arXiv subjects

Chi Zhang

Publications and source records attributed to Chi Zhang.

At least 847 records · Page 47Linked to original sources

arXiv2018

Batch-normalized Recurrent Highway Networks

Gradient control plays an important role in feed-forward networks applied to various computer vision tasks. Previous work has shown that Recurrent Highway Networks minimize the problem of vanishing or exploding gradients. They achieve this by setting the eigenvalues of the temporal Jacobian to 1 across the time steps. In this work, batch normalized recurrent highway networks are proposed to control the gradient flow in an improved way for network convergence. Specifically, the introduced model can be formed by batch normalizing the inputs at each recurrence loop. The proposed model is tested on an image captioning task using MSCOCO dataset. Experimental results indicate that the batch normalized recurrent highway networks converge faster and performs better compared with the traditional LSTM and RHN based models.

cs.LG↗

arXiv2018

Notes on Scattering Amplitudes as Differential Forms

Inspired by the idea of viewing amplitudes in ${\cal N}=4$ SYM as differential forms on momentum twistor space, we introduce differential forms on the space of spinor variables, which combine helicity amplitudes in any four-dimensional gauge theory as a single object. In this note we focus on such differential forms in ${\cal N}=4$ SYM, which can also be thought of as "bosonizing" superamplitudes in non-chiral superspace. Remarkably all tree-level amplitudes in ${\cal N}=4$ SYM combine to a $d\log$ form in spinor variables, which is given by pushforward of canonical forms of Grassmannian cells, the tree forms can also be obtained using BCFW or inverse-soft construction, and we present all-multiplicity expression for MHV and NMHV forms to illustrate their simplicity. Similarly all-loop planar integrands can be naturally written as $d\log$ forms in the Grassmannian/on-shell-diagram picture, and we expect the same to hold beyond the planar limit. Just as the form in momentum twistor space reveals underlying positive geometry of the amplituhedron, the form in terms of spinor variables strongly suggests an "amplituhedron in momentum space". We initiate the study of its geometry by connecting it to the moduli space of Witten's twistor-string theory, which provides a pushforward formula for tree forms in ${\cal N}=4$ SYM.

hep-th↗

arXiv2018

Scattering Forms, Worldsheet Forms and Amplitudes from Subspaces

We present a general construction of two types of differential forms, based on any $(n{-}3)$-dimensional subspace in the kinematic space of $n$ massless particles. The first type is the so-called projective, scattering forms in kinematic space, while the second is defined in the moduli space of $n$-punctured Riemann spheres which we call worldsheet forms. We show that the pushforward of worldsheet forms, by summing over solutions of scattering equations, gives the corresponding scattering forms, which generalizes the results of [1711.09102]. The pullback of scattering forms to subspaces can have natural interpretations as amplitudes in terms of Bern-Carrasco-Johansson double-copy construction or Cachazo-He-Yuan formula. As an application of our formalism, we construct in this way a large class of $d\log$ scattering forms and worldsheet forms, which are in one-to-one correspondence with non-planar MHV leading singularities in ${\cal N}=4$ super-Yang-Mills. For every leading singularity function, we present a new determinant formula in moduli space, as well as a (combinatoric) polytope and associated scattering form in kinematic space. These include the so-called Cayley cases, where in each case the scattering form is the canonical forms of a convex polytope in the subspace, and scattering equations admit elegant rewritings as a map from the moduli space to the subspace.

hep-th↗

arXiv2018

An Integration of Bottom-up and Top-Down Salient Cues on RGB-D Data: Saliency from Objectness vs. Non-Objectness

Bottom-up and top-down visual cues are two types of information that helps the visual saliency models. These salient cues can be from spatial distributions of the features (space-based saliency) or contextual / task-dependent features (object based saliency). Saliency models generally incorporate salient cues either in bottom-up or top-down norm separately. In this work, we combine bottom-up and top-down cues from both space and object based salient features on RGB-D data. In addition, we also investigated the ability of various pre-trained convolutional neural networks for extracting top-down saliency on color images based on the object dependent feature activation. We demonstrate that combining salient features from color and dept through bottom-up and top-down methods gives significant improvement on the salient object detection with space based and object based salient cues. RGB-D saliency integration framework yields promising results compared with the several state-of-the-art-models.

