra of CTB 37B and CTB 37A, which may be attributed to shock interaction with molecular clouds.

astro-ph.HE↗

arXiv2016

Private Empirical Risk Minimization Beyond the Worst Case: The Effect of the Constraint Set Geometry

Empirical Risk Minimization (ERM) is a standard technique in machine learning, where a model is selected by minimizing a loss function over constraint set. When the training dataset consists of private information, it is natural to use a differentially private ERM algorithm, and this problem has been the subject of a long line of work started with Chaudhuri and Monteleoni 2008. A private ERM algorithm outputs an approximate minimizer of the loss function and its error can be measured as the difference from the optimal value of the loss function. When the constraint set is arbitrary, the required error bounds are fairly well understood \cite{BassilyST14}. In this work, we show that the geometric properties of the constraint set can be used to derive significantly better results. Specifically, we show that a differentially private version of Mirror Descent leads to error bounds of the form $\tilde{O}(G_{\mathcal{C}}/n)$ for a lipschitz loss function, improving on the $\tilde{O}(\sqrt{p}/n)$ bounds in Bassily, Smith and Thakurta 2014. Here $p$ is the dimensionality of the problem, $n$ is the number of data points in the training set, and $G_{\mathcal{C}}$ denotes the Gaussian width of the constraint set that we optimize over. We show similar improvements for strongly convex functions, and for smooth functions. In addition, we show that when the loss function is Lipschitz with respect to the $\ell_1$ norm and $\mathcal{C}$ is $\ell_1$-bounded, a differentially private version of the Frank-Wolfe algorithm gives error bounds of the form $\tilde{O}(n^{-2/3})$. This captures the important and common case of sparse linear regression (LASSO), when the data $x_i$ satisfies $|x_i|_{\infty} \leq 1$ and we optimize over the $\ell_1$ ball. We show new lower bounds for this setting, that together with known bounds, imply that all our upper bounds are tight.

cs.LG↗

arXiv2016

Deep Learning with Differential Privacy

Machine learning techniques based on neural networks are achieving remarkable results in a wide variety of domains. Often, the training of models requires large, representative datasets, which may be crowdsourced and contain sensitive information. The models should not expose private information in these datasets. Addressing this goal, we develop new algorithmic techniques for learning and a refined analysis of privacy costs within the framework of differential privacy. Our implementation and experiments demonstrate that we can train deep neural networks with non-convex objectives, under a modest privacy budget, and at a manageable cost in software complexity, training efficiency, and model quality.

stat.ML↗

arXiv2016

Testing one-zone synchrotron-self-Compton models with spectral energy distributions of Mrk 421

We test one-zone synchrotron self-Compton (SSC) models with high-quality multiwavelength spectral energy distribution (SED) data of Mrk 421. We use Markov chain Monte Carlo (MCMC) technique to fit twelve day-scale SEDs of Mrk 421 with one-zone SSC models. Three types of electron energy distribution (EED), a log-parabola (LP) EED, a power-law log-parabola (PLLP) EED and a broken power-law (BPL) EED, are assumed in fits. We find that the one-zone SSC model with the PLLP EED provides successful fits to all the twelve SEDs. However, the one-zone SSC model with the LP and BPL EEDs fail to provide acceptable fits to the highest energy X-ray data or GeV data in several states. We therefore conclude that the one-zone SSC model works well in explaining the SEDs of Mrk 421, and the PLLP EED is preferred over the LP and BPL EEDs for Mrk 421 during the flare in March 2010. We derive magnetic field $B'\sim0.01$\ G, Doppler factor $δ_{\rm D}\sim$30--50, and the curvature parameter of EED $r\sim1$--$10$ in the model with the PLLP EED. The evolutions of model parameters are explored. The physical implications of our results are discussed.

astro-ph.HE↗

arXiv2016

Tracking Completion

A fundamental component of modern trackers is an online learned tracking model, which is typically modeled either globally or locally. The two kinds of models perform differently in terms of effectiveness and robustness under different challenging situations. This work exploits the advantages of both models. A subspace model, from a global perspective, is learned from previously obtained targets via rank-minimization to address the tracking, and a pixel-level local observation is leveraged si- multaneously, from a local point of view, to augment the subspace model. A matrix completion method is employed to integrate the two models. Unlike previous tracking methods, which locate the target among all fully observed target candidates, the proposed approach first estimates an expected target via the matrix completion through partially observed target candidates, and then, identifies the target according to the estimation accuracy with respect to the target candidates. Specifically, the tracking is formulated as a problem of target appearance estimation. Extensive experiments on various challenging video sequences verify the effectiveness of the proposed approach and demonstrate that the proposed tracker outperforms other popular state-of-the-art trackers.

cs.CV↗

arXiv2016

Real-Time Visual Tracking: Promoting the Robustness of Correlation Filter Learning

Correlation filtering based tracking model has received lots of attention and achieved great success in real-time tracking, however, the lost function in current correlation filtering paradigm could not reliably response to the appearance changes caused by occlusion and illumination variations. This study intends to promote the robustness of the correlation filter learning. By exploiting the anisotropy of the filter response, three sparsity related loss functions are proposed to alleviate the overfitting issue of previous methods and improve the overall tracking performance. As a result, three real-time trackers are implemented. Extensive experiments in various challenging situations demonstrate that the robustness of the learned correlation filter has been greatly improved via the designed loss functions. In addition, the study reveals, from an experimental perspective, how different loss functions essentially influence the tracking performance. An important conclusion is that the sensitivity of the peak values of the filter in successive frames is consistent with the tracking performance. This is a useful reference criterion in designing a robust correlation filter for visual tracking.

cs.CV↗

arXiv2016

Cluster Mean-Field Signature of Entanglement Entropy in Bosonic Superfluid-Insulator Transitions

Entanglement entropy (EE), a fundamental conception in quantum information for characterizing entanglement, has been extensively employed to explore quantum phase transitions (QPTs). Although the conventional single-site mean-field (MF) approach successfully predicts the emergence of QPTs, it fails to include any entanglement. Here, for the first time, in the framework of a cluster MF treatment, we extract the signature of EE in the bosonic superfluid-insulator transitions. We consider a trimerized Kagome lattice of interacting bosons, in which each trimer is treated as a cluster, and implement the cluster MF treatment by decoupling all inter-trimer hopping. In addition to superfluid and integer insulator phases, we find that fractional insulator phases appear when the tunneling is dominated by the intra-trimer part. To quantify the residual bipartite entanglement in a cluster, we calculate the second-order Renyi entropy, which can be experimentally measured by quantum interference of many-body twins. The second-order Renyi entropy itself is continuous everywhere, however, the continuousness of its first-order derivative breaks down at the phase boundary. This means that the bosonic superfluid-insulator transitions can still be efficiently captured by the residual entanglement in our cluster MF treatment. Besides to the bosonic superfluid-insulator transitions, our cluster MF treatment may also be used to capture the signature of EE for other QPTs in quantum superlattice models.

cond-mat.quant-gas↗