cs.CV↗

arXiv2018

Thermopower and Nernst measurements in a half-filled lowest Landau level

Motivated by recent proposal by Potter et al. [Phys. Rev. X 6, 031026 (2016)] concerning possible thermoelectric signatures of Dirac composite fermions, we perform a systematic experimental study of thermoelectric transport of an ultrahigh-mobility GaAs/AlxGa1-xAs two dimensional electron system at filling factor v = 1/2. We demonstrate that the thermopower Sxx and Nernst Sxy are symmetric and anti-symmetric with respect to B = 0 T, respectively. The measured properties of thermopower Sxx at v = 1/2 are consistent with previous experimental results. The Nernst signals Sxy of v = 1/2, which have not been reported previously, are non-zero and show a power law relation with temperature in the phonon-drag dominant region. In the electron-diffusion dominant region, the Nernst signals Sxy of v = 1/2 are found to be significantly smaller than the linear temperature dependent values predicted by Potter et al., and decreasing with temperature faster than linear dependence.

cond-mat.mes-hall↗

arXiv2018

Learning with Non-Convex Truncated Losses by SGD

Learning with a {\it convex loss} function has been a dominating paradigm for many years. It remains an interesting question how non-convex loss functions help improve the generalization of learning with broad applicability. In this paper, we study a family of objective functions formed by truncating traditional loss functions, which is applicable to both shallow learning and deep learning. Truncating loss functions has potential to be less vulnerable and more robust to large noise in observations that could be adversarial. More importantly, it is a generic technique without assuming the knowledge of noise distribution. To justify non-convex learning with truncated losses, we establish excess risk bounds of empirical risk minimization based on truncated losses for heavy-tailed output, and statistical error of an approximate stationary point found by stochastic gradient descent (SGD) method. Our experiments for shallow and deep learning for regression with outliers, corrupted data and heavy-tailed noise further justify the proposed method.

cs.LG↗

arXiv2018

On an open problem of Skiba

Let $σ=\{σ_{i}|i\in I\}$ be some partition of the set $\mathbb{P}$ of all primes, that is, $\mathbb{P}=\bigcup_{i\in I}σ_{i}$ and $σ_{i}\cap σ_{j}=\emptyset$ for all $i\neq j$. Let $G$ be a finite group. A set $\mathcal {H}$ of subgroups of $G$ is said to be a complete Hall $σ$-set of $G$ if every non-identity member of $\mathcal {H}$ is a Hall $σ_{i}$-subgroup of $G$ and $\mathcal {H}$ contains exactly one Hall $σ_{i}$-subgroup of $G$ for every $σ_{i}\in σ(G)$. $G$ is said to be a $σ$-group if it possesses a complete Hall $σ$-set. A $σ$-group $G$ is said to be $σ$-dispersive provided $G$ has a normal series $1 = G_1<G_2<\cdots< G_t< G_{t+1} = G$ and a complete Hall $σ$-set $\{H_{1}, H_{2}, \cdots, H_{t}\}$ such that $G_iH_i = G_{i+1}$ for all $i= 1,2,\ldots t$. In this paper, we give a characterizations of $σ$-dispersive group, which give a positive answer to an open problem of Skiba in the paper.

math.GR↗

arXiv2018

AlignedReID: Surpassing Human-Level Performance in Person Re-Identification

In this paper, we propose a novel method called AlignedReID that extracts a global feature which is jointly learned with local features. Global feature learning benefits greatly from local feature learning, which performs an alignment/matching by calculating the shortest path between two sets of local features, without requiring extra supervision. After the joint learning, we only keep the global feature to compute the similarities between images. Our method achieves rank-1 accuracy of 94.4% on Market1501 and 97.8% on CUHK03, outperforming state-of-the-art methods by a large margin. We also evaluate human-level performance and demonstrate that our method is the first to surpass human-level performance on Market1501 and CUHK03, two widely used Person ReID datasets.

cs.CV↗

arXiv2018

Constraint-free Natural Image Reconstruction from fMRI Signals Based on Convolutional Neural Network

In recent years, research on decoding brain activity based on functional magnetic resonance imaging (fMRI) has made remarkable achievements. However, constraint-free natural image reconstruction from brain activity is still a challenge. The existing methods simplified the problem by using semantic prior information or just reconstructing simple images such as letters and digitals. Without semantic prior information, we present a novel method to reconstruct nature images from fMRI signals of human visual cortex based on the computation model of convolutional neural network (CNN). Firstly, we extracted the units output of viewed natural images in each layer of a pre-trained CNN as CNN features. Secondly, we transformed image reconstruction from fMRI signals into the problem of CNN feature visualizations by training a sparse linear regression to map from the fMRI patterns to CNN features. By iteratively optimization to find the matched image, whose CNN unit features become most similar to those predicted from the brain activity, we finally achieved the promising results for the challenging constraint-free natural image reconstruction. As there was no use of semantic prior information of the stimuli when training decoding model, any category of images (not constraint by the training set) could be reconstructed theoretically. We found that the reconstructed images resembled the natural stimuli, especially in position and shape. The experimental results suggest that hierarchical visual features can effectively express the visual perception process of human brain.

cs.CV↗

arXiv2018

Accurate reconstruction of image stimuli from human fMRI based on the decoding model with capsule network architecture

In neuroscience, all kinds of computation models were designed to answer the open question of how sensory stimuli are encoded by neurons and conversely, how sensory stimuli can be decoded from neuronal activities. Especially, functional Magnetic Resonance Imaging (fMRI) studies have made many great achievements with the rapid development of the deep network computation. However, comparing with the goal of decoding orientation, position and object category from activities in visual cortex, accurate reconstruction of image stimuli from human fMRI is a still challenging work. In this paper, the capsule network (CapsNet) architecture based visual reconstruction (CNAVR) method is developed to reconstruct image stimuli. The capsule means containing a group of neurons to perform the better organization of feature structure and representation, inspired by the structure of cortical mini column including several hundred neurons in primates. The high-level capsule features in the CapsNet includes diverse features of image stimuli such as semantic class, orientation, location and so on. We used these features to bridge between human fMRI and image stimuli. We firstly employed the CapsNet to train the nonlinear mapping from image stimuli to high-level capsule features, and from high-level capsule features to image stimuli again in an end-to-end manner. After estimating the serviceability of each voxel by encoding performance to accomplish the selecting of voxels, we secondly trained the nonlinear mapping from dimension-decreasing fMRI data to high-level capsule features. Finally, we can predict the high-level capsule features with fMRI data, and reconstruct image stimuli with the CapsNet. We evaluated the proposed CNAVR method on the dataset of handwritten digital images, and exceeded about 10% than the accuracy of all existing state-of-the-art methods on the structural similarity index (SSIM).

cs.CV↗

arXiv2017

Multi-Target, Multi-Camera Tracking by Hierarchical Clustering: Recent Progress on DukeMTMC Project

Although many methods perform well in single camera tracking, multi-camera tracking remains a challenging problem with less attention. DukeMTMC is a large-scale, well-annotated multi-camera tracking benchmark which makes great progress in this field. This report is dedicated to briefly introduce our method on DukeMTMC and show that simple hierarchical clustering with well-trained person re-identification features can get good results on this dataset.

cs.CV↗

arXiv2017

SOT for MOT

In this paper we present a robust tracker to solve the multiple object tracking (MOT) problem, under the framework of tracking-by-detection. As the first contribution, we innovatively combine single object tracking (SOT) algorithms with multiple object tracking algorithms, and our results show that SOT is a general way to strongly reduce the number of false negatives, regardless of the quality of detection. Another contribution is that we show with a deep learning based appearance model, it is easy to associate detections of the same object efficiently and also with high accuracy. This appearance model plays an important role in our MOT algorithm to correctly associate detections into long trajectories, and also in our SOT algorithm to discover new detections mistakenly missed by the detector. The deep neural network based model ensures the robustness of our tracking algorithm, which can perform data association in a wide variety of scenes. We ran comprehensive experiments on a large-scale and challenging dataset, the MOT16 benchmark, and results showed that our tracker achieved state-of-the-art performance based on both public and private detections.

cs.CV↗

arXiv2017

Molecular Clock Dating using MrBayes

This paper provides an overview and a tutorial of molecular clock dating using MrBayes, which is a software for Bayesian inference of phylogeny. Two modern approaches, total-evidence dating and node dating, are demonstrated using a dataset of Hymenoptera with molecular sequences and morphological characters. The similarity and difference of the two methods are compared and discussed. Besides, a non-clock analysis is performed on the same dataset to compare with the molecular clock dating analyses.

q-bio.PE↗

arXiv2017

Anisotropic Pseudopotential Characterization of Quantum Hall Systems under Tilted Magnetic Field

We analytically derived the effective two-body interaction for a finite thickness quantum Hall system with a harmonic perpendicular confinement and an in-plane magnetic field. The anisotropic effective interaction in the lowest Landau level (LLL) and first Landau level (1LL) are expanded in the basis of the generalized pseudopotentials (PPs), and we analyze how the coefficients of some prominent isotropic and anisotropic PPs depend on the thickness of the sample and the strength of the in-plane magnetic field. We also investigate the stability of the topological quantum Hall states, especially the Laughlin state and its emergent guiding center metric, which we can now compute analytically. An interesting reorientation of the anisotropy direction of the Laughlin state in the 1LL is revealed, and we also discuss various possible experimental ramifications for this quantum Hall system with broken rotational symmetry.

cond-mat.str-el↗

arXiv2017

Improved diffusing wave spectroscopy based on the automatized determination of the optical transport and absorption mean free path

Diffusing wave spectroscopy (DWS) can be employed as an optical rheology tool with numerous applications for studying the structure, dynamics and linear viscoelastic properties of complex fluids, foams, glasses and gels. To carry out DWS measurements, one first needs to quantify the static optical properties of the sample under investigation, i.e. the transport mean free path $l^\ast$ and the absorption length $l_a$. In the absence of absorption this can be done by comparing the diffuse optical transmission to a calibration sample whose $l^\ast$ is known. Performing this comparison however is cumbersome, time consuming and prone to mistakes by the operator. Moreover, already weak absorption can lead to significant errors. In this paper, we demonstrate the implementation of an automatized approach, based on which the DWS measurement procedure can be simplified significantly. By comparison with a comprehensive set of calibration measurements we cover the entire parameter space relating measured count rates (CR$_t$, CR$_b$) to ($l^\ast$, $l_a$). Based on this approach we can determine $l^\ast$ and $l_a$ of an unknown sample accurately thus making the additional measurement of a calibration sample obsolete. We illustrate the use of this approach by monitoring the coarsening of a commercially available shaving foam with DWS.

cond-mat.soft↗

arXiv2017

Margin Sample Mining Loss: A Deep Learning Based Method for Person Re-identification

Person re-identification (ReID) is an important task in computer vision. Recently, deep learning with a metric learning loss has become a common framework for ReID. In this paper, we also propose a new metric learning loss with hard sample mining called margin smaple mining loss (MSML) which can achieve better accuracy compared with other metric learning losses, such as triplet loss. In experi- ments, our proposed methods outperforms most of the state-of-the-art algorithms on Market1501, MARS, CUHK03 and CUHK-SYSU.

cs.CV↗

arXiv2017

Dynamical and structural signatures of the glass transition in emulsions

We investigate structural and dynamical properties of moderately polydisperse emulsions across an extended range of droplet volume fractions phgr, encompassing fluid and glassy states up to jamming. Combining experiments and simulations, we show that when $ϕ$ approaches the glass transition volume fraction ${ϕ_{g}}$ , dynamical heterogeneities and amorphous order arise within the emulsion. In particular, we find an increasing number of clusters of particles having five-fold symmetry (i.e. the so-called locally favoured structures, LFS) as $ϕ$ approaches ${ϕ_{g}}$ , saturating to a roughly constant value in the glassy regime. However, contrary to previous studies, we do not observe a corresponding growth of medium-range crystalline order; instead, the emergence of LFS is decoupled from the appearance of more ordered regions in our system. We also find that the static correlation lengths associated with the LFS and with the fastest particles can be successfully related to the relaxation time of the system. By contrast, this does not hold for the length associated with the orientational order. Our study reveals the existence of a link between dynamics and structure close to the glass transition even in the absence of crystalline precursors or crystallization. Furthermore, the quantitative agreement between our confocal microscopy experiments and Brownian dynamics simulations indicates that emulsions are and will continue to be important model systems for the investigation of the glass transition and beyond.

cond-mat.soft↗

arXiv2017

An silicon nanocrystal laser

Silicon lasers have been the most challenging element for the monolithic integrated Si photonics. Here we report the first successful all-Si laser at room temperature based on silicon nanocrystals (Si NCs) with high optical gains. The active Si NC layer was made from hydrogen silsesquioxane (HSQ) that had undergone a phase separation annealing, followed by high-pressure hydrogen passivation. Using the nanoimprint technique, a second-order Bragg grating was made on the active layer as the distributed feedback (DFB) laser cavity. This Si NC DFB laser device was optically pumped with femtosecond pulses. Criteria for lasing action, i.e., the threshold effect, the polarization dependence and the significant spectral narrowing of stimulated emission were evidently fulfilled, which indicated the realization of an all-Si laser.

physics.optics↗